We’re very excited to announce the launch of two new Focused Research Organizations in the UK powered by ARIA: Meridial and Echo Labs!

These two organizations represent a new way to build scientific institutions around specific technical bottlenecks that are too engineering-heavy, operationally complex, or long-horizon for conventional labs or startups to address effectively. In Convergent Research’s role as an ARIA Activation Partner we developed the UK FRO Founder Residency with the aim of identifying and refining FRO-shaped projects aligned with ARIA opportunity spaces, to help build the capability to launch and support new FROs in the UK. The UK is Convergent’s first major expansion outside the US.

Each of the two new organizations is tackling a different bottleneck, but both are built around the same core premise: that some forms of scientific progress require purpose-built organizations, not just new grants or new labs. Both align with a distinct ARIA opportunity space, targeting areas where new infrastructure could unlock significant progress. These new organisations are:

Meridial, launching with an initial £14 million award from ARIA and aligned with its Scalable Neural Interfaces opportunity space, is building a microscopy platform designed to map and track synaptic connections in living animals over time. By making it possible to observe how brain connectivity changes across development, disease, learning, and therapeutic intervention, Meridial aims to help bridge an important gap between molecular mechanisms and circuit-level function. Over its funded period, Meridial will work to develop and operate a platform capable of mapping and longitudinally tracking synaptic connections across local and long-range brain circuits over extended time periods.

Echo Labs, launching with an initial £7 million award from ARIA and aligned with its Scoping Our Planet opportunity space, is building new infrastructure to represent the natural world and make it legible enough to model, compare, and forecast. If the state of an ecosystem can be measured as a dynamic system, the implications extend beyond observation. Just as weather and human health became understandable through shared measurements and modeling, ecosystem condition could become a measurable, continuously updated layer of intelligence.

Meridial and Echo Labs join a growing UK FRO landscape that includes Bind Research, a UK-based not-for-profit focused on making disordered proteins druggable. Together, these efforts suggest a broader institutional shift: one in which new scientific organizations are designed not around disciplines alone, but around bottlenecks, capabilities, and the shared infrastructure required to unlock downstream progress.

Both Meridial and Echo Labs are expanding their teams in 2026. Keep an eye on their websites (and this blog) to learn more: meridial.org and echolabs.org.

Subscribe now

We recently wrote about why we’re excited about the NSF Tech Labs initiative and how it could become a landmark moment for Focused Research Organizations. Today we’re sharing our full response to the NSF’s Request for Information.

We submitted a Program Design response, drawing from our experience launching ten FROs to offer concrete recommendations on team selection, governance structure, milestone design, IP management, and transition planning. We also suggested ways in which we could serve as an implementation partner — providing the operational infrastructure, governance oversight, and nonprofit scaffolding we’ve built over four years to help Tech Labs teams deploy funds and start building from day one rather than spending their first year on administrative setup.

Separately, we submitted responses on translational bottlenecks in focus areas the NSF might select. Based on our work exploring high-leverage opportunities for science and technology, we recommend the following areas as especially impactful for US competitiveness. Within each area, we also highlight 10+ teams with promising Tech Labs concepts.

Subscribe now

Artificial intelligence: As AI capabilities accelerate, we have a narrow window to build the defense-dominant technologies that will help determine whether the transition goes well. Provably secure cyber-physical systems will be essential if AI supercharges offensive capabilities — yet AI can also accelerate provable security. Meanwhile, the infrastructure we build may determine the extent to which AI can accelerate beneficial science, and the brain still holds important secrets that may inspire the next generations of more energy efficient and prosocial AI. We highlight teams working on formal verification for trustworthy human-AI agreements, brain-inspired architectures grounded in large-scale neural recordings and connectivity maps, and federated infrastructure for scientific synthesis that moves knowledge out of PDFs and into structured, queryable networks of claims and evidence. Read our full response*.

Biotechnology: Even superhuman AI reasoning cannot substitute for measuring contingent physical facts about biology — mapping “biological dark matter”, developing more powerful and cost effective observational tools, and generating high-quality empirical datasets become more valuable as AI advances and meaningful, predictive data becomes the bottleneck. We highlight teams working on turning connectomics and neural recording data into predictive brain models across species, unlocking Earth’s vast unculturable microbial diversity for the bioeconomy, building fully defined and reproducible organoid platforms, scaling tissue biomanufacturing from artisanal lab practice to automated production, and generating large-scale immunogenicity datasets to predict how engineered therapeutics will behave in the human body. Read our full response*.

Quantum, materials, and semiconductors: Quantum computing is increasingly constrained less by fundamental physics than by engineering challenges in hardware and software infrastructure. Materials discovery is bottlenecked by the gap between computational prediction and experimental validation, including physical materials synthesis and testing. And semiconductor based innovation in new domains is gated by the prohibitive cost and timeline of custom chip development. Meanwhile, fundamental experiments and technology development to underpin atomically precise manufacturing remain underexplored. We highlight teams working on an open-source quantum operating system and unified quantum-classical compute fabric, an AI-native materials data foundry that captures the “dark data” of failed experiments, molecular-scale nanomanufacturing using engineered proteins and DNA origami, and open-source ASIC design tools that let scientists and small startups build custom chips without a need for massive investments. Read our full response*.

(*Please note that some of the “potential tech lab” descriptions have been modified or redacted slightly, at the request of the proposing teams.)

How does a child learn about the world so much, from so little, so quickly? Why does it take so much more data to train today’s leading AI models? How do you grow a mind?

One answer might come from probabilistic programming.

Probabilistic programs are generative models written as code. Probabilistic programming systems allow users to specify how (sometimes very noisy) observations could have been produced, then infer what hidden causes of those observations are the most plausible. They do this by developing and then leveraging domain knowledge and rules about the world, like an understanding of physics or biology or economics. The models and programs produced by probabilistic learning can also be refined by “metaprograms” that can transfer their learning from one domain to another. This ability to specify structured models is meant to allow probabilistic programs to be significantly more energy and data-efficient, compared to unstructured deep learning models.

Deep learning, by contrast, often learns its “rules” mostly through gradient-descent based self-supervised learning: a model is presented with massive piles of examples (often strings of tokenized text), and the model adjusts its weights until its outputs match the training signal. This method of developing artificial intelligence has worked astonishingly well by leveraging newfound scales — of data, compute, and training modes. And it has produced a kind of superhuman performance, where a single model can achieve scores on tests and benchmarks across fields that no single person could reasonably master in their lifetime.

But this scale has a price. Even as mechanistic interpretability work helps us glimpse the vast latent structure these models discover, today’s systems often spend huge amounts of compute learning regularities that humans seem to acquire with a few examples and a much more modest metabolic budget. The brain runs on roughly as much power as a dim light bulb — yet it can learn quickly, adapt fluidly, and generalize in ways we still don’t fully know how to engineer.

The quest to develop better algorithms and implementations of probabilistic programming is, in part, an attempt to close that gap and answer the question of how the human brain became such an efficient learner. How can we build models that discover the world’s structure from scratch? Can we sustain those models with energy or compute scales that are closer to the human brain than a data center?

Here’s Vikash Mansinghka, a Principal Research Scientist in the Department of Brain and Cognitive Sciences at MIT - describing this vision for probabilistic programming in a TEDx talk:

Around the same time as Vikash gave this talk, he was also gearing up to co-found and lead CHI-FRO, a FRO we’ve just added to our ecosystem page.

CHI-FRO (pronounced “kai-fro”, like the Greek letter χ) launched with a mission to develop an open source platform that shows how probabilistic programming can scale cheaply and effectively. As with any fast-moving and ambitious effort, the design decisions made at the outset of an FRO shape the arc of its growth and impact.

In the case of CHI-FRO, we believe that three of those choices really stand out and contributed to the product development successes and remarkably well-fitting early transition that the organization saw during its time at Convergent. (After support from Convergent from 2023 to 2025, CHI-FRO transitioned to a home with another non-profit.)

In addition to covering some of the technical vision and origins of the FRO, we hope that the reusable institutional lessons we share can help the ecosystem — especially in light of the upcoming Tech Labs RFP from NSF and other opportunities in the metascience space.

Subscribe now

1. Consider the Counterfactual

How did the CHI-FRO come to be?

We are in the middle of a wave of new technologies driven by deep learning and GPUs. These two technologies took off together: GPUs had already been heavily engineered as graphics cards for computer games and video-intensive workflows. Because they enabled matrix calculations (like those done by the transformer neural network architecture) to be done in a massively parallel way, they became a natural hardware platform for a new software paradigm that was beginning to show a lot of promise.

But what if the path of technology had been different?

Sara Hooker has a great piece called “The Hardware Lottery” about the history of computer science. Per Hooker, a hardware lottery is when “a research idea wins because it is compatible with available software and hardware and not because the idea is superior to alternative research directions.”

Nearly a decade before CHI-FRO, Adam was intrigued when Vikash posed a deceptively simple question. Consider the counterfactual: what could we learn if we applied a similar engineering mindset to probabilistic programming as the one driving deep learning? What if we could get probabilistic programming running ultra-efficiently on powerful hardware, and make it ultra-accessible with languages as usable and widespread as TensorFlow or PyTorch were when deep learning took off?

We didn’t know the answers, but we knew one way to start pursuing them: build better tools for researchers working on probabilistic programming. And in this case, that scope of ambition and technical roadmap indicated an FRO was needed, in addition to and in support of academic research.

2. Build like a Startup, not a Side Project

Engineering projects often need the right organizational infrastructure to become exceptionally productive. Take as one example the productivity of the Lean FRO as measured by its annual roadmaps after it was able to be structured more like a software start-up and less like an ambitious side-project.

When CHI-FRO started, there had already been numerous intellectual achievements in the probabilistic programming literature. Scaling the promise of those early papers to tackling modern, high-dimensional problems required a particular kind of engineering team structure and labor force. In particular, a robust, well-maintained programming language can provide that kind of needed scale and act as a real force multiplier for a field.

And that’s exactly what CHI-FRO’s engineering team worked on, enabling the continuing development of the Gen language and its ecosystem of open-source libraries, which has been expanded to a number of domains.

Here are three of the platforms developed by CHI-FRO:

• Gen3D: 3D scene perception via GPU-accelerated probabilistic programming

• GenLM: Controlled Generation via Langauge-model probabilistic programming

• GenJAX, see this paper: probabilistic programming with vectorized programmable inference

Together, these achievements pave the way for technologies that can leverage probabilistic program inference using hundreds or a thousand times less computation than deep learning for similar performance.

3. Engineer for (and with) the Ecosystem

Creating essential technology for probabilistic programming means helping answer some very big questions for humans and machines alike. As Josh Tenenbaum, Vikash’s co-founder at CHI-FRO, expressed in the conclusion of his seminal paper from 2011:

The key research questions are as follows: What approximate algorithms does the mind use, how do they relate to engineering approximations in probabilistic AI, and how are they implemented in neural circuits?

These questions have a notable structure that comes from the late British cognitive scientist David Marr. Marr famously articulated three connected levels at which one can analyze cognitive systems: the computational level (which describes what the system does and why); the algorithmic level (which describes how the system does those computations and what representations it uses); and the implementational level (which is how the system physically manifests those computations and representations).

In the case of probabilistic programming, Vikash and Josh wanted to advance all three levels. They already were: Josh’s research program had long asked what kinds of structured generative models could underwrite human learning and common sense. In a more applied way, Vikash was driving his efforts towards making probabilistic programming work on modern, high-dimensional tasks. And hovering in the background was a still deeper question: how could these models be implemented efficiently in neural circuits?

Because each of Marr’s levels tend to demand a different kind of labor, these levels are helpful not only for intellectually categorizing research questions, but also point towards how science and engineering can be organized institutionally. Vikash, Josh, and their colleagues at MIT and other universities were already pushing the bleeding edge of this field in their theoretical and computational papers, building models and running experiments with subjects in their labs. At the same time, they were building some very promising proofs-of-concept through the ambitious Missions of the MIT Siegel Family Quest for Intelligence.

Complementary to these efforts, CHI-FRO’s structure as a Focused Research Organization allowed it to organize a software team to attack the engineering and scalability bottlenecks facing the researchers at MIT and elsewhere. Their contributions sat in the critical algorithmic and implementational layers, helping turn ideas about probabilistic programming into functions, libraries, and language contributions that could be leveraged by the wider research community.

This critical partnership, as part of a broader Project CHI, was patterned into CHI-FRO’s roadmap and organizational structure at the outset, ensuring that what CHI-FRO made was responsive to and coordinated with the research community that could benefit from it.

At the outset of each FRO, we also urge our founders to think about their post-FRO transitions. A helpful framing for thinking about this can be choosing to moving your tools and work into a “.org” (a perpetual foundation), a “.gov” (becoming a state-supported public good), or a “.com” (commercializing your work and spinning it out), or some mix of the three. It was another critical part of the ecosystem — the Probabilistic Computing Foundation — which gave CHI-FRO the chance to pursue a “.org” transition, moving from an incubation period under Convergent into a new home that is even more closely co-integrated with the pursuit of scientific questions. There, CHI-FRO continues its work, and the team’s focus has shifted to building field-wide standard machine-executable models of the human mind that explain the workings of the brain, rooted in the tenets of probabilistic programming.

We are grateful to the CHI-FRO team for all their hard work, and are very proud of what they accomplished in building critical infrastructure just two years under the Convergent umbrella. And we continue to be quite excited about their future, and the continued accomplishments that we believe will emerge in this exciting intersection of computer science, cognitive science, neuroscience, and much else!

Merge Labs’ new team and initial technical direction have roots in Forest Neurotech, a Focused Research Organization (FRO). Forest and Merge will continue to partner, carrying forward the transformative vision we shared when we started Forest just a few years ago. We are thrilled by the pace with which Forest was able to prove the efficacy of the FRO model, and that this proof has spurred deep tech investment at the scale required to take it to the next level.

Moving forward, Merge Labs will invest deeply in long-horizon R&D. In parallel, the ongoing nonprofit effort led by Forest will continue to expand scientific partnerships.

Making Waves

The human brain is a few pounds of soft tissue comprising about one hundred billion neurons (and tens of billions of other cells) packed into an opaque object about a dozen centimeters on each side. Pulses of electricity, a diverse range of chemistries, and dynamic subcellular growths and contractions animate the brain’s connections. These phenomena underlie the entirety of the human experience and make us who we are.

As a result, we’ve built an extraordinary toolkit to measure and stimulate the brain. We put people inside magnetic chambers cooled with liquid helium to infer blood oxygenation (functional MRI); we apply dense electrode arrays to pick up faint electrical signals that have traveled through skull and scalp and propagate across skin (EEG); we shine near-infrared light to estimate blood flow (fNIRS); and, in the highest-stakes cases, we perform neurosurgeries that place electrodes directly on or in the brain to record and modulate activity in patients.

And yet, for something so important, we still have an incredibly limited understanding of how it works, and each of our tools comes with hard tradeoffs.

No single modality hits the “Goldilocks” combination of coverage, resolution, invasiveness, portability, and cost that would make brain interfaces truly transformative.

MRI can be powerful but loud, expensive, and immovable. Electrode placement can be exquisitely precise but invasive, localized, and sparse. EEG and fNIRS are accessible but biased toward signals near the scalp, and they struggle to reach the deep brain regions that shape motivation, sensation, movement, and cognition, and which Adam discussed recently on Dwarkesh Patel’s podcast:

Ultrasound technology works by emitting high-frequency sound waves and listening for the echoes that come back. Some of those waves reflect off hard boundaries, while others travel through softer tissues (like the brain) and scatter off gradients and textures in those tissues. By measuring the timing and strength of the returning waves, we can reconstruct what’s happening beneath the surface. Think of the classic ultrasound image of a fetus and the sound of its heartbeat that gives expecting parents the first picture of their child.

How can this technology help us see what’s happening in the brain?

Functional ultrasound imaging is a modality for using these high-frequency waves of sound not only to see biological structures but to get a window into neural function. The physics of ultrasound and its backscatter allow us to produce an image with a resolution that is on the order of a hundred microns in width (with 100 µm being roughly the width of a human hair).

At the same time, what makes ultrasound especially compelling is its field of view. It can cover large volumes, with a field of view that can span from the surface of the brain all the way to its very center. Hence the name “Forest”: ultrasound as a brain interfacing modality uniquely lets us not lose sight of the forest for the trees, nor vice versa. Here’s Sumner Norman, Forest’s founding CEO, describing it on Ashlee Vance’s podcast earlier this year:

Core Memory

The History And Future Of Brain Implants - EP 31 Sumner Norman

Into the brain we go…

Listen now

9 months ago · 18 likes · 1 comment · Ashlee Vance

While there is no law of physics that rules out ultrasound as a high-resolution readout modality for brain activity, there is one practical obstacle: the thick bone of the human skull. That is why Forest chose to start in cases where a “window to the brain” already existed. In patients who’ve had a craniectomy, usually for relief of swelling or surgical access, Forest was able to deliver clean ultrasound into the brain and recover the highest quality signals possible for scientific discovery. This strategy catalyzed Forest’s ability to demonstrate key milestones for the field: ultrasound can measure function across wide swaths of the human brain, with incredible sensitivity and resolution, on the path to a true whole-brain interface.

