Lessons from Adam Marblestone

Lessons from Adam Marblestone

Biophysicist Adam Marblestone developed the concept of Focused Research Organizations (FROs) to fund engineering projects that fall into the gap between academic grants and venture capital. As CEO of Convergent Research, his work ranges from mapping the mammalian brain to identifying structural components missing from current artificial intelligence. This compilation gathers his thinking on institutional design, neuroscience, and the practical mechanics of funding science.

Part 1: The Focused Research Organization (FRO) Model

1. On the FRO definition: "An FRO is a special purpose organization to pursue a defined problem over a finite period of time, irrespective of any financial gain, like in a startup and separate from any existing academic structure or existing national lab." — Source: [Idea Machines Podcast]
2. On structural purpose: FROs exist to tackle problems that are "a little bit too research for startups and too engineering coordination heavy for academia." — Source: The FRO Podcast
3. On public goods: The ultimate goal of an FRO is to create public goods like datasets, tools, or scalable technologies that make research faster and easier. — Source: [Effective Altruism]
4. On organizational independence: FROs should operate as stand-alone moonshot organizations insulated from both academic and commercial incentive structures. — Source: [Essential Technology Blog]
5. On engineering intensity: Academics can rarely muster the time, focus, and workforce coordination needed to turn a proof-of-principle technology into a reliable, scalable technique. — Source: [Effective Altruism]
6. On project scope: Not everything should be a FRO; they are meant only for specific, time-limited engineering efforts that do not fit anywhere else. — Source: [Science Plus Plus]
7. On startup parallels: While FROs operate with the intensity of startups, they are accountable to their funders rather than shareholders. — Source: [Essential Technology Blog]
8. On ideal frequency: The global scientific ecosystem could benefit from launching roughly ten FROs per year to tackle large-scale bottlenecks. — Source: Existential Hope
9. On finite lifespans: FROs are explicitly designed to have an expiration date, preventing institutional bloat and keeping the focus entirely on a specific milestone. — Source: [Convergent Research]
10. On scale: These organizations are built to execute projects that require tightly coordinated teams of 10 to 30 scientists and engineers working on a single goal. — Source: [Convergent Research]

Part 2: Structural Gaps in Science Funding

1. On misaligned incentives: Many researchers would drastically change their scientific agendas if they were not constrained by current funding structures. — Source: [Second Best]
2. On the philosophy of time in academia: "If you have an academic audience, we say, 'Oh, it's only five years.'" — Source: [Institute for Progress]
3. On the philosophy of time in industry: "But if you have someone working a software engineering job in Silicon Valley, it's more like, 'Well, I've never stayed anywhere for more than two years.'" — Source: [Institute for Progress]
4. On engineering in biology: The tools that allow us to ask fundamental biological questions often require massive, unglamorous engineering efforts that no grant wants to fund. — Source: [Essential Technology Blog]
5. On the gap between discovery and product: There is a valley of death for projects that require millions of dollars of engineering but will never yield a profitable consumer product. — Source: [Clearer Thinking Podcast]
6. On grant constraints: Traditional grants favor hypothesis-driven research, punishing efforts that purely aim to build a new measurement device or map a dataset. — Source: [ChinaTalk]
7. On the limitations of university labs: A single principal investigator with a rotating cast of graduate students is structurally incapable of building industrial-grade scientific infrastructure. — Source: [Bioinformatics CRO Podcast]
8. On venture capital's blind spots: Venture capital funding demands a path to rapid revenue, which immediately disqualifies foundational scientific tool-building. — Source: [Existential Hope Podcast]
9. On reforming science: We do not simply need more money in science; we need new organizational shapes to deploy that money effectively. — Source: [Convergent Research]

