← back to problem page
Whitepaper · v0.1·progress & abundance·open to refutation

An optimism.fun request for startups

Scientific productivity

Each research dollar buys less progress than it used to.

Published

2026-04-24

Authors

optimism.fun

Status

Draft · v0.1

License

CC BY 4.0

§0weekly drop · week 1

Editor’s drop

Published 2026-08-20 · optimism.fun

blackpaper · the problem

Bloom, Jones, Van Reenen and Webb published a paper in 2020 with a title that states the finding plainly: "Are Ideas Getting Harder to Find?" (American Economic Review, https://www.aeaweb.org/articles?id=10.1257/aer.20180338). The answer, across every field they measured, is yes. To keep Moore's Law doubling transistor density on the same schedule, the semiconductor industry now needs roughly 18 times more researchers than it did in the early 1970s. Agricultural yield growth shows the same pattern. So does biomedical research productivity per dollar. Research output per researcher is falling, broadly and for decades, not in one field but across the ones that have been carefully measured. This is the quest that sits underneath every other quest on this site. Malaria vaccines, AI tutors, fusion reactors, cancer therapies: all of it is downstream of how efficiently humanity turns research dollars and research hours into new, working ideas. If that conversion rate keeps declining, every other problem gets harder to solve, not easier, even as more money gets thrown at it. The world spends roughly $2.5 trillion a year on R&D (UNESCO Institute for Statistics, https://uis.unesco.org/apps/visualisations/research-and-development-spending/). That spend has grown for decades. Output per researcher, per the Bloom-Jones-Van Reenen-Webb data, has not grown to match it, it has shrunk. The number of practicing researchers worldwide has grown to roughly 8.8 million (UNESCO Institute for Statistics, via Science|Business reporting), and the growth in headcount is precisely what has been absorbing the productivity decline: it now takes many more people to produce the same rate of new ideas that a smaller cohort produced two generations ago. Separately, the replication crisis remains substantially unresolved across psychology, medicine, and parts of biology: a material share of published, peer-reviewed findings do not reproduce when independently retested, at rates that vary by field but are consistently uncomfortable (this is a widely documented pattern in metascience literature, not a single clean statistic, and the honest confidence level here is medium, not high). Grant-application overhead consumes a large and rising share of a researcher's working hours, hours not spent doing the research the grant is meant to fund. Neither problem is new. Neither has been fixed at the system level. The compounding structure is what makes this different from a normal line item: a 10% improvement in how efficiently research turns into working knowledge does not save 10% of $2.5 trillion once, it raises the trajectory of every field that depends on new ideas, indefinitely, for as long as the improvement holds. That is also what makes it hard to fund: the return is diffuse, long-horizon, and does not show up as an obviously fundable line item the way a specific vaccine or a specific reactor design does.

whitepaper · the proposal

The proposal is not one invention, it is a set of structural changes to how research gets funded, executed, and verified, several of which are already running as live experiments rather than proposals. What would actually have to be built or changed: (1) more Focused Research Organizations, time-boxed, fully staffed, mission-specific teams that build a piece of infrastructure a field needs and then hand it off, rather than the open-ended academic-lab model; (2) fast-grant and lottery-based funding mechanisms that cut the months-to-years grant-review cycle down to weeks, freeing researcher time currently spent writing and reviewing proposals rather than doing research; (3) AI-assisted literature review and experiment design tooling that lets one researcher cover the search space a team used to need; and (4) systematic, funded replication infrastructure, so a published finding's reliability is a known, checkable number rather than an assumption. None of these four is speculative technology. All four exist today in early, unscaled form. The unlock is that two of the four (AI-assisted literature synthesis and automated/semi-automated experiment execution) only became viable in the last few years, as language models became competent enough to reliably summarize, cross-reference, and flag contradictions across a literature too large for any one person to read, and as lab automation hardware became cheap enough to run overnight, unattended experiment queues. Convergent Research's FRO model demonstrates the funding-structure half of the fix is also working: FROs typically run $20-50 million initial grants over 3-7 years, roughly ten times the funding density of a typical academic grant, for a fully staffed team of 15-30 researchers pursuing one specific deliverable (Convergent Research, https://www.convergentresearch.org/about-fros; Federation of American Scientists). Convergent Research has launched eight of them since 2021. Sixteen FROs at that scale is roughly a $1 billion program, a real, funded number, not a hypothetical. **The market, worked from real numbers.** The addressable population is the roughly 8.8 million researchers working worldwide (UNESCO Institute for Statistics, reported via Science|Business, https://sciencebusiness.net/news/number-scientists-worldwide-reaches-88m-global-research-spending-grows-faster-economy). The realistic price point is harder to pin to one hard source than the other two problems on this site, and that should be stated plainly rather than papered over: single-tool academic research software (statistical packages, qualitative-analysis tools, reference managers) typically prices in the $250-1,600/seat/year range at the academic tier, scaling several multiples higher at commercial/enterprise tiers. A full AI-research-acceleration bundle, combining literature synthesis, experiment design assistance, and automated lab-queue tooling, would plausibly price as a multi-tool stack rather than a single seat license. Using a conservative planning estimate of $3,000/researcher/year for that bundle: 8,800,000 researchers x $3,000/researcher/year = $26,400,000,000/year. That is a rough $26 billion/year addressable market for AI-research-acceleration tooling specifically, a small slice of the $2.5 trillion/year global R&D spend this site's problem page cites for context, but a real, buildable software-and-services market on its own, comparable in size to the enterprise developer-tools market. The FRO funding-structure fix runs on a separate, smaller track: roughly $1 billion for sixteen organizations at current scale, which is a policy and philanthropy financing question, not a product-market question, and should not be added to the $26 billion figure since they solve different halves of the problem (execution capacity versus funding-allocation efficiency). Conjecture, open to refutation, more than either of the other two entries here: the $3,000/researcher/year figure is triangulated from adjacent academic-software pricing, not observed directly for a bundled AI-research product, because that product category is still forming. The 18x figure and the FRO budget figures are hard, cited numbers. The $26 billion market-size figure is a genuine estimate, and the single most useful thing a reader could do with this page is find a harder number and correct it.
§1abstract