Forest’s vision includes many downstream applications: a single ultrasonic implant that can monitor and treat multiple diseases and tackle many use cases, high-performance brain-computer interfaces that don’t require penetrating brain tissue, and one of the most powerful new scientific tools that every clinician and neuroscientist can access.

Fast FROward

Our core thesis for Forest was that raising the bandwidth and coverage of brain interfaces while lowering invasiveness could open up a large design space for clinical and, eventually, broader applications.

It had been theoretically clear for some time that physics permitted such a technology. And when we met the Forest founders, they had already begun to publish papers that demonstrated that the technology could be used to get real, decodable, tractable information about brain activity.

The state of the art in 2023 for this technology was unwieldy, large, and unsuitable for most of the work that clinicians, neuroscientists, and others could hope to use it for. The idea that functional ultrasound imaging in the brain could be usefully done with a small wearable was largely unproven, but plausible in principle.

Before Forest, the view that this was worth pursuing was decidedly not widespread or accepted in the broader neurotech and BCI community. In 2021 it was clear that neither academia nor industry was both able and willing to take a concerted risk on developing this technology.

The Forest founders wanted to make ultrasound a practical modality for next-generation brain sensing and neuromodulation, to ultimately develop a safer, less invasive, and more scalable alternative to what existed, and they sought to make this capability broadly accessible to the neuroscience, neurology, and neuropsychiatry research communities.

We believed those initial physical possibilities of ultrasound could be channeled into useful observational tools and potentially, as ongoing trials funded by ARIA are exploring in the UK, neuromodulation-based therapeutics.

This vision became a reality because of a group of purpose-driven philanthropists – Eric and Wendy Schmidt, Griffin Catalyst (the civic engagement initiative of Citadel Founder and CEO Kenneth C. Griffin), the Riley & Susan Bechtel Foundation, James Fickel, the United Kingdom’s Advanced Research and Invention Agency, and an anonymous donor – whose generous support enabled us to start Forest in 2023 as a Convergent Research Focused Research Organization (FRO) pursuing a clear fundamental capability goal in neurotechnology.

As a FRO, the Forest team developed and pursued an ambitious roadmap for scientific development, software, and hardware advancements. They were dogged in this pursuit, and the improvements they were after came fast: we believe that the Forest device went from lab to human patient in the shortest time ever achieved by comparable neurotech efforts, moving from launch to utilizing devices in patients through multiple IRB-backed clinical pathways in a matter of months.

During their time as a FRO, Forest produced the first-ever functional brain imaging data in humans using ultrasound-on-chip, proving that it can drive a minimally invasive, functional human brain imaging modality.

Forest quickly set up collaborations with clinicians at UCLA Neuro-ICU and Cedars-Sinai, and began demonstrating meaningful capabilities across the tech stack, including ongoing human studies that detected auditory-related signals in traumatic brain injury patients.

And they’ve built a foundation for hardware, software, and signal-processing that is increasingly scalable for trials and clinical use. Last year, Forest released the open-source ultrafast CUDA-accelerated ultrasound beamformer which enabled practical and near-real-time ultrasound data processing for them and their collaborators, and is now available to researchers across the ecosystem. This technology is an order-of-magnitude improvement on the state of the art that came before.

Their early in-human data has shown correlations that meet or exceed published results recorded via fMRI on the same paradigms, strengthening confidence that ultrasound can capture task-relevant functional signals. We believe that this is only the beginning of what can be done with this technology.

A Sonic Boom

We are also deeply motivated by what this technology can mean for people around the world. Millions of people live with severe or treatment-resistant variants of mental illnesses like depression, bipolar disorder, and OCD, meaning that several drugs or treatments failed to provide them with sufficient relief. Billions more people could benefit from a deeper understanding of how the brain works and a greater ability to tune its functioning in health and in disease.

One set of results that we’ve been following firsthand stands out here. As Sumner described on Core Memory this summer, Forest reported bedside detection of “cognitive motor dissociation” (sometimes called “hidden consciousness”) in a comatose patient using ultrasound.

Miniaturized ultrasound is particularly powerful here: it’s impractical and cost-prohibitive to constantly bring comatose patients into the MRI suite from the ICU. A bedside-ready modality for these patients could be revolutionary for our understanding of (and potentially the treatments available for) their conditions, as well as deeply meaningful to their loved ones.

This is just one transformative application among many.

New Growth

We believe that this branching point demonstrates one of many models of success for a FRO.

Forest started as a single organization pursuing deep platform R&D along both a technical development roadmap on the one hand, and what we’d call “impact discovery” on the other. That impact includes public dissemination of the technology, research collaborations and partnerships with clinicians, and practical pathways to getting devices tested and used in medicine. As both the technical milestones and impact proof points began to manifest, we saw new surges of interest in this modality, which we believe will be critical for clinicians and scientists around the world.

We hope to see many more. As highlighted by last month’s Tech Labs Initiative announcement by the NSF, integrated teams pursuing transformative engineering-heavy foundational capabilities with speed and focus can be a powerful way to unlock larger fields. We can’t wait to see how the scaled revolution we dreamt of when we planted the first seeds of Forest continues to grow both through Merge Labs’ work and the ongoing work of the Forest nonprofit.

We are proud of all that Forest’s small, integrated team has accomplished. We congratulate Forest and Merge Labs, and look forward to sharing more about the ongoing Forest nonprofit’s upcoming priorities and leadership here soon.

Subscribe now

It’s been an incredibly exciting year in metascience at large, and for us at Convergent.

Last week, the National Science Foundation announced their Tech Labs Initiative.

The week before, we joined teams from our UK FRO Founder Residency (powered by ) in London as they presented the proposals they’ve been working on since June.

All year, our existing Focused Research Organizations (we’ve launched almost a dozen) have been sharing the fruits of their labor, and we’re hard at work on many more new FROs in the pipeline.

And this October, like we do every year, we brought our Focused Research Organization leaders, team members, and supporters together for FROCon.

Over a few cold, bright, autumnal days in Falmouth, we shared breakthroughs & challenges, identified opportunities for collaboration, and reconnected amidst the day-to-day busy-ness of pursuing our mission.

And while we were all in one place, we took a few hours to pull some members of the team aside, to ask them some questions about what we’re working on, and why.

Here’s the video we made about Convergent:

We hope it is a useful artifact for understanding our work, and a helpful preview for what’s to come.

This year has been a big one, and next year will be even more so.

We’re so excited about what’s to come, and grateful for everyone who’s joined us along the way.

Happy holidays - from our team to you and yours.

tl;dr: If you’re drafting an NSF Tech Labs RFI response around a team-built public good that removes a major bottleneck in a field, we’re happy to talk. Convergent has incubated 10+ FROs (and learned a lot in the process) and can help pressure-test whether your idea is “FRO-shaped,” and, if it is, explore partnering (including potentially referencing your team in our RFI as a prospective collaborator). If you’re determined to advance this kind of effort, we have an opportunity to amplify each other’s impact at a uniquely exciting moment for unconventional scientific projects. If this interests you, you can reach us via this form, and read on…

Last Friday, the National Science Foundation released a request for information about the “launch of a new initiative designed to launch and scale a new generation of independent research organizations.”

In their announcement, they wrote that:

The Tech Labs initiative will support full-time teams of researchers, scientists, and engineers who will enjoy operational autonomy and milestone-based funding as they pursue technical breakthroughs that have the potential to reshape or create entire technology sectors. Tech Labs teams will move beyond traditional research outputs (e.g., publications and datasets), with sufficient resources, financial runway, and independence to transition critical technology from early concept or prototypes to commercially viable platforms ready for private investment to scale and deploy.

We’re thrilled that new models for doing vital infrastructure-building for science – including some that may be Focused Research Organizations (FROs) – will be (hopefully!) implemented by the federal government next year.

When we wrote the original blueprint proposing FROs in 2020, we expressed the importance of enabling:

… the U.S. government to fund centralized research programs, termed Focused Research Organizations (FROs), to address well-defined challenges that require scale and coordination but that are not immediately profitable. FROs would be stand-alone “moonshot organizations” insulated from both academic and commercial incentive structures. FROs would be organized like startups, but they would pursue well-defined R&D goals in the public interest and would be accountable to their funding organizations rather than to shareholders. Each FRO would strive to accelerate a key R&D area via “multiplier effects” (such as dramatically reducing the cost of collecting critical scientific data), provide the United States with a decisive competitive advantage in that area, and de-risk substantial follow-on investment from the private and/or public sectors. Some FROs would lay the engineering foundations for subsequent government investment in programs similar in scope to the Human Genome Project.

Since then, Convergent Research has launched almost a dozen of these non-profit startups in the US to build critical infrastructure for science and technology. Just last week, we had a symposium to cap off our first FRO Founder Residency program, powered by ARIA in the UK. And all year, we’ve been working on new FROs in line with our priorities for the Intelligence Age.

The Possibility of Institutional Forms

World War II and the fiery competition of the Cold War gave way to a surge of institutional innovation in American science funding. Overseen by Vannevar Bush’s Office of Scientific Research and Development, the Manhattan Project showed that the government could create a performer that was built-for-purpose: a team assembled to build a specific product and get a specific outcome. With the end of the war and, a decade later, the launch of Sputnik, wave after wave of institutions came on the scene.

These organizations spanned basic and applied research, R&D that was curiosity-driven and motivated by national need, and it took place across public and private labs. During this time, there was also major investment in research at corporate labs. Think of the famed efforts to push the envelope of computing and more at giants like Bell Labs and Xerox PARC.

But after a while, the explosion of new forms just… slowed or contracted. Today, a vast majority of research funding has come to take the form of small project-based grants, to academic faculty PIs working with a handful of students and postdocs in a university, each aiming at their own peer reviewed papers, theses and future small grants.

Universities have been around for a thousand years and may well be around for another thousand. They’re complex institutions with many critical roles in our society. And they will have a meaningful role in shaping this new initiative from the NSF, as well as any parallel efforts at the NIH and elsewhere.

Our existing FROs have often spun out of universities, and they often collaborate closely with them. Some of our FRO founders maintain joint appointments at universities. And all of them consider university researchers to be key users of their technologies and datasets and a source of critical talent, interaction and input.

But a growing consensus is clear: we have tried to make one type of institution be everything for almost everyone and for every problem. Rather than evolving new institutional forms, we spent decades primarily evolving the university.

For a long time, if you were a scientist or engineer who felt a calling to create transformative public goods, the university was your best and most legible home. If you wanted to build something that would outlast a paper and outscale use in a lab - a new instrument, a reference dataset, a platform that makes whole fields cheaper and faster - you still typically had to smuggle it into an institution and grant structure that was optimized for teaching, tenure files, and individual credit for papers. There were exceptions, of course: national labs, a few corporate and philanthropic institutes, the occasional moonshot program. But as a default career trajectory, “doing an ambitious thing in science that isn’t obviously monetizable” typically meant going through the gauntlet of becoming a professor, navigating the university’s incentives, and pushing against the grain, one grant cycle and one grad student/post-doc at a time.

With the emergence of FROs, other new kinds of institutes, and now NSF’s Tech Labs, we now have the possibility to spur on new institutional forms for innovation that can fill precise and important gaps in what the modern academic system offers.

F is for Focused: Avoiding Near Misses

We are especially excited that the Tech Labs Initiative captures the spirit of targeted focus, integrated teamwork and transformative purpose that have animated the first generations of FROs. They write that:

Proposing teams should describe a clear challenge, gap, or bottleneck motivated by practical use considerations that they are uniquely suited to tackle. Successful teams should have a clear vision of how their defined mission is technically challenging, economically significant, and advantageous, and currently unmet by existing funding mechanisms. A proposal should justify why the approach espoused by the Tech Labs program – including dedicated resources for a full-time team of experts – may be uniquely suited to fostering high-impact breakthroughs while addressing the challenge, gap, or bottleneck over several years.

These elements – identification of a gap or bottleneck; a clear vision that is unmet by existing funding mechanisms and team structures; a full-time team of technical and operational/management experts; a transformative technical roadmap to be pursued over a finite time horizon – are exactly what set the FRO apart from other ways of organizing R&D. FROs are by design not just “academia with more cash” nor are they just startups that are funded with public or philanthropic money. We hope that, in the coming expansion of innovative approaches to team-based science, these specific design principles are further developed and experimented on with care.

Put simply: FROs aren’t for everything.

At Convergent, we design our FROs to fill in specific, missing pieces in scientific infrastructure or transformative technology. In each case, this required us extensively mapping the gaps in a particular field, and building up a vision of how the future could be better in a plausible and mechanistic way, a kind of definite optimism.

We spend a great deal of time refining our impact theses and roadmaps, and it has become quite clear that not every problem needs a FRO, and that FROs are not the best solution for many, missing pieces of R&D infrastructure and transformative technology. Accordingly, the Tech Labs RFI does not suggest a replacement for existing work – rather, it carves out a new ecological niche, on the order of 1-2% of the current NSF yearly budget, for specific problems that need a new shape and are transformative for the other 98% if solved.

Accordingly, there are a lot of things we believe FROs are not well-suited for, as seen in the catalog below of near misses. Avoiding these misdirections of effort is what takes up much of our conversation time with proposers who want to start a FRO with Convergent, suggesting that more intuitive or concrete stories and examples may be needed to help people build a concrete mental model of Convergent-style FROs. We’ll be sharing much more of our learnings, but in the meantime, it may be helpful for anyone reaching out to us to check their idea against these common patterns of “near misses”.

“Just do what my lab does, but with more money.”

This approach does not take advantage of the FRO’s independent, non-profit status – for example, the ability to hire engineers or other non-trainee talent at industry wages or do non-publishable work. This also often applies to the idea of converting a lab into a broader “institute”. Sometimes that can be a great thing to do, but it isn’t a FRO.

“Hire 20 of the best researchers and let them work on blue-sky, individual projects in the same building.”

This does not take advantage of multidisciplinary hiring opportunities or the startup-like organizational structure of a FRO, or its laser focus on time-bounded milestones to unlock a very specific bottleneck (what you specifically build).

“Directly work on the biggest intellectual/commercial challenge in the field.”

Ultimately, FROs exist to catalyze problems getting solved – and that is not always the same as directly working on the biggest problem. FROs are justified not by the raw value of their output but by their comparative advantage: doing things academia and industry cannot do.

For example, if the biggest problem in transcranial ultrasound stimulation is identifying its mechanism of action, the right FRO project might be to build hardware, establish experimental protocols, and give them to labs, in order for labs to go figure out the mechanism.

That is not to say that FROs are not working on some of the greatest direct intellectual or commercial challenges in a given field – often, they are – but that this would not be assumed by default.

“Founders and partners don’t pass the police interview test”

Everyone proposing a FRO should be able to describe, independently, the same set of core goals and milestones for the FRO and why they are the right ones; this isn’t just a loose opportunistic collaboration or consortium. FROs benefit from having a unified focus, and that focus needs to be shared by the founders.

“Define the FRO as ‘fill in everything missing in the field’”

This can be too diffuse. You should carve out which part of the field’s needs are specifically FRO-shaped. Other mechanisms (e.g. ARPA programs or academic research can solve for other areas.)

“Lacking the startup non-profit culture and just wanting to be a regular non-profit.”

See for instance this piece for a bit on the elements of being “startup-like”, even beyond the specifics of the FRO model.

“Work on something that requires professional engineering staff and project management, but isn’t transformative or doesn’t unlock one of the top bottlenecks holding back a field”

There are, unfortunately, a lot of projects that require engineering teams that don’t look like self-organized collections of grad students and postdocs, and also don’t return the fund for investors. In a given area, though, there will only be a handful of the most transformative and fundamental bottleneck unlocks that can massively accelerate progress field wide once built and deployed. In neuroscience, for instance, developing technology for mapping the wiring of the brain is a key, fundamental capability that requires tightly coordinated engineering, making it a good fit for a FRO. In contrast, making improvements to any given niche software tool used somewhere in science may be a positive contribution, and may require professional engineering, but often doesn’t justify the resourcing level of a FRO.

“Asking a question versus building something.”

FROs are inspired by the (somewhat apocryphal) quote by Sydney Brenner that “Progress in science depends on new techniques, new discoveries and new ideas, probably in that order.” FROs are not themselves designed to answer a single scientific question. Instead, they are engineering efforts that build something that allows many scientists to ask many questions.

See also: Dyson, F. J. (2012). Is Science Mostly Driven by Ideas or by Tools? Science, 338(6113), 1426–1427. DOI:10.1126/science.1232773

“Being clear why this is not just a startup.”

Many aspects of startups inform the design of FROs, but the key difference is that FROs are driven by a mission to solve a scientific and technical problem to generate public benefit, not by near-term commercial value capture.

The most valuable FRO opportunities often suggest opportunities for startup companies, yet this similarity can paradoxically undercut their prospects for support. Potential supporters are likely to ask, “Why not do a startup instead?”, yet regardless of founding aspirations, a startup’s inherent focus on specific viable products can in some cases turn its effort away from broadly applicable research results. As a consequence, potential startup companies can sometimes cast a shadow over the landscape of research objectives, drawing attention and support away from some FRO opportunities that may promise the greatest and broadest practical applications.