Part 3: Brain Mapping and the Rosetta Brain

1. On Rosetta Brains: We need Rosetta Brains, which are datasets that integrate spatial wiring diagrams, molecular markers, and transcriptomic data within the exact same brain tissue. — Source: [Brain Preservation Foundation]
2. On structural context: Knowing the connectome alone is insufficient; we must also know which genes and proteins are expressed at each specific synapse. — Source: [arXiv]
3. On multi-modal datasets: To understand the brain, we must move beyond isolated datasets and map multiple biological modalities simultaneously. — Source: [E11 Bio]
4. On tool building in neuroscience: Progress in mapping the brain is currently limited more by our physical measurement tools than by our theoretical models. — Source: [Adam Marblestone's Blog]
5. On fluorescent in-situ sequencing (FISSEQ): Technologies like FISSEQ represent the kind of molecular recording required to capture brain activity at scale. — Source: [arXiv]
6. On scaling connectomics: Mapping a whole mammalian brain with molecular annotation requires an industrial effort, rather than a scattered academic one. — Source: [E11 Bio]
7. On making the brain machine-readable: The ultimate engineering challenge of neurotechnology is rendering the brain's dense, physical structure into high-fidelity digital data. — Source: [Kernel]
8. On molecular recording: We need to find ways to record neural activity using molecular tape recorders rather than relying solely on physical electrodes. — Source: [Hertz Foundation]
9. On reverse engineering biology: You cannot successfully reverse-engineer a complex system if you are only allowed to measure one variable at a time. — Source: [Brain Preservation Foundation]
10. On physical constraints: The density of the brain means any effort to map it must solve severe physics and optics challenges simultaneously. — Source: [Adam Marblestone's Blog]

Part 4: Neuroscience and Biological Intelligence

1. On the brain's architecture: The specific physical architecture of the brain allows it to perform amortized inference in ways that purely digital systems cannot yet match. — Source: [Dwarkesh Podcast]
2. On biological hardware: Biology computes in a fundamentally different way than silicon, relying on dense, low-power, parallel chemistry. — Source: [Adam Marblestone's Blog]
3. On learning from limited data: One of the central mysteries of biological intelligence is its ability to learn complex behaviors from incredibly sparse data. — Source: [Dwarkesh Podcast]
4. On world models: The brain is constantly constructing and updating a rich causal model of the world, rather than predicting the next sensory input. — Source: Theories of Everything
5. On neural interfaces: Building high-bandwidth neural interfaces requires solving the physics of interacting with delicate, moving tissue without causing harm. — Source: [Kernel]
6. On the limits of current AI models: LLMs are powerful, but they lack the intrinsic, biologically-grounded structures that allow humans to reason about physical reality. — Source: [Dwarkesh Podcast]
7. On evolutionary priors: The brain does not start as a blank slate; it comes equipped with evolutionary priors that heavily constrain and guide learning. — Source: Theories of Everything
8. On energy efficiency: The human brain operates on roughly 20 watts of power, a level of efficiency that current AI hardware is nowhere near achieving. — Source: [Adam Marblestone's Blog]
9. On the brain as an engineering target: To understand biological intelligence, we must approach the brain as a complex engineered system, rather than viewing it purely as a biological organ. — Source: [Hertz Foundation]

Part 5: The Missing Primitives of AI

1. On innate reward systems: Evolution encodes high-level desires and intentions into the brain via innate reward systems, something currently missing or poorly implemented in AI. — Source: Theories of Everything
2. On the limitations of next-token prediction: Biological brains are not simply next-token predictors; they have distinct architectures for episodic memory, planning, and sensory processing. — Source: [Dwarkesh Podcast]
3. On missing primitives: There are fundamental architectural primitives in biological brains that have not yet been successfully translated into artificial neural networks. — Source: [Convergent Research]
4. On neuromodulation: AI models lack the equivalent of chemical neuromodulators that can globally alter the state and learning rate of the system. — Source: [Dwarkesh Podcast]
5. On feedback loops: The brain relies heavily on massive top-down feedback connections, whereas most current AI relies primarily on feed-forward architectures. — Source: Theories of Everything
6. On continuous learning: Biological intelligence learns continuously without catastrophic forgetting, a capability that remains elusive in machine learning. — Source: [Dwarkesh Podcast]
7. On separate memory systems: The brain clearly separates working memory from long-term storage, using the hippocampus to index and consolidate information over time. — Source: Theories of Everything
8. On grounding: True intelligence requires physical and temporal grounding in an environment, as opposed to passive exposure to static text corpora. — Source: [Adam Marblestone's Blog]
9. On algorithmic diversity: The brain is not a uniform web of neurons; it is a highly structured ensemble of diverse algorithms working in concert. — Source: [Dwarkesh Podcast]
10. On the necessity of neuroscience for AI: We will likely need to understand the missing biological primitives before we can build artificial systems capable of human-like reasoning. — Source: [Convergent Research]