The four-axis ranking

We rank humanity’s most important problems on four quantifiable dimensions — quantity of humans affected, severity per person, current solution quality, and addressable market size — and package each as a proposal in the spirit of Musk’s Hyperloop Alpha. This document is the proposal for scientific productivity. Every number below is sourced and tagged with confidence. Every ranking is a conjecture, open to refutation.

Quantity · humans affected

8.1B

med

Severity · 0–1

unknown

Current solutions

2.5 / 10

med

Market size · TAM

$2.5T

high
§2problem statement

What we are trying to solve

Bloom, Jones, Van Reenen and Webb showed research productivity across semiconductors, agriculture, and biomedicine has been falling for decades, we spend far more scientist-hours for each new idea. The meta-problem: every other quest on this list depends on scientific throughput. Focused Research Organizations, AI-assisted research, metascience reform, and funding structure changes all compound across every domain.

§3why it persists

The gap between the world and the world that is physically possible

Today: Per-researcher productivity declining for 50+ years (BTGD). Most grant time spent on paperwork. Replication crisis unresolved. New PhDs needed to maintain Moore’s-law-style progress doubling every ~7 years.

Current solution quality is rated 2.5 / 10 (med confidence) — meaning there is substantial unclaimed ground between what exists and what is possible. estimated — per-researcher productivity declining ("Are Ideas Getting Harder to Find?", Bloom-Jones-Reenen 2020); replication crisis and grant-admin overhead unresolved.

§4existing alternatives

Who is already working on this

9 entities are currently working on this problem across public markets, private companies, and research orgs. Each is evidence the market is real; none has obviously solved it.

AI-driven drug discovery. Industrializes biology through automated phenotypic screening at scale.

$2.4B

Ginkgo Bioworks

public · USA

Cell programming platform. Biosecurity Division runs metagenomic surveillance and pathogen detection at scale.

$350M

OpenAI

private · USA

Originally nonprofit research lab, now capped-profit. Safety and superalignment teams alongside capabilities work.

$157.0B

Replit

private · USA

Browser-based coding platform. 100M+ learners worldwide; AI agent for app-building compresses learn-to-ship cycle.

$3.0B

Insitro

private · USA

ML-driven drug discovery from Daphne Koller. Combines stem-cell biology with deep learning.

$2.5B

Hadrian

private · USA

Fully automated precision-parts factories for aerospace, defense, and infrastructure. Removing labor cost from the BOM.

undisclosed

Isomorphic Labs

private · UK

Alphabet-spinoff applying AlphaFold-class ML to drug design. Deals with Novartis and Eli Lilly.

undisclosed

§5proposed direction

If we solve this, here is the world we get

After · 20 years

Scientific output per dollar 10× higher via automated experiments, AI literature review, FROs, lottery + fast-grants reform. Trustworthy literature. New disciplines created on AI timescales.

Requests for startups · 3 concrete companies to build

The autonomous lab

A PhD spends most of their hours pipetting, not thinking. Build the closed-loop robotic wet-lab that runs experiments 24/7 and proposes the next one.

why now
Lab robotics + foundation models crossed the line where the design-build-test-learn loop can close without a human in the inner loop.
shape
A self-driving lab as a service: researchers submit hypotheses, the system designs, executes, and iterates experiments autonomously, returning results and next-best experiments.
success
Experimental throughput per scientist rises an order of magnitude in adopting fields.