To be clear, we’re very pro-startup. Any project that can achieve its core mission by being a traditional venture-backed startup should do so – we’re specifically interested in those problems where the market alone doesn’t get the necessary fundamental research (and the platform technologies and tools that enable that research) over the hump.

This isn’t to say there aren’t many great platform technology companies, but it also means there are many good FROs that couldn’t be VC backed.

Building Something New

The idea of the FRO is to bring together a big enough team to build a mini-Hubble Space Telescope – a critical mass of well-organized, tightly integrated cross-disciplinary scientific and organizational talent, for a critical duration of time, all pointing at the same target.

Infinite frontiers, serendipitous discovery and novel individual perspectives in science still matter as much as ever – and academic research is a key part of how we get these. But we have overlooked a finite set of gaps that an institutional monoculture has left open for transformation. These are gaps researchers have spent decades feeling, but which they have been unable to cross, for lack of a critical mass of teams organized in truly new forms and incentive structures.

If Tech Labs can help us bridge these gaps, providing new and shared fundamental capabilities for humanity, then we will have made something rare: a new kind of organization that expands the space of possible science for everyone.

How to Work With Us On This

We plan to submit to this RFI, and we hope many, many others do too.

Our sense is that the NSF plans to move fast on this. So: if you have an appropriately shaped idea and nascent team - don’t wait, this is your chance! Read the RFI, attend the webinars, and submit your thoughts on the initiative to the NSF – and hopefully your idea to their subsequent open call for proposals, if they have one.

If you have an idea, but no institutional home (or want to change your institutional home) - you might want to consider working with Convergent Research. As we’re submitting our own RFI, we may list your team in our own RFI submission as such a prospective partner. Please reach out to us to explore this possibility by submitting to this form.

You can see some of our recent priorities here, but we also sometimes think beyond these strict categories.

Structurally, Convergent provides a great deal of support to its subsidiary FROs, ranging from experienced governance and speedy impact-oriented tech transfer, to operational support on non-profit setup, accounting, legal and HR, to a community of other FROs developing an emerging shared practice and culture. We are a 501(c)(3) non-profit separate from any university – but generally excited about working with universities as close collaborators. We were created with the sole purpose of accelerating beneficial science in this way. We’ve stewarded hundreds of millions of dollars of grants already. FROs are set up as LLCs that are subsidiaries of the Convergent parent, but operate with a high degree of autonomy and deep tech startup-like speed. Think of us as a mission control that supports many launches and ongoing efforts. We believe the meta-structure we’ve set up for FROs can, in many cases, help teams align well with the goals of the NSF Tech Labs program. Please feel free to reach out to us to see if our experience supercharging FROs might be a fit for what you want to do.

For your awareness, our FROs tend to involve 4-7 year projects, about 20-30 full-time staff, and about $6-10M/year burn rates, which is on the lower end of NSF’s spectrum; we’re also open to discussing larger projects. To be clear, from our initial read, the Tech Labs RFI suggests a set of possible structures that is compatible with FROs but may also accommodate different or larger projects, inclusive of forms closer to university-adjacent institutes like the Arc or Broad Institutes. We’re especially interested in laser-focused, directed projects to build transformative systems, tools, datasets and so on - what we consider the more “FRO-shaped” end of the spectrum. In the parlance of the X-Labs proposal from the Institute for Progress, we’re an X03 that incubates and supports X02s. We also often do projects well outside the technology areas that NSF’s TIP Directorate is focused on, like our projects in building new tools for astronomy and for fundamental neuroscience. But it is clear that FROs are one of the core types of vehicle this support mechanism is designed to enable, and we’re excited to hear about your ideas for it!

So, feel free to reach out to us at this form. And whether you’re planning on reaching out or going another way, we’d love to add your ideas to our Gap Map.

Subscribe now

You can think of any cyber-physical system as consisting of

1. the hardware that is running the computation or infrastructure,

2. the software that is controlling the hardware, and

3. the people who are running, operating, or communicating with the software.

For the system to work well, you need all of these components to work flawlessly, as compromising any single component can often compromise the entire system.

But things rarely work flawlessly in the real world. Software has bugs, leading it to do things that the system operators don’t intend it to do. Some of these are critical vulnerabilities that can be exploited by attackers. Hardware might have accidental and intended vulnerabilities. It can be hard, expensive, or illegal to pop the hood on commercial hardware and software systems to examine their inner mechanics. And sometimes people try to get the system to do the wrong thing, or people try to gain unauthorized access to the system.

Unsurprisingly, if a cyberattacker wants to compromise a cyber-physical system and can deploy powerful AI systems, then each of these vulnerabilities become easier to exploit. However, a defender could also leverage a powerful AI system themselves. The remainder of this post will discuss specific projects that would enable defenders to better leverage AI systems.

To structure these potential projects, we’ve developed a tiling tree structure, which intends to be mutually exclusive and completely exhaustive, by asking questions that each have a small number of possible answers.

The first question is “What part of the system are we trying to secure?”

Subscribe now

There are three choices here: software, hardware, and the people. Let’s take each in turn:

Securing Software

The world of software moves very quickly, so the deciding factor for what kind of technology you need to secure your deployed codebase is likely your answer to the question of “how fast do you want to - and how fast can you - deploy changes to your software?” Some software is in continuous deployment, some can only be updated with physical contact, and other software require intermediaries to help deploy it. Think about the difference between a webpage hosted on your home server, an iPhone app, and the control software on a nuclear submarine.

To radically improve software security with near-term AI capabilities, we can then ask if we’re willing to build new security infrastructure and/or force operators of various cyberphysical systems to change their workflows. This infrastructure is composed of tools, libraries, definitions, and datasets that enable humans to reason about wholly novel workflows. Creating this infrastructure takes time and human effort because much of it is sourcing tacit knowledge and building consensus around definitions. Doing this work takes time and attention, which will only become more costly as the pace of technological development accelerates.

If groups are not interested in using new tools or changing their workflows yet want better security now, the best we can do is to provide operators of critical cyberphysical systems with AI-based red teaming and blue teaming beyond their human security experts, like the kind Evan wrote about in “Preventing AI Sleeper Agents”.

Red- and blue-teaming is a standard practice in which a so-called “red-team” simulates attacks, and works to discover vulnerabilities, while the blue team defends, and both teams afterward collaboratively improve the system security.

These security audit workflows already exist in software engineering efforts and AI systems have already been shown to be quite effective at identifying and fixing various types of vulnerabilities. The leading AI labs each have an effort to build AI agents to perform this work. And a number of startups and longstanding companies have begun to offer this as a service.

However, leaning into this AI-based red-and-blue-teaming approach wholesale risks triggering an arms race, since the attacker only has to find one vulnerability, but the defender has to fix every bug in every system. Because this effort requires creating AI agents that are very good at finding vulnerabilities, it produces useful tools and compelling targets for bad actors as well as good ones. Because these AI tools can be weaponized, they’re (understandably) generally not accessible to the public, which adds a major operational barrier to broad adoption as well.

–

If operators are willing to use new workflows, but they do not have time to build significant security infrastructure, then we should use AI to drive adoption of existing memory safety technologies.

Most computer programs rely on accessing a region of computer memory on the system in which they are running. If a computer program reads or writes to memory outside a narrow scope allocated to it, this can lead to the program behaving unpredictably, crashing the computer, or modifying the system’s behavior in catastrophic ways. This problem is exacerbated by the incredible complexity in the abstractions with which memory is treated at a hardware level, in how complex computations are executed in hardware, and how different operating systems manage different programs. These kinds of memory handling errors have been reported to account for roughly 70% of exploitable vulnerabilities from Google and Microsoft.

This problem is fixable. For instance, the Rust programming language has been engineered to rule out memory errors. Sadly, most software is not written in Rust, but it is increasingly possible to make the translation from other languages into Rust, with the help of AI.

Happily, there has been movement on the federal level on this topic, and support for continued movement is bipartisan. In 2024, the Office of the National Cyber Director released a report which the White House announced with a press release titled “Future Software Should Be Memory Safe.” In June 2025, CISA and NSA issued a joint guide explicitly advancing adoption of memory-safe languages, citing the 2024 ONCD report. Likewise, DARPA has a program focused on building tools to Translate All C To Rust (TRACTOR), and there’s an FRO proposal (also part of that IFP Launch Sequence collection) to dramatically scale this effort.

One limitation of this memory safety approach, however, is that it does little to mitigate the other ~30% of security vulnerabilities. In addition, many operators of critical infrastructure like electrical grids currently seem unlikely to accept and run AI-generated translations of their software into a new language with which they have no expertise. Of course, this may change over time, but it poses a meaningful barrier to adoption currently.

–

If someone is willing to invest in both new security infrastructure and new workflows, they should build and deploy formal verification as the default for software.

Formal verification brings together three critical components:

1. The code that runs the program

2. A specification of what that software should do

3. A mathematical proof that shows that the code satisfies the specification

Formal verification provides the strongest security by mathematically proving specific security properties in a way that guarantees they will hold against any and every adversary, even an advanced AI. It makes code provably secure. This is a powerful approach that depends on cutting edge research, and we can immediately identify several projects that would meaningfully derisk and accelerate it.

Automating the first and third items, i.e., generating the software and constructing the proof, are already being accelerated by AI. The former have been demonstrated by Cursor and Claude Code, while the later has been demonstrated in pure math with recent projects like AlphaProof and Aristotle that leverage Lean. But more progress is needed before this can meaningfully improve cybersecurity.

For one, we should make more evals, benchmarks, and RL environments for common formal verification tasks. Several leaps in large language model capabilities have been accelerated by the creation of evals and benchmarks which can guide progress in the model’s capability at a particular genre of task, and the development of RL environments in which a model can self-train in a reliable way without needing a great deal of additional data. This should include making models better at using formal verification languages and tools (like Lean or Rocq), fixing proofs with mistakes, or identifying gaps in specifications. The same scaffolding could help us develop even more powerful automated formal verification models.

Another project – which we are aiming to incubate as a FRO – would involve building a set of AI based tools to help humans generate and validate formal specifications. Humans - or at least most humans - don’t think or talk about their code in terms of set theoretic lemmas. But for something to be formally verified, even in Lean, it ultimately needs to devolve from the abstract and somewhat vague symbolic language we use when we speak or when we program software, to the language of formal mathematical proofs. Since the ground truth of “what should the program do” often only exists in the mind of the human operator or designer – and even then, may exist only in a fairly vague way - we need ways of helping humans convert their ideas about what software should ideally do into the exact language (Lean being an example) that a computer can prove theorems in. We believe this is complex, but ultimately quite doable given the current capabilities of LLMs.

One benefit to building formal verification infrastructure is that incremental adoption provides clear benefits. You might assume that replacing the software that runs existing critical infrastructure with secure alternatives must, at some point, involve taking away the old software and forcing operators to run new software they’ve never seen before. However, that isn’t true, because there are intermediary milestones leveraging formal specification and verification that are both helpful to operators and can increase trust.

An initial exploration for a particular system operator could involve using AI-based tools to generate a formal specification from their existing code, documentation, and user input. This specification can then be immediately useful for answering questions about the system, synchronizing code and documentation, and monitoring the active system without changing any operator behavior. This makes it easier for designers and users of engineered systems to understand aspects of system behavior, rather than having them simply replace their system with a version they’re told is secure.

Once the specification-based monitor is primarily finding bugs in the system, not bugs in the spec, the trusted specification can be used to filter anomalies, and eventually, AI systems should be able to generate verified code that is proven to have the security and functionality properties in the specification. This provides early value and fast derisking (especially compared to approaches like rewriting software in Rust), even for operators who run systems for critical infrastructure like electrical grids, regional utilities, and water plants.

Securing Hardware

We could also focus on hardening hardware systems.

Electronic hardware is essentially ubiquitous. A modern car contains thousands of different semiconductor chips. Now think about how many different chips there are in your house, from your internet router to your humidifier, baby monitor, and phone charger. Basically everything more complicated than a simple incandescent lamp has at least one.

Now, where were those chips made? Do you trust them not to do anything you don’t want them to do? Why? (Datacenters, hospitals, and power plants face a much harder version of these questions, yet may have no more dedicated resources than you do to answer them.)

Today, we typically treat hardware as a trusted, fungible commodity: as long as it runs the software we want, we don’t care what chip components power our cars and phones and home appliances. This may make it particularly unsettling for us when we learn that both software and hardware systems can induce nearly undetectable propagating vulnerabilities in cyber systems. As a result, if you want software to run reliably, you need to really trust your hardware as well.

For those of us looking to build resilience into our cyberphysical systems, the next question becomes whether we trust that our deployed hardware systems are secure, or we want to develop trust in it, or we want to build new systems from scratch so that we can trust them.

If we already trust our hardware, it stands to reason that we want to make sure we can continue to trust it in the future. That means protecting the hardware with tamper response measures, because hardware functionality can be compromised and (cryptographic) secrets can be extracted from devices if an attacker has physical access. If you want to make sure that attackers can’t steal an AI model, change your AI system, or read your private data, you need to make sure all the data is permanently erased if someone else gets ahold of it.

This technology could also be used to combat chip theft, export control violations, and other misuse, especially if combined with more destructive tamper response mechanisms and flexHEG monitoring and control capabilities.

The same tamper response hardware in a different form factor would also be useful for building sensors for situations in which you cannot afford for sensor data to be spoofed or fabricated, which could be useful for combating deep-fakes and disinformation. These sensors could make measurements or take images, and cryptographically sign that information with a private key; any tampering with the sensor would lead to the loss of that key, preventing adversaries from signing forged data on a sensor they’d compromised.

–

Beyond finding ways to perpetuate existing trust into the future, you can also build trust in existing deployed hardware that you can’t currently vouch for.

Fighting forged and knockoff chips - hardware that claims to be of a certain provenance but is actually a fake - is already a challenge today, and while defense capabilities here will likely grow faster than offense, the stakes of making a mistake will definitely rise, as cyberphysical systems become increasingly critical to the function of society.

There are demonstrations today that show how people can use infrared light to compare suspect chips to a reference and thereby identify chip forgeries. These demonstrations, however, only catch glaring differences between chips, wherein the two chips sit inside identical centipede-like black enclosures and have meaningfully different functionalities. A more ambitious version of chip validation would not simply compare a chip at a high level to a reference, but provide detailed analysis of the chip’s functionality. This is a much harder (and currently virtually intractable) problem, but it is theoretically possible and could be unlocked with AI capabilities. Additionally, this project has a compelling synergy with an earlier project on our list: you should be able to leverage gains in formal specification capabilities to help do the latter verification of a suspect chip and its functionality.

–

Lastly, if for whatever reason you can’t afford to build trust in existing hardware because you believe it to be too thoroughly compromised or just can’t take the risk that it is, then building de-novo hardware transparently may be the best option to achieve trustworthy security.

One option could be to re-develop high transparency and reliable supply chains, starting with relevant raw minerals, the processing of silicon into ingots, and running semiconductor manufacturing in a highly transparent, secure, and monitored way.

Another option would be to choose specific, high-impact chips (for instance, those monitoring AI usage), and design freely available and open, secure versions of those chips. This would enable anyone to trust-but-verify, providing the same level of security expectations that we have for the most trusted open source libraries. If competitors can all build trust in the same hardware for critical applications like AI governance (e.g., flexible hardware-enabled guarantees, flexHEG), these devices could enable new forms of corporate agreements, international treaties, and user rights/privacy.

Another factor in hardware resilience would be deploying back-up or duplicate systems. For instance, for electricity supplies, this could look like stockpiling transformers (which are notoriously expensive and have long lead times) or decentralizing the electrical grid by facilitating progress on microgrids, battery back-ups, geothermal, and small modular reactors.

Securing the Human Factor

If the hardware and software are both secure, then we still need to consider the human accessing and operating the system.

Humans provide a number of inputs to cyberphysical systems, from authentication information like protected passwords or other evidence of identity, to decision inputs that drive those systems into new states in the real world. While preventing humans from ever making a mistake would require dystopian overreach, new technology could increase our confidence that (a) the only agents with access to a system are humans who are supposed to have it and (b) those humans aren’t accidentally taking actions that are obviously inconsistent with their goals.

To protect against unauthorized access, we could build and deploy radically better identification systems by replicating the internet’s Domain Name Service (DNS) workflow, which gives your computer cryptographic assurances that you’re connecting to the website and server your expect, whether you’re looking for convergentresearch.org or google.com. There are relevant proofs-of-concept like Worldcoin, which provides robust biometric identification, and the Estonian e-ID, which demonstrates that citizens can benefit from government-issued cryptographic identities. In maturity, this system could apply the gold standard of security, which involves combining something you are (biometric), something you have (ie. a hardware key like an HSM, or hardware security module), and something you know (password) into a robust access request.