Part 6: Ecosystem Design and Metascience

1. On the science of science: We must treat the organizational structure of science as an engineering problem that can be deliberately redesigned and optimized. — Source: [Idea Machines Podcast]
2. On field building: True scientific progress often requires actively building a new field from scratch, rather than waiting for it to emerge organically. — Source: [Clearer Thinking Podcast]
3. On scientific roadmapping: We need explicit, detailed roadmaps for scientific disciplines to identify exactly which engineering bottlenecks are slowing progress. — Source: [SciBetter]
4. On the role of philanthropy: Philanthropists should fund the creation of public-good infrastructure, like FROs, instead of defaulting to the traditional grant lottery. — Source: [Essential Technology Blog]
5. On institutional inertia: Existing institutions naturally resist adopting new formats like FROs because they disrupt the established metrics of academic prestige. — Source: [Second Best]
6. On coordinating talent: The most valuable resource in science is not money, but the ability to tightly coordinate elite talent around a highly specific goal. — Source: [Convergent Research]
7. On measuring success: The success of a focused research project should be measured by how many other scientists adopt and use the tools it creates. — Source: [ChinaTalk]
8. On the limitations of consensus: Traditional peer review enforces consensus, which actively prevents the funding of high-risk, high-reward engineering moonshots. — Source: [Idea Machines Podcast]
9. On alternative incentive structures: We have to build career paths for scientists where they are rewarded for engineering reliable tools, rather than exclusively for publishing novel papers. — Source: [Bioinformatics CRO Podcast]

Part 7: Engineering vs. Discovery in Biology

1. On Sydney Brenner's hierarchy: "Progress in science depends on new techniques, new discoveries and new ideas, probably in that order." — Source: [Essential Technology Blog]
2. On asking questions vs. building: FROs are the engineering efforts that build the tools allowing others to ask fundamental scientific questions. — Source: [Essential Technology Blog]
3. On the necessity of engineering: Much of modern biology is currently bottlenecked by a lack of basic engineering, not a lack of hypotheses. — Source: [Convergent Research]
4. On scaling up biology: Biological research must transition from a bespoke, artisanal craft into an industrialized, highly engineered process. — Source: [Cultivarium]
5. On the illusion of biological complexity: Sometimes what appears as irreducible biological complexity is simply a lack of the right measurement tools. — Source: [Adam Marblestone's Blog]
6. On unglamorous work: The most important work in science is often the unglamorous task of characterizing and standardizing biological parts. — Source: [Cultivarium]
7. On the separation of concerns: We need to separate the people who invent new biological tools from the people who must engineer those tools to be reliable. — Source: [Convergent Research]
8. On hardware for biology: We lack the fundamental hardware devices necessary to manipulate and observe biological systems at the required speed and scale. — Source: [E11 Bio]
9. On technology overhangs: There are massive technology overhangs in science where the basic physics is understood, but no one has funded the engineering to build the machine. — Source: [Idea Machines Podcast]

Part 8: The Future of Scaling Research

1. On the trajectory of Convergent Research: The goal is to prove that the FRO model itself is a repeatable, scalable format, rather than stopping after launching one or two organizations. — Source: [Convergent Research]
2. On accelerating progress: If we can solve the institutional bottlenecks, we can dramatically accelerate the pace of scientific discovery across all disciplines. — Source: [Existential Hope Podcast]
3. On new organizational forms: The future of science depends on our willingness to experiment with new organizational forms that sit between academia and industry. — Source: [Second Best]
4. On the urgency of tool building: We cannot afford to wait decades for academic labs to slowly iterate on the tools we need to solve urgent biological problems today. — Source: [Clearer Thinking Podcast]
5. On the integration of AI and wet labs: The next frontier of research will require deeply integrating artificial intelligence with high-throughput biological wet labs. — Source: [Cultivarium]
6. On the ambition of science: We have collectively lost the ambition to execute massive, coordinated scientific projects outside of the defense sector. — Source: [Institute for Progress]
7. On overcoming the valley of death: FROs are specifically designed to bridge the valley of death, ensuring that necessary technologies actually reach the scientists who need them. — Source: [Idea Machines Podcast]
8. On the legacy of moonshots: We need to reclaim the spirit of the Apollo program, but apply it to foundational challenges in biology, neuroscience, and climate. — Source: [Existential Hope Podcast]
9. On the ultimate goal: The objective of reforming science funding is to ensure that humanity's brightest minds are working on the hardest physical bottlenecks, instead of fighting for small grants. — Source: [Convergent Research]