The replication layer

Most published findings are never independently checked, so the literature quietly rots. Build the company that makes high-value replication fast, funded, and a default step.

why now
Autonomous labs + AI literature analysis make systematic replication economically feasible for the first time.
shape
A replication-as-a-service org funded by journals, funders, and firms that need a result to be true before they bet on it; outputs a trust score the field actually uses.
success
High-stakes results carry a replication status, and the replication rate of new work rises measurably.

Fast grants as a standing product

It takes 6–12 months to fund a science idea. Fast Grants proved it can take 48 hours without quality loss. Build the always-on micro-grant rail for frontier research.

why now
The Fast Grants experiment produced the playbook; no one operationalized it as durable infrastructure.
shape
A funding platform with rolling micro-grants, expert triage, and radically compressed decision latency, capitalized by philanthropies and firms wanting optionality on frontier work.
success
Time from research idea to funded work drops from quarters to days as a standing option.

full rubric + framing on the Requests for Startups page.

§6cost & scale

What the market can pay

The world is already paying $2.5T per year against this problem (global R&D spend (UNESCO Institute for Statistics) — productivity uplift is leveraged across all of this; high confidence).

A successful solution does not need to capture more — it needs to redirect a meaningful slice of existing spend, plus the latent willingness-to-pay implied by the severity score above. The cost ceiling for a real solution is bounded by this number; everything cheaper is dominated, everything more expensive is a non-starter.

§7safety & considerations

What could go wrong, and how we know we are not wrong

Section in progress

Failure modes, ethical considerations, and the conditions under which this whitepaper would be falsified are being authored as the weekly cadence ships. The Deutschian commitment: every claim above is a conjecture; we publish the conditions under which we would update. Unpublished sections stay empty. A Monday newsletter is not promised. Subscribe to get the upgrade, or contribute on GitHub.

§8suggested investors

Who would back this

Capital allocators with a stated thesis or deployed portfolio in this domain. This is a starting list — Exa Websets enrichment will expand it to direct check-writers per company.

Grant

Emergent Ventures

Fast grants. High-variance, unconventional, talent-first.

Grant

1517 Fund · Medici Project

Minimum $1k, no strings, no equity. A five-minute video is the whole application.

Fellowship

Thiel Fellowship

$100k to stop out of school and build something important.

Fellowship

O'Shaughnessy Fellowships

$100k for creators across domains, art, science, history, startups.

Accelerator

Activate Fellowship

Two-year salary and national-lab access for scientists taking PhD research to market.

Focused Research Org

Speculative Technologies

ARPA-style coordinated research programs for big-if-true platform tech.

Venture Studio

Deep Science Ventures

Outcome-first venture creation. Define the holy-grail outcome, then recruit founders.

Venture Capital

Lux Capital

Counter-conventional science at the edges of physics and biology.

Venture Capital

Fifty Years

Every portfolio company must solve a UN SDG. Profit with purpose, no exceptions.

§9voices

What the thinkers say

“Progress depends on the unhindered creation of good explanations. Institutions that reward criticism outperform those that reward consensus.”

— David Deutsch · Physicist & Philosopher

“First-principles reasoning beats reasoning by analogy. Most industries are throttled by process, not physics. The idiot-index gap is where opportunities live.”

— Elon Musk · Engineer & Founder

“Research productivity has been falling for decades and we under-study the process of progress itself. Metascience and institution design matter as much as any single discovery.”

— Patrick Collison · Co-founder & CEO, Stripe

“We need a new philosophy of progress that makes technological and scientific advancement a positive cultural project again, not something to apologize for.”

— Jason Crawford · Founder

“The Great Stagnation thesis: since roughly 1973 the low-hanging fruit of science and industry has been picked. Restoring the innovation engine is the meta-problem.”

— Tyler Cowen · Economist & Writer
§10sources & criticism invite

Where this is wrong, tell us

Every number on this page carries a source and a confidence tag. Every section open to refutation. If a citation is wrong, a number is stale, or a conjecture is unfounded — file a correction.

corrections → use the feedback widget in the nav · open issue at github.com/adamtpang/optimism.fun

Published whitepapers

One problem.Published essays, not a Monday promise.

The essays that exist are listed at /whitepapers. A new essay is not promised every Monday. Weeks with no essay are empty on purpose. Nothing here invents a back catalog.

optimism.fun uses your email address only to send low-volume project updates through Resend. You can unsubscribe with one click at any time; read the privacy policy.