To protect against outcomes where the right person makes the wrong choice when they interact with a cyberphysical system, there might be reasonable low-hanging fruit in detecting and preventing errors. It might be finally possible for us to introduce privacy-preserving AI assistants that help people making critical decisions to receive relevant, timely input of additional information. This could look like a worker’s AI-powered safety goggles alerting them that in the course of a routine inspection they just set the dial to 1101, which they have never done before and which could lead to the system departing from nominal behavior, and perhaps they meant 1011 instead. You can imagine deploying unblinking, unfaltering AI assistants at work across areas like transit, facility monitoring, and medicine, radically increasing the awareness and knowledge of drivers, monitors, and doctors, but without taking the humans out of the loop. It’s worth noting generally, but especially for applications like this, that the trustworthiness of the AI system as well as the full AI stack become critical, especially as humans increasingly rely on AI assistance.

Lastly, there is a question of recovery after failure. If we’re truly preparing to become resilient against even the worst cybersecurity scenarios, it’s important to acknowledge that failures might still happen even after all the precautions above are taken, and so it’s worth investing in making those failures less bad if they do happen. For a tiny fraction of the US defense budget, we can ensure that America retains the ability to rapidly rebuild critical infrastructure using basic tools and domestic materials. In a world where supply chains can vanish overnight, this capability may determine national survival. We ought to establish a robust baseline repository of tools and knowledge across the US that would enable a cold-start of technology supply chains in the event of catastrophe, rapidly bringing America back to the level of technological capabilities it had in the mid 20th century.

If you would like to see these ideas and more in table format, check out our Google Sheet here. We also have notes for interested funders with potential performers and estimated budgets; please reach out if you’re interested!

Evan Miyazono joined us earlier this year as a research fellow at Convergent. Evan is a physicist, metascientist and founder of Atlas Computing, which has been building the field of AI-based formal verification and hardware-enabled governance mechanisms.

Good Numbers, Bad Numbers

How will science and technology change with abundant intelligence? How will people tap into a “country of geniuses in a data center” to bring about new technologies and cures? These are among questions that motivate us in our work.

But what’s to prevent the people tapping into the data center — or the “geniuses” themselves — from doing intentional or accidental harm?

Computer systems innervate every aspect of our day-to-day lives, undergirding the core infrastructure that supports society. Beneath all our digital possessions — e-mails, smartphone apps, bank accounts, electronic health records — is a world of real things: electrical grids, oil and gas pipelines, buildings and their access controls, hospitals and medical equipment, water treatment plants and air traffic control systems.

Somewhere, in each of those pieces of infrastructure, are deeply embedded cyberphysical systems. To put it overly simply, computers take in data, run calculations, and produce numbers. Trains don’t run into each other, and planes land — for the most part — when and where they should because the computers are good at outputting the right numbers at the right times.

When we converted phone lines into the internet, we made it almost as easy to hack into a computer as it was to make a prank phone call. Even though the systems have evolved significantly since, we still exist in the midst of an arms race between the information security experts who help develop products and systems and malicious actors who want to exploit holes in them. For some of us, that world may only reach us when a video from DEFCON or James Mickens makes its way into our YouTube queues. For some of us, our (or our loved ones’) access to medical care, energy, flights, online content, or financial security have already been affected by mistakes, hacks and breaches. Either way, it is always there, humming along as we tap our cards to check out or get on a train, as we check our e-mail or play a song on our phone, as we visit a loved one in the hospital or turn on our water faucet.

Today, trillions of dollars and thousands of companies are working to build technology that can serve as a drop-in replacement for people, whether they are call center employees, truck drivers, or junior software developers. These machines are increasingly trusted to write code and agentically interact with the world. We’re entering a new stage of vulnerability to the electronic systems that make our fast-moving, deeply interconnected lives possible, with many humans moving onto or even out of the loop.

By creating synthetic labor at scale, labor that can interact with the real world in countless ways, we are creating a new category of vulnerability and attack vectors.

As AI develops, we have a critical window to steer its use so that it promotes resilience rather than increasing vulnerability.

Raising Resilience

Some AI capabilities are growing exponentially. As this trajectory continues, it becomes increasingly valuable to rigorously map what breakthrough technologies and organizations are needed to increase our resilience and ensure that the things we care about remain secure and robust.

After we map the gaps — the missing capabilities — we have the opportunity to decide which should be bridged and crossed. Our principle here is to use the steering wheel, not just the pedals:

We believe that we have a relatively narrow window to steer technological progress in beneficial directions - not just in AI itself, but in the surrounding fields. We believe right action during this critical window will shape many of the enduring impacts of this technology. This means that advancing “defensive” technologies faster today could spell the difference between stability and chaos in an accelerated world. We’ve previously made reference to “differential technological development” and “defensive (or decentralization, or differential) acceleration”; this principle says that we must consider whether the technologies we are accelerating are defense-dominant or not, and that we ought to steer new technology development away from equilibria in which technology systems that impinge upon human wellbeing negatively can proliferate unchecked.

We’ve been talking about this as “AI Resilience”: an approach of identifying features of a world where humans are flourishing with powerful AI, and identifying what needs to be built to move toward it. We believe this approach can be pursued by differentially accelerating defense-dominant technologies.

Critically, some approaches that might seem defense-dominant, like using AI systems to find and fix vulnerabilities, can actually quickly become dual-use, because it can be far easier to find and exploit a vulnerability with powerful AI tools than to widely deploy a patch that AI helped us find.

We’ve already begun to roadmap using this approach:

Earlier this year, we supported a report on this topic by Nora Ammann and Eddie Kembery, which you can read at airesilience.net. In it, they frame a set of technologies and practices that could create defense-dominant scenarios for each of cybersecurity, biosecurity, and epistemic security, which combines protection of mental health and defense against mis/disinformation.

Atlas Computing also recently completed an AI Resilience Gap Map exercise (building on the Convergent Research Fundamental Development Gap Map) with dozens of ideas for potential organizations that the team intends to explore by hiring a cohort of field strategists to design and source founding teams for the orgs. This map provides a framework for thinking about opportunities, but there are many that still need to be identified, so much more work is needed.

Atlas and Convergent also collaborated on a piece for the Institute for Progress on securing what is commonly referred to as the US AI tech stack. Another piece of ours in that collection considers the value of brain connectomic data for developing better-aligned AIs, and several other pieces bear on key opportunities in the AI resilience gap map.

This roadmapping work suggests some clearly useful paths forward.

Provably Secure

Even precious gems, fossils, and deep stone weather over millennia. But logic holds perfectly over time; once you mathematically prove that something is true for a given set of assumptions, it’s true, given those assumptions, forever. Proofs persist. This permanence offers a foundation for building for cyberphysical resilience.

This strategy builds upon our capacity to find, formalize, and verify proofs. We support this at one of our existing Focused Research Organizations, called Lean FRO: the Lean FRO is developing an open source language that allows anyone (human or AI) to guarantee absolute correctness in their mathematical proofs by boiling down mathematical abstractions into watertight chains of logic that a computer can decisively check.

Ultimately, every circuit, program, and chip that does calculations can be represented by mathematical rules. Because we can mathematically prove facts about what these rules can and cannot produce, we can know durable truths about each of those objects and the calculations they do. We can start from reasonable assumptions and prove things about how those combined hardware and software systems work. This could range from a proof that the total amount of money being moved in a transaction remains constant at every step in its execution from start to finish, or that a particular power plant will never blow a transformer as long as no more than one other component fails. When you can begin layering these proven security elements on top of each other, you can have a more robust and resilient system overall.

Even in a world in which the outputs of LLMs are the results of complex neuronal activations and training parameters that we don’t fully understand, code produced by an LLM can be checked and verified relative to a safety or performance specification all the same – a process called formal verification. Even if that code is longer than any human or group of humans can parse, if a fact is proved about it, that fact is true. And because proofs are much, much easier to verify than find, humans can still verify these proofs.

In this way, we can have deterministic knowledge about code. And these proofs constrain not just human actions, but also potential actions of a superintelligent future AI system; if we’ve proven that a program will never crash for any input, then no AI system can find an input that will make it crash, no matter how intelligent the AI is.

In a world that is increasingly dependent on code that no human can review and in which hardware pervades every system we depend on, we can create digital systems that are provably secure.

Here’s discussing this on ’s podcast, :

So far at Convergent, Evan has helped mentor a team working on flexHEG-adjacent technology for compute governance through our UK FRO residency, powered by ARIA, and is helping to design and recruit for a FRO to generate formal specifications, which we’ll talk about below. These are just two early examples of efforts that we believe can meaningfully advance cyberphysical system security.

But there are many more.

A Resilience Renaissance

AI agents are going to open many gates to futures that we thought were locked or didn’t know existed. Even as frontier labs work to harden their systems against misuse, and even as leading companies use frontier models to develop better security against fraud, terrorism, and acts of cyberwar, there will still be poorly protected systems that will crack under the pressure. Market demand and national security might address most of these, but each of these are typically governed on timescales slower than AI actions.

But AI tools can also give us new capabilities in this arms race. As Eric Drexler writes:

Not all gates matter equally. Some unlock applications, for better or worse. Some unlock tool rooms, and some tools open more gates. Sequences matter. Whether interpretability precedes capabilities, whether steering methods precede autonomy, whether knowledge integration pulls ahead of epistemic collapse — these differentials in technology development can shape outcomes for the world.

We can live in a vastly more safe and resilient future, supported by the ballast of defense-dominant technologies. AI-enabled developments could introduce suites of technologies that make the world dramatically safer, healthier, richer, and stronger.

Beyond some sequence of those gates could be the tools that usher in a renaissance of resilience. In our next post, we describe the most promising gates to begin with.

Let’s get knocking.

Read the follow up to this post here:

Thanks to and for their comments on drafts of this piece.

As an editorial in Nature put it, “On 2 July, a science initiative was born in a lecture hall in London.” That morning, the Metascience Alliance, a collaboration of dozens of organizations in the metascience community, was announced onstage at the Metascience 2025 conference.

We were honored to be invited to join the coalition at its launch and to speak at the conference, which brought together 830 participants from around 65 countries.

Anastasia Gamick, our co-founder and President, gave the opening keynote for that morning’s session, “Forging new institutions and alliances.” You can read it below.

–

Hi everyone - welcome to Day 3. I’m impressed that you’re all warriors and you’re all here this morning!

I want to flag that I’m not an academic - I’m a builder or an entrepreneur. I don’t often cite papers, and I don’t actually often read papers, but I wanted to bring a couple to share with you.

The first is an important piece of research published by a single author and it points at a really critical opportunity.

The second, Many Labs, represents the moment psychology turned a statistical warning into an operational solution - a publicly auditable, institution‑backed, globally collaborative stress‑test of its own evidence base.

Now, I asked GPT if the ManyLabs results would have been possible without various members of that alliance. It said:

“Without COS’s purpose‑built Open Science Framework and its central coordination staff, the dozens of labs involved would have had no common platform, or logistical backbone, for preregistering protocols, managing data, and synchronising a global replication effort.

Without the Registered Report format, participating researchers would have faced high career risk that null or contradictory results would be rejected by journals, stalling Many Labs at the proposal stages.

 Without Arnold Ventures’ early multi‑million‑dollar grants covering staff salaries, servers, even the supplies that got shipped all over the world, the volunteer‑driven vision behind Many Labs would have lacked the material resources to move from idea to execution.

I would argue that ManyLabs is an example of how novel institutions and alliances are critical for many, many novel outcomes and experiments that we want.

At Convergent Research - we are busy creating one new type of this novel type of institutional types.

How many of you have heard of FROs? Cool!

We’re inspired by the large-scale science projects that created the fundamental capabilities needed to change fields, like the Human Genome Project, the Large Hadron Collider at CERN and the Hubble Space Telescope. These are examples of grand projects that were pivotal in accelerating their fields and that produced critical research resources. They were platforms that teams of biologists, astronomers, and physicists built on for decades after they were completed.

But, such large-scale projects are often once-a-decade efforts. They require huge feats of coordination and capital and consensus. The Human Genome Project, for instance, cost the federal government $5.6 billion in 2010 dollars and took almost 13 years to complete.

That’s large-scale science.

But what about midscale science?

Across every field, there are problems that are not easily solved within existing institutions, like corporations, venture-backed startups, or academic labs. These problems cost more than the typical NSF grant, in the range of tens of millions of dollars. They often have a bad dollar-to-paper ratio. You require engineers and many different scientists across many different areas, and you try things over and over again as you try to get to something that is executable and not just publishable. And because there is no existing institution, or there are very few institutions to solve these problems, they have been stacking up, choking off many fields we look at.

Here’s one way that we look at it.

Imagine that science is a tree. And that progress goes from the core of that tree, where it’s a very core understanding - something like math or physics - out to the branches, where the fruit are the things that we use in our everyday lives. Let’s say that you want the fruit of that tree, something looks like curing psychiatric disease and neurodegenerative illness, training better and more aligned neural nets, answering deep questions about consciousness.

All of those fruits may depend on the existence of a single branch of science that is not mature enough or advanced enough to bear those fruits. It just might not exist.

In the case of understanding all of these things, we probably need to understand the synaptic connections of the brain. How do different neurons talk both to their neighbors and to neurons all the way across the brain to do things like reward signalling? And often, for that branch to exist, it often needs to come out from the trunk of the tree, and there basic tools and machines that exist at the center of science and technology to support that branch.

If you follow all the way back you come to something we call a fundamental capability. In this case of connectomics, it is reading and error-proofing the synaptic connections in a cheap and scalable way. And so we identified a couple of pieces of technology that would be really important to scale neuro-connectomics, and we built an FRO to tackle just those problems.

Soon, E11, which launched in 2021, will be releasing a preprint with data and the outline for a technology that they’re calling PRISM, which should enable scaling connectomics at an order of magnitude cost reduction from what was previously available.

[ed. note: This preprint, along with plasmids, reagents, and code, was released by E11 yesterday!]

Like all our FROs, E11 is doggedly focused on its particular goal. They want to unblock progress further on in the tech tree, and so they are pursuing pre-specified, quantifiable technical milestones that lay squarely between what neuroscience can do now - and what we want it to be able to do in a short period of time.

As a philanthropically funded organization, its job is to produce a public good. And so they are publishing as they go, they are blogging about their work, they are bringing in the community and sharing the data for their high-impact research.

And it’s led by full-time executives. We had somebody out of Ed Boyden’s lab move from his academic career and come and be CSO and then CEO. They have hired in experts from Janelia, Hopkins, IST Austria, MIT to come do their neuroscience. They have hired automation engineers and barcoding scientists to work collaboratively together towards these end-goals.

We’ve launched ten organizations tackling everything from building tools, datasets, platform technologies, and they’re all sprinting towards their goals.

The blueprint of all of our FROs is trying to unblock those key nodes at the trunk of the tech tree. Let me give you a couple more examples.

Cultivarium is making tools and data accessible through their website and their portal, which is a one-stop shop for biologists looking for information for over 170,000 organisms. Cultivarium is helping scientists figure out how to work with the hundreds of millions of organisms that are available in the ecosystem.

Right now, most R&D in labs and in industry is done only in a handful of organisms rather than working with what is available in the wild, purely because it’s really hard. It took our founders - they worked with one novel organism - it took them their entire postdoc to be able to grow it at all, which means it is inaccessible for people in academic labs or people in industry to make that investment, all to barely be able to get to do the science.

So they’re working on building a set of platform technologies, where you can bring them an organism or a handful of organisms, or you can bring them a use-case that you want to work in, and they can generate how to work with that organism. Recently, they announced a $10m award from the Wellcome Trust, which is dedicating over £50 million in fungal research across their Discovery Research, Climate and Health and Infectious Disease programmes. With this support, Cultivarium is expanding from bacteria to archaea and fungi.

Off Media

Filaments of the Future

At Cultivarium, we are on a mission to open the microbial universe to exploration at scale. A significant part of that universe is comprised of fungi and archaea, whose superpowers have only begun to be understood and harnessed by researchers worldwide…

Read more

a year ago · 12 likes · Henry Lee

So you can log on to their portal and access their data, go in and explore the tree of life, go read their preprints about the different technologies they have built.

Another FRO is called EvE Bio.

Owl Posting

Mapping the off-target effects of every FDA-approved drug in existence (EvE Bio)

Note: Thank you to Bill Busa, CEO and co-founder of EvE Bio, for an extremely helpful discussion while working on this essay…

Read more

10 months ago · 51 likes · 5 comments · Abhishaike Mahajan

Good science depends on good data, especially in the age of AI. (As we know, not all data is good data!)

EvE Bio is building a world-leading dataset of the pharmome. They’re taking every single small molecule drug that the FDA has approved that they can get their hands on - there are a couple where you physically cannot get that small molecule - and they are testing them against hundreds of druggable targets, genetic targets in the genome, and they are open-sourcing that dataset. They’ve developed and are running a high-throughput screening assay about every day and a half, and have now screened more of these combinations than anyone else in the world.

And the data is all available online, under a Creative Commons license. You can check out that data at data.evebio.org.

Those are a couple of the new organizations we have launched in the last few years to tackle gaps we’ve identified where fundamental capabilities would unlock transformative new research. Thanks to the support of philanthropists, we’re able to run this major meta-experiment in institution building.

When we started this, my cofounder, Adam Marblestone, would have guessed that there were on the order of ~10 really good low hanging FRO shaped problems out there. And so we started a systematic search for those problems.

We talked to leading researchers in their field and we would say “what would you do with $50 million dollars to fundamentally change your field?”

And they would say “more money for my lab!”

And we would say “OK, but what if you didn’t have to pay new grad students? What if you didn’t have to publish a paper with a grad student every six months? What if you didn’t have to worry about tenure? What if you could build any facility you want? What would you do with $50 million dollars?”

And they said “more money for my lab!”

And then they walked away, and took a shower or took a walk, and they came back later, and they said, “hey, actually, this dataset is fundamental. This microscope would change the way I could look at things. We have a climate modeling project that will unlock the way that we do marine-based carbon removal.”

All of these things took people and scientists asking themselves new questions. Scientists have frequently been asking them the questions that you have the ability to answer. So if you can answer “what is my next grant proposal?” or “where am I going to get tenure?” - those are the questions you ask yourself.

By asking “what would you do with $50 million to fundamentally change your field?”, people are coming up with new answers, and it’s so exciting.

We took that series of conversations and turned it into the Gap Map. We put out a database of what we believed were gaps and missing foundational capabilities across a range of almost two dozen fields of science. And we’ve been pleasantly surprised by the adoption and use of the Gap Map. I’ve had policy makers from DC to Singapore to Berlin tell me that they are using it in their discourse, as they’re trying to think about their strategic avenues.

You can scan this QR code, if you’d like, or find it gap dash map dot org. There’s a download tool available if you want to put it into a model or query it with your AI assistant. We love adding new things to it - this is a collaborative project. So if you have areas that you think would have fundamental gaps or missing capabilities, let us know. If we’re wrong - we really love it when people tell us we’re wrong - please feel free to reach out.

At Convergent, we believe that new institutions and alliances are often the missing piece in solving critical gaps in scientific progress.

FROs are only one way to solve a set of problems that existing institutions - like traditional academia, startups, large corporate, or governmental research centers - cannot address because of their unique institutional incentives.

They are one new type of organizational form for producing scientific public goods. But we are not the only organization building FROs anymore - there’s a new one in the UK called BIND that just launched with Research Ventures Catalyst. We are working with ARIA now to launch one in the UK.

Other people are working to stand up their own FROs. And - I don’t think that FROs are right for every gap, or that all science should be done with an FRO. But I do believe that we’ve prematurely standardized on a set of R&D institutions - this ecosystem has come some way, and we’re really good at doing research, but we’ve left things that are incentivized by other shapes of organizations on the table.

I am so excited to see everyone here really thinking about how to do science better. I’m excited to hear funders and policy makers in the UK and Europe pushing to run experiments about how we fund things and trying new and different ways to practice science.

I’m most excited by new questions that might be answered by new types of organizations. I look forward to joining the Metascience Alliance, a powerful new opportunity to bring together old and new institutions and work towards a better science. That work is critical, because it shapes how we do science during the fastest-moving developmental period in human history. I really think that what we are doing now will impact humanity for the rest of its existence, and those outcomes are dependent on all of us.

So I’d ask you to please, be bold. Take it from me - you don’t need to be limited by what came before, because we can build the institutions and alliances we need, and then we can get to the exciting work of doing science and R&D. Then, we’ll live in a world in which scientific progress is hopefully only limited by the creativity and hard work that we put in.

At Convergent, we match founders, funders, and ideas to help tackle the critical bottlenecks slowing down scientific and technological progress. We do this through a new type of scientific institution - the focused research organization, or FRO - a non-profit startup that we help design, launch, and operate. Each FRO is a time-bound, technically ambitious effort designed to unblock a whole domain of technical progress.

Sourcing these projects is an active, iterative process that blends scientific roadmapping, founder development, and funder alignment. We talk with hundreds of scientists and technologists each year, working with some of them to reshape initial concepts into “FRO-shaped” projects – tightly scoped efforts aimed at producing a product that fills a critical gap in the scientific process. We intend for those products to seed a whole ecosystem of users who can drive a scaled revolution in their field, and we focus on the type of engineering-heavy public goods creation that couldn’t otherwise happen in academia, industry, or startups.

Over the past four years, this process has launched almost a dozen FROs in domains from brain mapping to climate tech. Typically, we’ve started with a founder or team who has a defined target and a rough plan. Then we work side by side with them to sharpen the technical roadmap, stress-test their theory of change, assemble a founding team, and raise a runway of philanthropic funding.

A New Approach

This June, we kicked off our first-ever UK FRO Founder Residency as an Activation Partner for the United Kingdom’s Advanced Research and Invention Agency (ARIA). This founder residency programme is a deliberate expansion of our playbook, and builds on all of our past learnings.

Early conversations suggested to us that our two (very!) nascent organizations had a set of deeply overlapping interests in pushing science and technology forward. Years later, we were very happy when – after ARIA put out a call for Activation Partners and rigorously vetted a slew of organizations in the space - we were selected to work together, a project which has culminated in this open FRO call.

As Ilan Gur, ARIA’s CEO, said to Eric Gilliam recently on a podcast:

“I think probably the most important thing to accelerate progress in the research ecosystem is to increase the diversity of institutional types. Convergent is one of the few examples we have of a model that is taking root, that’s saying, “Here’s a new institutional type. Let’s figure out how to scale it!”

We wholeheartedly agree about the need for more institutional types! We’ve always had a rolling open call for abstract submissions for FROs from scientists around the world. But over the past year, thanks to ARIA’s interest in supporting FROs across their full range of opportunity spaces, we were able to run our first systematic cohort-based selection process.

Driven by ARIA’s interest in supporting new and experimental programmes (not “come to us and do more of what you’ve always done”, but “come and run an experiment with us based on what you’ve learnt so far!”), we called for entrepreneurial scientists, engineers, and builders eager to tackle FRO-shaped challenges in the UK. Selected teams would participate in a 6-month bootcamp as a cohort, refining and developing their FRO concepts. They would receive funding and travel expenses to cover events like roadmapping workshops, and partake in weekly programming to develop their projects.

After our public presentations sharing the open call - on webinars and at partner institutions - over 90 proposals came in over the stretch of a few weeks this spring, complete with technical concepts, organizational plans, and sketches of how each team wanted to execute their potentially groundbreaking idea. We were pleasantly surprised by the quality of the ideas and the talent, as the applicant pool really seemed to understand the scope and scale of project we were looking for!

We put together a team of expert readers to assess the applications together with the relevant ARIA technologists, taking those 90 or so applications down to a set of 20, then 10, then 5 finalist teams via iterations on concept notes, virtual pitches, and an in-person finalist day in London in June.

After all was said and done, the teams we selected are working towards:

• Democratising access and facilitating analysis of Earth-scale environmental data by providing open-source, AI-powered software infrastructure and data-as-AI models that eliminate current fragmentation and complexity bottlenecks

• Saving coral reefs from climate change by engineering heat-resistant symbiotic algae using advanced live imaging technologies for screening.

• Providing trustworthy assurance for AI by building verifiable hardware-enabled governance systems.

• Revolutionizing ecological monitoring and forecasting by integrating ground-truth data collection with Earth observation

• Unlocking understanding of brain connectivity changes over time in diseases by building a high-throughput in vivo platform for tracking synaptic connections

The Residency

Fresh off their selection gauntlet, our five teams kicked off their residencies, which they’ve been deeply engaging with this summer. As they work on topics ranging from neuroscience to coral reef restoration, they’re developing their ideas through weekly programming that will culminate with a live UK FRO Residency Symposium in London this December 9th (please contact Corey Ruzicano if you are a funder or ecosystem builder interested in attending).

Why structure it as a residency?

Because turning a bold technical idea into a fundable, time-boxed, public-goods organization requires different skills than succeeding in academia or with a conventional startup. FRO founders must simultaneously do a number of difficult things: crystallize a vision for a tractable, gap-filling product; recruit and organize a mission-driven team; and align an early ecosystem of users and funders who will transform that product into scaled impact. That's too much for a single lab or PI working in isolation. The residency provides a crucible that is both helpful and purposefully intense by bringing teams together under structured conditions, with expert coaching on hand, so that they can pressure-test ideas at startup speed without sacrificing scientific rigor.

We see this first FRO Founder Residency in the UK as a really exciting new modality for how we can source and incubate FROs. It extends our playbook using lessons learned from the past four years of launches (which we’ve written more about and will be sharing with you, very soon!) When we reviewed these learnings, we found that our most successful teams coupled thorough roadmaps with early adopter engagement, saw operations as a primary design problem in and of itself (execution, after all, is everything), and shared their theory of change openly with diverse stakeholders, iterating early and often. These insights inflected both the structure and tempo of the residency we designed.

As the leaders of one team put it:

Participating in the residency with the guidance of Convergent Research has challenged us to maintain an ambitious vision while staying disciplined about both our technical roadmap and our plans to build a truly ”FRO-shaped” organization. This balance has been both demanding and deeply rewarding, ultimately sharpening our thinking considerably. With the support of both ARIA and Convergent Research, we have expanded our network of experts for brainstorming and feedback, encouraging growth at a rate that would not be otherwise possible.

So, what does that look like week to week? We’ve structured a curriculum that evolves over time, blending a diverse programming slate (expert-led sessions, one-on-one coaching, cohort “lab meetings,” and focused sprints) that’s designed to take founding teams from an idea at the start of the residency, all the way to an externally-presentable FRO proposal by December.

That compressed evolution is something we put together intentionally, and it comes in three parts. Teams first work to translate their inspirations and raw ideas into detailed roadmaps complete with milestones, explicit technical risks, and user maps. Then, for the second part, they shift from what to how and focus on the questions involved in building initial organizations, assembling recruitment plans, and putting together budgets that correspond to the needs of their roadmaps. Finally, the teams move into focusing on readiness for fundraising, putting together all the collateral they’ll need for live pitches (like the above-mentioned event on December 9th) and follow-up conversations.

We’ve found that the cohort dynamic is an integral part of the residency - residents share their experiences, pool together resources as needs and asks come up, and build on one another’s momentum.

Our delivery team for the residency includes Convergent members who've helped launch numerous FROs across biology, astronomy, and beyond, alongside external experts (FRO and startup founders, venture investors) who provide real-world examples. And, by design, our programming is focused on not only the development of the technical roadmaps, but also the concrete leadership and operational skills needed to launch a FRO, so that teams are prepared to leap from concept to fully funded organization.

Learnings and Unlearnings

One of the most interesting parts of developing this residency has been reflecting on what our FRO founders learn and unlearn as they launch and run a FRO, and observing a similar process in the resident teams. A critical part of the start of any Convergent FRO is how teams - coming from university labs or other settings - adjust to the operational norms and tempos that we set to help them succeed.

The key learnings gravitate around things like:

• Starting from bottlenecks, not “topics”: When scoping a FRO idea, successful attempts are the ones that take aim at catalytic nodes on the tech tree, where founders can build up conviction that unblocking one capability has great potential to unlock many downstream wins.

• Defining a product and its users from day one: Strong FRO concepts clearly articulate the product that will be built based on milestones achieved from the technical roadmap, the first adopters that can get tangible benefit from that product, and the process by which those users can develop an ecosystem.

• Moving from vision to execution: Ideas can be hard-won, but execution is everything. You need to spend attention and energy on developing mid-level technical plans, milestone ladders, and plans to get early adopter engagement to create a tractable path from a “big if true” concept to a tractable product build that’s motivating for funders

• The critical pillars of organizational excellence: Hiring, budgets, and governance decisions are all early considerations that have huge leverage in org design. Founders that think carefully about their networks and understand how bringing in industry-grade and other critical talent early can accelerate the whole org’s speed from its earliest days

• The importance of milestones as a tool: This is a point that comes up time and again when the team at Convergent thinks about how FROs grow. Using different kinds of milestones (high level impact targets, intermediate checkpoints, and go/ no-gos) to structure a roadmap allows you to to keep ambition high while providing directionality and alignment, both internal and external

• Defining an “impact exit” early: Well-defined ideas for how a FRO can eventually transition (whether it’s to a foundation, open source releases, spinning out startups, etc.) is a critical site for attention from the very beginning

And the critical unlearnings we often see are around shifting from:

• Academic → startup speed: We’ve noticed that bias to action, and thinking in sprints rather than long academic cycles is incredibly important. Teams learn to plan out decisive experiments that can quickly build agreement about how to move forward. We encourage them to make reversible calls quickly, and run compatible processes in parallel whenever possible

• Perfection before shipping → ship to learn: The most efficient founders use fast iteration cycles with external input to learn and improve.

• Networking within your comfort zone → world class reach: Founders need to reach beyond familiar circles to test their hypotheses, recruit advisors and hires, and build real ecosystem momentum

• Budgeting for austerity → investing for escape velocity: FROs exist to tackle substantial, moonshot problems. Being overly conservative with budgets can starve momentum, especially given the time-bounds and run-rates of most FROs. Early investments in critical hires and tooling are essential for lift-off, and waiting too long to make critical decisions wastes fuel that could be used to get there.

Our residents discovered that they could effectively execute a high-impact workshop in just eight to twelve weeks. They've also moved from doing outreach within familiar scientific circles to enthusiastically seeking connections across a truly diverse range of sectors and stakeholders. We think both of these (learning to move fast, and unlearning some previously-held barriers to reaching out) reflect a meaningful adjustment from the teams to the cadence we work to cultivate at Convergent.

We’ve also enjoyed seeing the spontaneous emergence of camaraderie within the residency cohort. Even though each team is working on highly specialized technical projects, they've grown into a culture of mutual encouragement and collective problem-solving. Running a FRO and encountering the problems of doing something hard and new can feel burdensome and demoralizing, and just as our current set of FRO CEOs find support in reaching out to one another, we’ve found that residents in lab meetings regularly swap insights, offer support, and celebrate each other's victories.

Many of us on the Convergent team, especially those of us who have been around the block in incubator and fellowship environments, recognize the same spark that animated the communities we remember fondly looking back. We’re hopeful that - even if the residents in our program don’t end up founding FROs - the skills and relationships they’ve gained will lead to better outcomes across their scientific, technological, and entrepreneurial pursuits! Having big dreams and pursuing them isn’t easy, but it’s meaningful and worthwhile - and providing the space to realize what those dreams are for the wonderful group of people who have come together for this residency is something we’re very grateful to be part of.

The Path Ahead

By the end of the year, the teams that make it through the program will present and field questions on stage in London with a fully fledged proposal for a multi-year, high-impact scientific sprint.

We’ll be sharing their progress in the months ahead, and we invite anyone curious to join us on Oct 9, 2025 at a Convergent UK Launch Party co-hosted with Venture Cafe London and ARIA, which will feature presentations from us and some of our resident teams, and a chance for all of us to connect with our readers and community.

ARIA’s mandate is “high‑risk, high‑reward” and we’re very excited to see how our residents’ proposed FROs ultimately measure up to our rubric and the expectations of the ARIA Programme Directors this December. From this cohort, we aim to launch 1 - 2 FROs in early 2026 with anchor funding from ARIA.

As one of them (Jacques Carolan, founding Programme Director for Precision Neurotechnologies) recently put it:

Convergent Research’s FRO programme is expanding the ‘art of the possible’ - enabling a new generation of scientists to dream bigger. The importance of bringing this ambition and vision to the UK through the residency programme cannot be overstated, and I am so excited to see what the residents come up with.

This residency is a joint experiment: can an agency with such ambitious scientific goals make use of the FRO model to bridge relevant capability gaps in the opportunity spaces they’ve targeted? We were very excited to see Bind launch as a FRO in the UK. We can’t wait to apply our learnings from our US FROs to helping the next UK FROs take flight.

Subscribe now

Enormous thanks to , Avery Holland, Corey Ruzicano, and the whole team supporting the residency, and - of course - to our wonderful partners across the ARIA team, and our brave & enterprising cohort of residents!

Since Convergent Research launched as a non-profit in 2021, AI has advanced at a speed that has surprised many experts.

While we don’t know what will happen, it’s our view that the next two decades – a relevant timeline for many mid-to-long range R&D projects – will likely overlap with an “AI transition” in which AI systems come to transform much of science, the economy, and society

So we have been trying to ask ourselves: what does it mean to do high-impact science as we head into this transition period?

How we do science will certainly change if AIs become more powerful and widespread: some research avenues that once seemed impossibly difficult will likely become more tractable and much cheaper; others will almost certainly be made redundant by AI-driven automation; still others will need additional resources and development for us to leverage the opportunities presented by this wave of technologies.

How can we identify what capabilities will matter most in an increasingly intelligence-saturated world?

At Convergent, we think it is possible to shape the trajectory of science and technology by strategically identifying and launching projects that unlock progress at critical leverage points.

Since 2021, and alongside the current explosion in AI, we’ve been building a launch platform for new organizations – FROs, or Focused Research Organizations – engineered to break through bottlenecks in science and engineering.

In this post, we’d like to share our thinking about how we’ve been incorporating the strategic landscape of AI into our FRO roadmapping process.

(Other organizations across the ecosystem have been asking this question too; we participated in - and read with great interest the other contributions to - IFP’s Launch Sequence essay collection, released yesterday. We link to relevant proposals from that collection, as well as other resources, throughout this post.)

Directed Science

Some advances, like DNA sequencing or the transistor, act as keystone developments that unlock entire fields of progress downstream. (Some economists have called the most powerful of these “general purpose technologies” – the original GPTs – which drive “whole eras of technical progress and growth”.) By identifying these crucial leverage points and hastening their development, we believe we can influence the trajectory of science and technology. We believe that, through careful roadmapping, we can identify core bottlenecks to progress in fields of science and technology, as well as the fundamental capabilities that could bridge those gaps.

But Convergent isn’t just a think tank for scientific roadmapping.

Think of us as a mission control for frontier technology. We launch, support and oversee missions in the form of coordinated teams to build the critical infrastructure to bridge those gaps. To date, our scientific bottleneck analysis and roadmapping led us to launch FROs like E11 Bio, Forest Neurotech and Lean FRO, each of which is unlocking a major scientific bottleneck that we believe is pivotal during a period of rapidly advancing AI.

New types of scientific institutions offer an opportunity to pursue directed R&D that rapidly responds to a changing technological and societal landscape. This is reflected in the White House’s recent report on AI, “Winning the Race: America’s AI Action Plan”, which noted that “Focused Research Organizations… using AI… to make fundamental scientific advancements” should be a key implementational strategy for federal investments in AI-enabled science in America.

Broad AI Acceleration and Its Implications for Research

The rapid advance of AI changes how we should think about prioritization for such directed efforts. Many traditional research projects are built on implicit assumptions about the speed and cost of human effort, about how the economy responds to scientific advances, and about the broader context that determines the relative benefits and risks of new technologies. When AI becomes a dominant force, these assumptions need to be revisited.

As we consider the path ahead, four principles guide our approach to scientific and technological development during this period:

1. Avoid fast obsolescence — Technological advances may render existing research directions, tools, or methods obsolete much sooner than expected. For example, many traditional analyses of critical global challenges assume slow technological progress, but in an AI-accelerated world, these timelines could collapse. Under conventional assumptions, solving climate change might focus on incremental improvements in renewable energy deployment, but if AI drastically accelerates materials discovery, new primitives like room-temperature superconductors could emerge far sooner than expected, reshaping the option space. Similarly, in software infrastructure, efforts to manually refactor legacy climate modeling codebases - once considered decade-long, expensive and difficult-to-organize projects - could become straightforward due to AI-powered software engineering. Don’t rely on fragile assumptions about the future.

2. Complement AI — As AI capabilities grow, some research areas will become even more critical. The key here is to identify work that AI cannot easily replace but instead depends on humans for, which may quickly become the bottleneck. Even superhuman reasoning cannot substitute for measuring contingent facts about the universe or about biology - so projects like mapping biological dark matter (i.e., new kinds of previously-unseen biological data layers, like classes of molecules we haven’t been able to systematically map before), developing new observational tools to do so, and generating high-quality empirical datasets can become even more valuable. Here, the highest-impact work could lie in creating the foundational inputs AI needs to function, and giving AI the tools to do creative science, as Ben Reinhardt describes in “Teaching AI How Science Actually Works: How block-grant labs can generate the real-world data AI needs to do science”; Cultivarium’s PRISM tool is a prototype example of this type of work. These kinds of large-scale scientific efforts remain crucial.

3. Use the steering wheel, not just the pedals — We believe that we have a relatively narrow window to steer technological progress in beneficial directions - not just in AI itself, but in the surrounding fields. We believe right action during this critical window will shape many of the enduring impacts of this technology. This means that advancing “defensive” technologies faster today could spell the difference between stability and chaos in an accelerated world. We’ve previously made reference to “differential technological development” and “defensive (or decentralization, or differential) acceleration”; this principle says that we must consider whether the technologies we are accelerating are defense-dominant or not, and that we ought to steer new technology development away from equilibria in which technology systems that impinge upon human wellbeing negatively can proliferate unchecked. (See, for example, Miles Brundage’s ‘Operation Patchlight: How to leverage advanced AI to give defenders an asymmetric advantage in cybersecurity’ essay, on this topic.)
   

   For example, robust biosecurity capabilities - like the ability to do universal pathogen detection - become far more urgent as AI accelerates the ability to design synthetic viruses. (And note that the major AI labs have recently activated their safety contingencies for bioweapons production ‘uplift’.) Similarly, provably secure cyber-physical systems will be essential if innovations in AI supercharge the deployment of offensive cyber-weapons – yet AI can also be used to accelerate the development of provable security measures.

   
   We also need to unlock faster progress in lagging scientific domains like neuroscience, which could make it possible to study the neural basis of human value formation at an unprecedented level. This is work that could prove pivotal for AI alignment in the medium term and for the role of humans through brain computer interfaces and other neurotechnologies, as we argue in this proposal released by the Institute for Progress yesterday.

   

   
   At the same time, the explosion of AI-generated content and the difficulty that democratic societies have in keeping pace will make our ‘epistemic resilience’ more critical than ever. Building tools that support robust human reasoning, self-governance, scientific synthesis, fact checking, and democratic participation will shape our ability to navigate this transition with our values intact. We believe that building such tooling (across areas like defensive capabilities, epistemic resilience, and democratic coordination) rapidly is a critical domain for future FROs and other coordinated research programs. Now is the time to invest in the foundations of a world in which AI can amplify beneficial resilience measures.

4. Imagine good futures — In times of rapid change, it would do us well to expand our creative toolkit and imagination without being overly anchored on what was previously considered possible. We should pursue high-variance projects that could be transformative in a rapidly evolving world (even if their immediate benefits in today’s world may be meaningful but more modest). Just as Xerox PARC not only invented modern computing but also pioneered new forms of human-computer interaction, we should invest today in research that expands the range of human agency in ways we can’t yet fully predict. Can AI help mathematicians solve the Riemann Hypothesis? What are the design principles of new interfaces by which we can digitize and transmit elements of human mental imagery and experience? Can we create new forms of scientific exploration, like a real world Sim City for the global economy. Can we design new systems for revitalizing ecosystems that have been previously thought unrecoverable? Many doors may be unexpectedly unlocked, so rather than just banging against the scarce few that have historically seemed like they might budge, giving long-range thought to which doors we might most want to open and where we might go on the other side has renewed value.

Steering under Acceleration

So far, Convergent Research has launched almost a dozen FROs across biology, neuroscience, astronomy, software infrastructure for formal mathematics, and climate. As we enter the next phase of our work, our goal continues to be the generation of strong counterfactual impact by directing scientific and technological progress in beneficial ways.

Research Priorities for the Next Phase of Convergent

Based on our principles, we believe that the following areas are poised for outsized counterfactual impact. Over the coming months, we hope to engage scientists, funders, and policymakers in exploring, refining, and pursuing them.

Neuroscience and Brain-Computer Interfaces

The importance of understanding the brain is overdetermined. As we explored with E11’s Andrew Payne in “Mapping the Brain for Alignment: How to map the mammalian brainʼs connectome to solve fundamental problems in neuroscience, psychology, and AI robustness”, better understanding the brain can impact our ability to design safer AIs, tackle massive amounts of suffering experienced by people around the world, create low energy chips, and much else.

• Neuroscience for AI Safety – Understanding how learning in the brain is steered by innate, evolved brain systems could inform AI alignment, which faces an analogous steering problem. We recently participated in a roadmapping exercise in this area and plan to deepen our work, in addition to the Launch Sequence essay above.

• BCIs – A closer integration between brains and computers may be needed to ensure that human intentionalities remain central in an AI-dominated world. We’ve already launched Forest Neurotech, a transformative BCI FRO and we see additional FRO opportunities in this space.

• Fundamental Brain Mapping and Modeling – Progress in functional connectomics could help explain how rationality, irrationality, and ethics emerge from the brain, and could let us explore new possibilities for AI based on more direct brain simulation. This could include:

  • Connectomics in the live brain (our FRO E11 Bio is scaling our ability to map static connectomes)

  • Simulating entire small nervous systems digitally – and the new kinds of datasets that might enable this

• Neural Basis of Consciousness – Understanding consciousness may become critical if AI systems approach sentience. While early-stage, progress in neuroscience is needed to have a chance at unlocking this.

Revealing Biological and Ecological Dark Matter

As we discussed above, AI’s impact in science depends on the availability of high-quality data, yet many crucial biological and environmental domains remain largely unmapped. Creating better measurement devices and analytical tools to help bring these areas of functional importance into view is the kind of effort that may be well-suited to FROs.

• An example from biology: Glycomics – Glycoproteins (sugar molecules that decorate cells, viruses, and antibodies) are fundamental to biology, yet while we have genomics and proteomics, we lack an equivalent field of glycomics to systematically study them.

• An example from climate: Non-CO₂ Greenhouse Effects – Key climate processes, such as nitric oxide emissions and aerosol-cloud interactions, remain poorly characterized, limiting our ability to model and respond to climate change.

• And even more new observational tools – unlocking biological dark matter could involve research directions like:

  • Magnetic sensing and control of biological systems – a new type of control knob on biology that is just beginning to be unearthed

  • Quantum biology – what is the role of the electron spin dimension in biology, see this non-Convergent project

  • Foundation datasets for new modalities like metabolomics, lipidomics and glycomics that haven’t previously seen the kind of investment genomics has (we’ve been starting here with PTI, our FRO on proteome-wide and proteoform-resolved single-cell proteomics)

  • Collecting data from unusual organisms/species (building on our work with Cultivarium)

  • 4D intact tissue cell observatories – see Nobel Prize Winner Eric Betzig’s take on a FRO shaped project

AI for Epistemics

The world is changing faster than ever, and humans augmented by AI systems (or AI systems acting fairly autonomously) will be capable of generating overwhelming amounts of information, some of which will be dangerously misleading or difficult to interpret. Even today, many channels of information exchange are clogged with AI slop generated by actors motivated by platform and market incentives. Better epistemic infrastructure is needed to improve human reasoning, coordination, and truth-seeking. Projects in this area that we’re excited about include:

• AI for Scientific Fact-Checking – Automated error or fraud detection in scientific literature, improving reproducibility and statistical design of studies.

• “System 2” Recommenders – Imagine a feed powered by your long-term goals, not just your immediate engagement. Perhaps AI-powered recommendation systems that help audit your existing feed, and suggest content that you will view as valuable upon reflection while reducing your propensity to be stuck in a filter bubble could be produced as a public good.

• Knowledge Synthesis and Verification –:

  • How can AI-generated knowledge become more productive and cumulative?

  • What should a Wikipedia that is shared by humans and machines look like?

  • Human-in-the-loop Fact-Checking – Automating community-driven misinformation checks, surfacing divergent perspectives to reduce polarization.

• AI for Strategic Decision-Making – AI tools for geopolitical strategy, economic modeling, and coordination in high-stakes scenarios.

• AI tools for scaling deliberative democracy – Tools to enable group decision making and collective intelligence

Provably Secure and Safe Cyber-Physical Systems

Many risks from AI misuse or loss of control manifest through cyber-physical systems in which AIs can digitally effectuate changes to the real world —whether in cybersecurity, biotechnology, or supply chains. Formal verification and hardware-based constraints could provide provably defense dominant security infrastructure necessary to prevent AI-enabled exploits. (This builds on our prior work on formal verification for math with the Lean FRO. See also: ‘The Great Refactor’ by Herbie Bradley & Girish Sastry.) Examples of projects to diligence could include:

• Hardware Locks for DNA Synthesis – Preventing unauthorized synthesis of dangerous pathogens.

• Secure Compute Governance – FlexHEG and other cryptographically enabled mechanisms for controlling how and where AI runs at the hardware level. (See also: “Faster AI Diffusion Through Hardware-Based Verification: How to use privacy-preserving verification in the AI hardware stack to build trust and limit misuse” by Nora Ammann and David ‘davidadʼ Dalrymple.)

• Formal Verification of Contracts and AI Behavior – Borrowing from formal verification in math (e.g., Lean) to develop provably compliant governance mechanisms.

• Privacy-Preserving AI Enclaves – Enabling local, secure AI processing without unnecessary data exposure.

• Structured transparency systems – Creating a larger negotiating space for coordination, assurance, auditing and other selective information sharing actors dealing with advanced technologies. (We’re looking forward to Andrew Trask and Lacey Strahm’s forthcoming essay related to this, “Unlocking a Million Times More Data for AI”.)

• Automating Secure Software Development – AI-driven workflows for producing formally verified, bug-free software at scale, to improve security and make software defense dominant.

Defining the Technological Basis of Post-Scarcity

New research directions could prototype technology for a world of material post-scarcity by providing programmable control over matter at the atomic and molecular scale. This could change both what is materially possible for us to manufacture and grow in the short-term, and help calibrate our planning for how AI-enabled science development could open up a future with much broader shared material prosperity.

• Modular Molecular Legos – While AI-driven protein design has advanced medicine and biomanufacturing, it has largely been within the rubric of biological templates. What if we created programmable, Lego-like molecular structures (and mechanisms for assembling them) for nonorganic advanced material production and compositional fabrication?

• AI-Driven Materials Science – Self-driving labs and AI-guided search for room-temperature superconductors and other transformative materials.

• Programming Developmental Biology – The AI-driven modeling of cells and tissues could extend the tools of synthetic biology from single cells to complex organism development. (Could we engineer “brainless cows” to produce meat without factory farming? Could mushrooms be grown with the taste and texture of filet mignon?)

• Enabling Autarky - More prosaically, how self-contained and self-generating can our manufacturing processes and supply chains become, for applications ranging from space colonization to civilizational resilience? Can AI help us preserve and rapidly propagate tacit knowledge in practical manufacturing, materials, research techniques and other critical areas where human knowledge is scarce and vulnerable? Can bioengineering help us turn waste into robustly domestically produced medicines? Can we make food without agriculture?

Complex Systems Economics

AI may drive large-scale and rapid changes to labor markets and global production, but economic modeling remains hamstrung in gaming out the multifarious impacts of these shifts. Moving toward agent-based simulations and more data-driven, fine-grained economic modeling could allow better policy design.

• Complexity Economics – Developing AI-powered, agent-based, out-of-equilibrium models of the economy, moving beyond traditional econometrics or simplified analytically tractable economic models. This could build on the work of Doyne Farmer.

• Simulating AI-Led Economic Transitions – A “real-world SimCity” to model AI-driven economic and labor shifts.

• New Economic Data Collection Methods – High-resolution tracking of economic behavior to train better economic models and improve decision-making.

Defensive Biosecurity Technologies

AI-driven advances in biotechnology raise the risk of engineered pathogens and of broader proliferation of offensive biotechnology, requiring rapid development of more universal, proactive biosecurity measures. Topics we have explored to date at Convergent include:

• Pathogen-Agnostic Detection – Low-cost, metagenomic sequencing or other novel physical modalities for detecting arbitrary biological threats before they spread. (See, for example, Simon Grimm’s “Scaling Pathogen Detection with Metagenomics: How to generate the data necessary to reliably detect new pathogen outbreaks with AI”.)

• Far-UVC for Air Sterilization – Deploying Far-UVC light to passively sterilize air and reduce pandemic risk. See e.g. this report by Blueprint Biosecurity.

The Road Ahead

Now is an important time to ask ourselves what futures we want to enable, and what this means for research directions today. While there are huge uncertainties about exactly how AI will play out, actively trying to envision and prioritize beneficial R&D directions – including ones that might previously have seemed outlandish – and bring about the coordination needed to approach them systematically is a muscle worth building, even as the resulting plans will need frequent revising.

What key directions and possibilities are we missing?

Acknowledgements: Thanks to Eric Drexler, Evan Miyazono, Steve Byrnes, Matt Botvinick, David Dalrymple, Ed Boyden, Adam Brown, Mary Wang, Ales Flidr, Viv Belenky, Kevin Esvelt, Doyne Farmer and Liana Paris for inspiration and discussions.

Subscribe now

Bridging gaps in fundamental R&D

In our first post, we set out a vision for accelerating beneficial scientific R&D:

Humanity could increase its rate of scientific and technological advancement by driving free energy into about 100 especially catalytic projects within reach. If there are ~20 fundamental fields of science, and within each there are ~3-5 pivotal elements of scientific infrastructure that the existing science establishment can’t or won’t build, that's an unmet need on the order of ~100 bridge-scale projects. If each costs $20-50M for a massively enabling sprint, that’s a few $B in input that could unlock trillions in societal value as outputs via accelerated R&D progress.

How do you find those pivotal projects?

In our most recent post, we described how researchers across fields come to create scientific roadmaps -- to identify what’s missing in their fields, and how they can use these roadmaps to make concerted strides forward:

At Convergent, we find these catalytic ideas by talking to brilliant scientists frustrated by systemic barriers to innovation. Motivated researchers are full of great ideas, especially when you give them permission to engage in out-of-the-box, deconstructive thinking.

Over the past few years, we’ve interacted with hundreds of scientists and technologists, and in the process we’ve identified a preliminary (and hopefully provocative!) list that we think illustrates the category of ‘projects that address key gaps in scientific infrastructure’.

To help you explore that list, we’ve built out an interactive database so you can get a bird’s-eye view of various scientific fields, their R&D gaps, and potential solutions requiring “fundamental development” (inspired by the term first introduced at Bell Labs) to break through or route around those gaps. We’re releasing this list now because many fantastic scientists and technologists have helped us identify gaps in our understanding of their fields – and we hope this list catalyzes many more such interactions.

You can engage by using the Contribute function on the site, by writing your own thoughts down and posting them or sending them to us at gapmap@convergentersearch.org, or by starting a conversation on social media and tagging us (we’re @convergent_FROs on X, Convergent Research on LinkedIn, and @convergentresearch.bsky.social on BlueSky).

The projects in this survey don’t fit neatly into either the “R” or the “D” of research and development. They don’t belong squarely within basic nor applied science. They don’t merely work toward solving a single problem and they aren’t just improvements on the previous state-of-the-art in a field; they fill gaps in what we call “fundamental development” - the infrastructure, tools, techniques, and technologies that enable entire fields of researchers and developers to do the work they need, and that need mid-scale coordinated projects to build.

Think about what high-energy particle accelerators and space telescopes mean for particle physics and astrophysics, or what the Human Genome Project catalyzed for biology; fields often don’t have the institutions necessary to build new infrastructure, especially if they are nascent, and producing these foundational capabilities takes real, coordinated effort. The gap-map is our way of mapping examples of projects that belong to this rough category, and represent a partial set of ones we think should exist.

Our aim in our first release definitely isn’t to be comprehensive - rather, we want to illustrate the category as broadly as we can so you can get a feel for how we and others think about this type of science focused development gap. This list has 100 or so R&D gaps with a few hundred foundational capabilities that need to be built or further developed, of which a few dozen are arguably most pivotal, and that we think are (excitingly!) within reach from where we stand now.

Some of the items in this survey are already being tackled, with dedicated teams doggedly engaging in R&D to acquire new essential technologies on this list - while some have initiatives going after them by building a proof-of-concept or even a group of people with the seed of an idea searching for more resources. Whether they’re already happening and could use more talent and resources, or have teams coalescing around them and need funding or additional strategic development, or are at an earlier stage – all of them started in someone’s head and evolved through conversation, writing, inquiry, and trial-and-error. And that’s exactly what we want to catalyze by releasing this database and by inviting you to contribute to it!

Solving these problems may require a range of new organizational and funding structures, some old and some new, of which Focused Research Organizations (like the FROs we’ve spun up at Convergent or helped inspire elsewhere) are one type. By no means, though, do we think FROs are a one-size-fits-all solution for every scientific bottleneck, nor even most of the R&D gaps on this list. For example, doing all the fundamental development needed to unlock our understanding of brain circuits or of AI for math may need to involve a mix of mechanisms like ARPA programs, companies, FROs, prize competitions and academic-style research and training. Ideally, institutions with meaningful funding to commit could work to drive progress across multiple types of organizations towards these ends in a coordinated way. But, in any case, we strongly believe progress in fundamental development requires some coordinated research programs, both FRO-shaped and otherwise, which in turn requires us to get good at working at the level of mid-scale science.

Patterns among the problems

After scrolling through a bit, you’ll no doubt notice that the fundamental capabilities required to solve the gaps highlighted in our preliminary list aren’t random. They reflect an underlying space of challenges and untapped opportunities that has a hidden structure.

To help guide your intuition, you can think of there being two broad areas of structural analysis that we find scientists and researchers returning to time and again when we have conversations like the ones that became this initial map.

The first is reflection about how nature works, borne out of the fact that there are fundamental, first principles physical limits that enable or constrain technological development. If light didn’t scatter in the brain, we wouldn’t contemplate there being a concomitant bottleneck of recording scalably from neurons. It’s because high temperature superconductors are rare in nature that researchers in the field might want to put together a self-driving lab to more efficiently traverse possibility space and find them.

The second source of structure for this list of fundamental development gaps is how the enterprise of science works today. Research has been organized and funded in a highly rigid format in recent decades across nearly all disciplines including the life sciences, materials science, climate science, astrophysics, social sciences, computation, and more. At a high-level, the recent political economy of these fields has lent itself to naturally funding certain types of developments and not others.

Take as an example, one key theme that emerges across the map when you zoom out (in graph view or otherwise): attaining three-dimensional read/write access to intact biological systems.

Viewed from an engineering perspective, there are obvious opportunities to accelerate progress on biological and molecular problems by being able to work with intact biological systems more robustly. If you had better physical and chemical tools, you could observe and perturb cells, organs (notably brains), and whole organisms with far greater spatiotemporal precision and multiplexing of targets/variables. With better sensing systems, both in hardware and software, you could glean a lot more about illness, disease, and behavior. Chemical and biological matter could be made much more simulatable and engineerable through better data collection, better control knobs, and by designing more interoperable building blocks to build with. For some of these tools, scalable development and application is imminently possible with systematic, centralized, cross-disciplinary engineering effort, yet such an effort often does not yet exist.

With more coordinated research programs operating at the level of mid-scale science, we can work to realize these engineering-intensive possibilities, and many more.

Pushing into the future

This database is focused on an illustrative survey of the landscape. Think of it and treat it like a crude map of an emerging space – not a prioritization of where you should go inside it.

Importantly, some of these advancements are not without significant risks that must be considered. We’ve done our best to mostly focus here on robustly beneficial advancements, and where there are grey areas or dual use potential, to tag them as follows:

• Safety Risk: Risks involving potential for the technology to cause harm or injury due to unintended behaviors or failures.

• Security Risks: Risks involving vulnerabilities within a system that malicious actors could exploit to cause harm.

• Governance/ Ethics Risk: Risks associated with how technology is managed, regulated, and aligned with societal values. Concerns around control, transparency, accountability and ethical implications of deploying the technology.

In future posts, we’ll discuss how we are starting to think about prioritizing which bridge-scale projects to pursue, in light of larger principles and observations. These are ideas like differential technology development, the need for “defensive acceleration”, or the need to adapt to the speed of change we could expect in a coming Intelligence Age.

And most of all - we’re sharing this because we want to hear from you! Whether you’ve been thinking about bottlenecks and gaps in your field, or you have ideas about what needs to be built using mid-scale science approaches, or you have other critical and/or constructive feedback based on what you’ve read here -- please reach out to us! Write to us at gapmap@convergentresearch.org, use the Contribute function on the Gap Map website, or start a conversation on social media and tag us. Let’s explore together!

Acknowledgements

Thanks to the entire Convergent Research team and the many external individuals who contributed their feedback and ideas to this project! If we missed your name, please let us know.

Matthew Akamatsu, Ben Blaiszik, Megan Blewett, Beck Brachman, Keith Brown, Steve Byrnes, George Church, Allegra Cohen, Eric Drexler, Ales Flidr, Lauren Gilbert, Spencer Greenberg, Victor Greiff, Casey Handmer, Sam Holton, Martin Borch Jensen, Tom Kalil, Edwin Kite, Konrad Kording, Tony Kulesa, Eirini Malliaraki, Patrick Mineault, Evan Miyazono, Anand Muthusamy, Manjari Narayan, Michael Nielsen, Daniel Oran, David Pfau, Dan Recht, Ben Reinhardt, Jeffrey Shainline, Sean Simonini, Nikolai Slavov, Ronen Tamari, Aaron Tohuvavohu, Brian Wang, Nicole Wheeler, Brad Zamft.

And thanks to the team that helped us implement the database and portal and get it off the ground: Christina Agapakis, Murray Cox, and Matt Tranzillo.

Our appreciation, as well, to the creators of the fantastic Frontier Climate Carbon Removal Knowledge Gaps database, which served as inspiration for database design.

Science needs more catalytic projects so that we can build essential technologies.

How do we find those catalytic projects? One way is by surveying the landscape of science and creating roadmaps that show us how to navigate the tech tree before us.

There are bottlenecks choking progress in virtually every scientific field, and you can identify these with a good roadmap. Once you know where they are, you can conduct research to break through these bottlenecks (or find clever ways to route around them) and unlock essential technologies. Often this unblocking and rerouting requires coordinated leaps forward — projects at such a scale that individual researchers, traditional institutions, and single grants simply cannot tackle alone.

Identifying scientific bottlenecks requires a special kind of forward-looking judgment. You have to look down the road at what will be made possible by unlocking the particular node on the tech tree you’re contemplating or routing around. Even if it’s unsexy or particularly thorny, you have to judge the value of solving a challenge now by the future unlocks it makes possible. Finding and funding catalytic projects means asking: If you knock over this one scientific domino, ‘how many other dominoes fall downstream?’ not ‘how intrinsically novel or interesting is the original domino?’

At Convergent, we find these catalytic ideas by talking to brilliant scientists frustrated by systemic barriers to innovation. Motivated researchers are full of great ideas, especially when you give them permission to engage in out-of-the-box, deconstructive thinking. Even scientists and innovators who have a lot of practice engaging in first-principles thinking can remain constrained by assumptions held by their field unless you set up a space where you can challenge them.

Astro Teller, head of Google X’s moonshot factory, has noted that achieving "10X" progress can sometimes be easier than obtaining incremental improvements because 10x thinking forces researchers to completely resurvey the landscape. By drawing a roadmap that helps you see promising paths to progress, the scientific community can establish a rigorous basis for justifying and directing the "multi-hop" investment needed to tackle the ambitious projects that lead to breakthroughs.

When you pull these conversations together, we’ve found the results typically reveal three key things: (1) which exciting avenues are being choked by a lack of coordinated research or infrastructure; (2) widely-held, and ultimately false assumptions about what advances are tractable; and (3) which destinations are especially compelling to target. Talk long enough with these folks about their ideas in the right ways and you end up with a roadmap towards a rich landscape of scientific and technological possibility.

In this post, I’ll sketch some preliminary design principles for five types of scientific roadmapping, and share some of our favorite examples. These principles and existing roadmaps have helped our team at Convergent Research to direct resources toward projects that we expect to be especially catalytic in the next five years.

There are many, many more transformative and fundable projects out there, and I’m excited for us to share more of our own map of them soon. In the meantime, we hope this serves as a guide that can help you read scientific roadmaps, conduct a survey of your own domain or, perhaps, even pitch a bottleneck-busting vision of your own!

How to build a scientific roadmap

To give you an intuition for how to interpret these artifacts, let’s first take a look at how scientific roadmaps get drawn, from creating a forum, to setting goals and choosing what type of roadmap is right for you. Whether it’s a group of intrepid researchers who want to leap-frog developments in their field or a collection of government funders determined to spur technological advancement, you have to get knowledgeable people together in a room to start thinking and writing and mapping together. Once you have a sense of how those discussions play out, you’ll have a better intuition for how to read the documents that come out of them.

Creating a Forum

Roadmapping begins by getting scientists to think about systematic barriers to innovation that they often don't get a chance to think about, let alone talk about together in any systematic way. Neuroscientists, for example, might wonder why it’s not yet feasible to map the whole mouse brain connectome; or why we can’t yet read and write to the whole human brain. But it’s not feasible to ask that question within a normal grant cycle.

Typical academic career and funding structures do not reward or provide regular venues for such work, unfortunately, so there is rarely sufficient local and shared incentive for individual researchers to pursue roadmapping projects on their own. (Of course, the history of science shows many moments where paradigm-shifting work is rewarded, but these moments aren’t usually produced by design.) Those that do often execute them as passion projects, labors of love that they lug from conference to conference or conduct well after their other critical grant-related work has been offloaded or has spun down.

So, if you want to help researchers get the kind of bird’s-eye view they need of their scientific landscape to do scientific roadmapping you can help by building a supportive forum. Workshops, brainstorming sessions, and collaborative writing exercises where researchers are encouraged to step beyond their immediate research agenda are all excellent venues for this — each is a temporary but potentially transformative environment that creates critical connections and durable artifacts for future use.

Consider ARPA program design workshops, for example, a signature element of the process for launching a high-powered research effort at government R&D agencies like DARPA and IARPA, and now ARPA-H and ARIA. At the inception of any research program a group of program managers, invited scientists, and industry experts gather to evaluate and align on “big if true” research objectives: What if we could predict drug safety and efficacy accurately before clinical trials even begin? What if we could bioprint any organ on demand?

These are hairy, audacious goals that nonetheless reflect real desires on the part of humanity, and could bring tangible benefits to billions of people. Negotiating these goals helps the group to map out research pathways and build a sense of shared ambition between scientists. Large coordinated projects thrive on a bedrock of healthy human relationships and a shared mission, so it’s no mistake that roadmapping starts here. Often, this is a chance for practitioners to revisit the dreams and aspirations that the founders of their field had decades ago, and to evaluate their progress towards reaching those lofty, important goals.

To curate a roadmapping workshop, select a set of participants who can both think outside the box, question assumptions, and ground the subject’s fundamental physical or scientific constraints. To find these people, consult experts and workers in fields who are not only highly knowledgeable but frustrated by systemic barriers to innovation and feel constrained by what’s possible. Often, when I meet these folks I have images of them just busting at the seams, like their lab coat - not to mention their whole lab infrastructure - is too tight. Balance them out with folks that are fresh in the field and can bring fresh perspectives. Each participant enters with a particular expertise and workshop problems are often interdisciplinary.

To borrow a term from the social study of science, you can think of roadmapping exercises as creating what Peter Galison calls “trading zones” between the contributing fields, and the roadmap documents as “boundary objects” that can move between those fields and set a framework for communication between these disciplines. (Lauren Greenspan and her co-authors have written on LessWrong about AI safety cases as useful boundary objects for developing big science efforts in the field.) Participants must therefore catch each other up on relevant background concepts, language, and the state-of-the-art in their contributing fields, align on the true scope of the questions being asked, and actively encourage one another to speculate. Leave the usual stump speeches at home!

You can encourage certain participants to give opening presentations that take risks, setting the tone for others and expanding the range of acceptable discussion and questioning. Future-looking speculation is often discouraged by journal reviewers; in this sort of setting, it’s the perfect tone. As scientists, many of us are pretty good at thinking about why some ideas won’t work; processes like these encourage us to ask what might be possible if this one does?

Roadmapping needs a diverse group of participants but also needs strong leadership. Typically, a cross-disciplinary leader will lead a larger writing exercise that establishes a framework, breaks down the problem into parts, helps each participant contribute to the most relevant parts, synthesizes the contributions from the participants, and melds them into a unified whole, questioning assumptions and making connections along the way.

Consensus is not the goal here, but rather an amalgam of perspectives that, when seen together, uncover a set of specific potential paths beyond the bottleneck — paths that no one participant may be able to fully articulate or even believe in. Whereas surveys of actual physical landscapes are meant to be definitive and durable, scientific roadmaps are wonderfully polyvocal and opinionated and change when new realities about how the universe works become known.

Roadmapping Goals, Structure, and Priorities

One indicator that a topic may be worth roadmapping is the presence of a productive tension: there is widespread agreement about the transformative potential of a goal, but significant debate around its near-term feasibility.

Say that you were those same neuroscientists from earlier, and considered it plausible (though uncertain) whether it was within the boundaries of the laws of physics to create a system that could read from the entire mouse brain. You’d want to move forward by really drilling into the physical limits, as well as figuring out if one could theoretically create such a system for less than all the funding currently available in the field. (One of my first roadmaps from over a decade ago was motivated by conversations like this.)

Or say that you had just gone through a global pandemic and wanted to move beyond pathogen-specific disease detection. You might wonder: can I leverage advances in metagenomic sequencing to detect emerging pathogens in such a discerning way that I can meaningfully separate early pandemic-potential signals from noise?

Finding a question that has this tension (‘this is physically possible and would be very impactful but seems profoundly challenging’) means that you’re in the right place and your topic sits at the frontier of current capabilities - a site where ambitious, innovative thinking can potentially bridge the gap between the goods you can envision and what’s presently achievable. At the same time, being able to define your very ambitious goals precisely is important - simply saying that you want to work on neuroanatomy or Alzheimers is insufficiently defined and will risk your group being pulled in too many directions. On the other hand, being able to say “I want to map the mouse connectome for a large but reasonable sum of money and effort” is quantitative enough to be tractable, posing a challenge that people can really dig their teeth into as they map out their plans to tackle it.

Roadmapping initiatives are designed to reveal these deep-seated gaps in research capabilities and pinpoint opportunities where strategic interventions could yield the most significant scientific returns. Questions central to scientific roadmapping can include:

• What common denominator technologies or capabilities could unlock the most progress across multiple future research areas?

• What core constraints define what is possible in a domain?

• Could there be assumption-violating workarounds for those constraints?

• Which problems should be targeted if we weren’t limited by resources, execution or coordination capacity?

Historically, roadmapping has led to ambitious projects like the Human Genome Project, which came out of a series of meetings at Cold Spring Harbor in the ‘80s. The BRAIN Initiative was born from the Brain Activity Map, which was created at a 2011 gathering organized by the Wellcome Trust, the Gatsby and Kavli Foundations, and the Allen Institute of Brain Science. Similarly, astrophysicist Kip Thorne’s theoretical work on detecting gravitational waves laid the foundation for the Laser Interferometer Gravitational-Wave Observatory (LIGO), a project that succeeded over multiple decades of development and collaboration. One inspirational point of view is reflected in how the old ARPA did “studies” — inexpensive, incisive investigations that revealed the deep, often hidden interconnections within complex systems and served as a trigger for directing larger investments toward the most impactful opportunities

Beyond large government initiatives, we believe roadmapping is vital to fund specific opportunities that address critical research gaps. They provide a strategic overview to see what could be achieved with a hypothetical (ambitious but realistic) technology - scientific infrastructure that the existing science establishment can’t build. Roadmaps guide researchers, and institutions toward high-impact, high-return areas. And they build a strong intellectual case for why particular fields and technologies warrant coordinated, large-scale investment.

Types of Roadmaps

Roadmap formats stand in contrast to the typical academic artifacts - journal articles, which report specific novel results, and literature reviews, which usually focus on the past. Instead, roadmaps often take other shapes such as landscape analysis, technical roadmap, pipeline analysis, opportunity analysis, agenda setting manifesto, and vision paper. Below you’ll find a summary of each, their objectives, and why we think they can be powerful.

The field of scientific roadmapping is a nascent one – this preliminary list is not meant to be exhaustive or a perfect categorization. Roadmaps can use hybrid formats, and it may take time to see which ones you need. But it will hopefully illustrate some of the texture of the explorations we think science needs more of. Let us know your favorite ones that we missed!

1. Landscape Analysis:

Think of this like a literature review, but going over the approaches being taken by people, labs, and organizations rather than just analyzing a collection of journal articles. You want to find both what exists and doesn’t so that you can see what should exist, but doesn’t yet. Landscape analyses of these activities and approaches reveal their assumptions, directions and the gaps between and among them. These high-level perspectives help us identify potential opportunities. New tools for AI assisted literature analysis through efforts like those at Future House and the release of OpenAI’s Deep Research means that the marginal cost of a landscape analysis of existing literature is going down fast, but nothing beats candid conversations with the people behind the work.

Purpose: List who is working in a space and what they’re doing, in order to reveal potential synergies or gaps.

Design Considerations:

• Comprehensiveness. Does this include all the work that could be relevant, even from non-obvious sources?

• Organization. Did the landscape usefully organize the players into a framework that can reveal gaps and opportunities, as well as key differences in assumptions or approaches? (Bonus points if the review introduces a new and useful categorization or framework for thinking about what is being tried and why.)

• Wayfinding. Did the review help identify interesting opportunities, assumptions or gaps?

Examples:

• RA Capital's Tech Atlases

• Positional Chemistry Analysis

• The Amaranth Foundation’s Bottlenecks of Aging

2. Technical Roadmap:

These roadmaps outline the potential paths to overcoming major scientific challenges. They focus on specific technical goals and quantitative constraints.

Purpose: State a technical goal and detail all the possible ways of achieving it through R&D: necessary breakthroughs, and possible solutions.

Design Considerations:

• Well-defined stretch goal. Is success unambiguously measurable? Is the goal still sufficiently far afield that it would truly stretch the existing scientific enterprise and the existing assumptions within the field to achieve it? Can the goal be decomposed into necessary or sufficient conditions? (e.g., Roadmapping to human-level AI performance on Atari or Go would have been hard in advance because we had no idea what nontrivial qualities were needed in a solution. We don’t have a “physics” of intelligence the way we have a physics of electromagnetism. Yet.)

• Completeness. Did you explore all possible routes, including ones far from what is being tried now? Are the approaches considered mutually exclusive and collectively exhaustive (MECE)?

• Constraining. Did you find nontrivial limiters (logical, quantitative, or probabilistic) for the different paths to the goal? Did you explore ways to work around those limiters and whether each one is truly fundamental?

Note about one subtype of Technical Roadmap that we love: Take something seemingly impossible and show it is actually just very very hard in a finite number of specific ways. Empower people with a quantitative framework for generating and triaging potential ways forward towards the un-impossible goal. Some examples include Casey Handmer’s writings - like this piece - on what it would take to do supersonic electric flight and a roadmap on whole mouse brain neural activity recording that I worked on. It could be a map of physics-based constraints and potential workarounds, or a MECE assessment (quantitative) for a hard technical challenge facing physical or engineering limits. For problems that have pretty well defined physics constraints, we’ve seen utility from this type of method.

Examples:

• Near-term climate risks and sunlight reflection modification: a roadmap approach for physical sciences research

• Principles of cryopreservation by vitrification

• Opportunities in pulsed magnetic fusion energy

• Imaging, Sensing, And Communication Through Highly Scattering Complex Media

• Reframing Superintelligence: Comprehensive AI Services as General Intelligence

• Physical Principles for Scalable Neural Recording

3. Pipeline Analysis:

These deconstruct complex research processes, such as drug development or materials science, into discrete stages. Pipeline analyses examine how resources are allocated and identify opportunities for optimization.

Purpose: Get more of some desired output/product from a given process.

Design Considerations:

• Definition. Is the output clearly defined and measurable? (E.g. FDA-approved drugs: yes. Scientific breakthroughs: less so.)

• Deconstruction. Is the process divided into conceptually clean steps?

• Careful accounting. How much time/energy/money/resources goes into each step? What is the yield at each step?

• “Gradient estimation.” How convincing are the estimates of what interventions would increase overall output?

Examples:

• Diagnosing the decline in pharmaceutical R&D efficiency

4. Opportunity Analysis:

These assess the potential impact of various initiatives, allowing funders to understand the likely outcomes and long-term value of different scientific investments.

Purpose: Estimate the expected value (on the margin) of a funding opportunity.

Design Considerations:

• Quantitativeness. Did the author produce convincing numerical estimates of impact?

• Thoroughness. Did the author consider all possible outcomes, failure-modes, and counterfactuals?

Examples:

• Open Phil uses the Importance, Tractability, Neglectedness framework, like New cause area: Violence against women and girls

• DARPA uses The Heilmeier Catechism as a methodology for generating this type of analysis for their programs

• VCs write investment memos, like Bessemer's Shopify investment memo

5. Agenda Setting manifesto

Purpose

Pose a concrete, “big if true” vision for a new capability, and sketch the elements of a coordinated research or engineering program that could plausibly solve it.

Design Considerations

• Direction. Does it lay out a transformative possibility for creation or improvement? Does it lay out a concrete set of necessary and sufficient action?

• Quantitativeness. Does it provide quantitative or at least concrete criteria for success?

• Holistic. Does it clearly identify limiting factors? Does it synthesize a holistic picture of how an otherwise dubious sounding goal could be achieved?

Examples

• Formal Mathematical Reasoning: A New Frontier in AI

• AI and the Future of Digital Public Squares

• Blueprint for Far-UVC Biosecurity

• The brain activity map project and the challenge of functional connectomics

• Organ Preservation Alliance Roadmap for solving the organ shortage

• Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems

• Lauren Greenspan on safety cases to motivate AI safety related research programs

• Scaling up connectomics

• Solving the DNA regulatory code

6. Vision papers:

A somewhat broader category laying out the shape of a largely uncharted field, often through introducing quite new concepts or frameworks. Michael Nielsen has a great essay about vision papers with some examples here. These go beyond classic agenda setting manifestos in the way they introduce new conceptual frameworks and they may not be as concretely actionable versus intellectually inspirational.

Recent examples in AI:

• A Manifesto for Multi-Agent Intelligence Research

• Building Machines that Learn and Think with People

• A Promising Path Towards Autoformalization and General Artificial Intelligence

Roadmaps in any of these formats are tools for identifying, coordinating, and advancing cross-disciplinary initiatives.

At Convergent, we have seen that the roadmaps like this can steer science toward massively enabling technologies. Take this 2018 paper published in ACS Journal of Proteome Research by Harrison Specht and Nikolai Slavov. Specht and Slavov (and others) had long recognized a bottleneck toward medical innovations: researchers’ inability to quantify proteins in single cells. The paper outlined specific ideas to improve single-cell mass spectrometry sensitivity and throughput by orders of magnitude. Think of this work as a technical roadmap (as outlined above) that uncovered key dimensions for improving proteomics. These ideas make a case for devoting resources to particularly catalytic R&D. And it led to the founding of Parallel Squared Technology Institute (PTI), a focused research organization.

A couple parting thoughts…

Roadmapping fits into a broader picture of enabling systems level research leadership. It is necessary because individuals and individual labs cannot make it through these bottlenecks alone.

In today’s research ecosystem, it can be hard to come by the resources, time, and justification for building a roadmap, but this leads to a negative feedback loop where we impoverish the culture of science by promoting thinking focused largely on projects the size of the most common grants (e.g., the NIH R01). Roadmapping promotes a forward-looking and agentic culture of research as well. Eric Drexler discussed this kind of culture and Space Age spirit in describing his inspiration to build a roadmap for nanotechnology:

…the space systems community taught me a way of thinking that harnessed creative vision to physical, quantitative reasoning, in order to explore what could be achieved in new domains of engineering… Satellite launchers and moonships grew out of quantifiable engineering visions: system-level concepts that could be sketched, assessed, and discarded at a rapid pace, evolving through a kind of Darwinian competition. The best concepts would win the resources of time and attention needed to fill in more details, to optimize designs, to apply closer analysis, and after this refinement and testing, to compete again. The prize at the end would be a design refined into fully detailed specifications, then metal cut on a factory floor, then a pillar of fire rising into the sky bearing a vision made real…

In future posts, we will share more about the outcomes of these kinds of roadmapping processes, detailing bottlenecks, landscapes, and the focused research organizations (FROs) addressing them. Stay tuned by subscribing here.

Subscribe now

Are you a scientist who wants to lead or join a different type of team? Are you a funder looking to accelerate science? Do you see a bottleneck in science that new bridge-scale tools, systems or datasets could transform?

At Convergent, we want to steer the engine of non-profit deep tech that we are building towards the most positively impactful problems for humanity now and in the long term. Our current open call is a request for FROs in the UK. The deadline for submitting a proposal is March 28. Check out the call and submit your ideas for critical bottlenecks and how to bridge them with essential technologies.

Acknowledgements

Thanks to Milan Cvitkovic and Ales Flidr for helping Convergent develop its culture of scientific roadmapping. Thanks to Anastasia Gamick, Mary Wang, Janelle Tam, Joseph Fridman, Christina Agapakis, and Max Levy for reviewing drafts of this piece.

And if you’ve read this far, here are even more examples that our team likes:

• JASON Studies (non-classified reports)

• IDA's Science and Technology Policy Institute

• International Roadmap for Devices and Systems (IRDS)

• Productive Nanosystems: A Technology Roadmap: Drexler, K. E., et al., 2007

• Battery Technology Roadmap: Ma, Jianmin, et al. "The 2021 battery technology roadmap." Journal of Physics D: Applied Physics 54.18 (2021): 183001.

• Nuclear Fission Power for 21st Century Needs: Hyde, R., et al., "Progress in Nuclear Energy," 50(2-6), 82-91, 2008.

• Revolution Stalled (Neuroscience Roadmap): Hyman, S. E., 2012, Science Translational Medicine, 4(155).

• Engineering Biology Research Consortium (EB RC) Roadmaps

• Opportunities in Pulsed Magnetic Fusion Energy

• Willy Chertman’s Fertility Whitepaper

• Transformative Opportunities for Single-Cell Proteomics

• Superconducting Optoelectronic Neurons V: Networks and Scaling

• Physical Principles for Scalable Neural Recording

• Supersonic electric flight

• Near-term climate risks and sunlight reflection modification: a roadmap approach for physical sciences research

• Our own Solid-state Far-UVC Roadmapping Workshop Report

• Towards ubiquitous metagenomic sequencing: a technology roadmap

• Principles of cryopreservation by vitrification

Scientific research has been held back by the very systems driving it.

There’s no shortage of intelligence among our scientists, nor creativity and hard work. But there are essential technologies missing across many of the most important scientific disciplines that would unlock significant breakthroughs—tools, datasets, and platforms that empower new discovery.

These are solvable bottlenecks. Consider how space telescopes, particle accelerators, and genome sequencers reveal a deeper world to us and open new domains of exploration and discovery. What has enabled these leaps? New ways of organizing science.

The teams that can build the tools we need to overcome today’s bottlenecks don’t have a natural home. Academic labs and startup companies will play an indispensable role in the future of technology, but university astronomy labs couldn’t have launched the Hubble Space Telescope on their own, nor would a venture-backed startup have built the Large Hadron Collider at CERN. Hubble and CERN illustrate a common pattern in science: a need for projects that are bigger than an academic lab can undertake, more coordinated than a loose consortium, and not directly profitable enough to be a venture-backed or industry R&D.

Focused Research Organizations (FROs) fit in this complementary role. Our non-profit has launched 10 FROs over the last three years to address the bottlenecks in areas essential to humanity, from neuroscience to synthetic biology to climate tech. All have a focused, five year mission to create new enabling infrastructures to break through scientific bottlenecks.

Breaking bottlenecks to build essential scientific infrastructure can have unexpected results. AI experienced transformational change when pivotal advances in infrastructure (cheap computing, software, abundant internet data, etc) unlocked its potential. We believe it is possible to transform many fields in the same way. Mountains of what Michael Nielsen calls “intellectual dark matter” are rumbling on the verge of some unknown technological eruption that will forever release their impact.

This is our vision for building essential technology.

Building “bridge-scale” science

Essential technology doesn’t have to be “big” science on the order of the Large Hadron Collider and the Human Genome Project. It can instead be a bridge between those once-in-a-generation megaprojects and the work from individual researchers in small labs.

Bridge scale teams operate more like startups, with nimble yet tightly coordinated teams of 20-30 people converging on a specific scientific bottleneck, with skills focused on building a particular technology that advances whole fields of science. These projects are outside of the scope of a PhD thesis or loose collaboration and they are also not VC-fundable commercial projects; rather than expecting a single novel result or an outsized financial return, these efforts generate an outsized scientific return on investment, unlocking productivity for an entire field.

Our research ecosystems are great at developing brilliant academics and commercial products, but are stymied by bottlenecks that need such a bridge. For example, theoretical neuroscientists can’t actually see the detailed structure of the brain, and have instead been forced to model it as a random network or make other oversimplified assumptions. But what if instead of having to speculate, we had tools that make brain mapping 100x cheaper and could get the actual circuit maps of the real brain? Theoretical neuroscience would see a renaissance, as would the study of intractable brain diseases.

These are essential technologies achievable by focused startup-sized teams. They are bridges that blur the boundaries between basic and applied science. To illustrate, is a tool as fundamental as CRISPR basic or applied science? The reality for this particular essential technology and many others is that it’s both at once. Applied work can speed up basic research and vice versa. Fundamental tools that give scientists more access to the brain, decode the language of the human immune system, create tools that advance mathematics with AI, or 100X our ability to map protein interactions in health and disease would all be catalytic for both fundamental advancement and applications.

Building up problem-solvers

In the middle of the 20th century, Vannevar Bush noticed a gap in the research ecosystem. Centralized wartime science was thriving, including the Manhattan Project, the Rad Lab, and breaking the Enigma code. But there wasn’t a large-scale decentralized ecosystem for training and curiosity. Bush responded by helping establish the National Science Foundation.

The scientific ecosystem that emerged has since become excellent at “planting seeds” by training scientists and funding curiosity in our academic institutions. And we still deeply need that part of the ecosystem. But now we also need better approaches for realizing the bridging advances that converge resources into the most fertile plots, and the people who can make that possible. For example, neuroscientists have uncovered many promising options for integrating brain-computer interfaces; Forest Neurotech focused resources on chip design for brain ultrasound devices, and quickly made large strides toward interfacing with human brain activity.

We aim to solve the three-way optimization problem of connecting talent and funding to the technology roadmaps that can break through scientific bottlenecks.

Humanity could increase its rate of scientific and technological advancement by driving free energy into about 100 especially catalytic projects within reach. If there are ~20 fundamental fields of science, and within each there are ~3-5 pivotal elements of scientific infrastructure that the existing science establishment can’t or won’t build, that's an unmet need on the order of ~100 bridge-scale projects. If each costs $20-50M for a massively enabling sprint, that’s a few $B in input that could unlock trillions in societal value as outputs via accelerated R&D progress.

As we create the organizational and financial means to build these solutions, scientists will gain a broader palette of career options. Many scientists want to combine the well-resourced, team-focused, and impact-targeted working environment of a deep-tech startup with the ability to create foundational public goods that “lift all boats” for scientists, regardless of potential returns to venture capitalists. Working in a FRO is a unique opportunity to use non-dilutive funding to build a transformative technology platform and become science entrepreneurs. An impactful platform should lead to post-FRO business and career opportunities as well. While we build tools to enable the science we need, bridge-scale projects attract and nurture a special breed of people who we expect to continue to impact science and society.

A call to action: Are you a scientist who wants to lead or join a different type of team? Are you a funder looking to accelerate science? Do you see a bottleneck in science that new bridge-scale tools, systems or datasets could transform?

At Convergent, we want to steer the engine of non-profit deep tech that we are building towards the most positively impactful problems for humanity now and in the long term.

We won’t choose randomly among these 100 or so essential technologies. We are motivated by society’s pressing challenges. For example, as biology and AI both accelerate at an unprecedented pace, we aim to drive differential progress in security and safety so that we can accelerate to the benefits without falling victim to unacceptable risks – leading to a focus on biosecurity and on fundamental methods supporting safe by design AI, like formal mathematical theorem proving.

Our current open call is a request for FROs in the UK. The deadline for submitting a concept paper has been extended to February 16. Check out the call and submit your ideas for critical bottlenecks and how to bridge them with essential technologies.