Charting AI’s Role in Scientific Discovery
Renaissance Philanthropy, with support from Google.org, is conducting a landscape study of AI integration in scientific research — and we want your perspective.
AI’s momentum in scientific research is hard to overstate. AlphaFold is accelerating progress in protein structure prediction and earned its inventors a Nobel Prize. Drug discovery is moving from screening existing molecules to using AI to design entirely new antibiotics that appear in no chemical library. Co-scientists, workbenches and long-horizon research agents are proliferating, searching, synthesizing, and reasoning over literature faster than any human team could. At the frontier, self-driving labs are closing the loop between hypothesis and experiment, running discovery cycles with minimal human intervention. Taken together, it looks like the scientific enterprise is being remade in real time.
Broad but shallow progress
Yet the transformation is uneven, both across domains and within individual domains. Even in fields like biology, real wins seem to cluster where data is clean and verifiable. Adoption stalls in harder, messier territory, like generative design or complex and regulated environments. Binding constraints are shifting away from model capability to data infrastructure, as fields lack both the right kinds of data and the incentives to produce it as a public good. Challenges are also being voiced at the metascience level, with some arguing that scientific systems and institutions are no longer fit for purpose, as AI becomes a forcing function laying bare structural gaps.
Progress, then, seems broad but shallow — and for the most part, concentrated in a few domains. It also remains unclear if bottlenecks to progress are consistent across fields. In one field, the binding constraint might be the absence of building blocks like shared benchmarks; in another, cultural or institutional blockers might translate into weaker adoption; and in others, progress might hinge upon translating in-silico results into trustworthy outputs for a lab or a decision-maker. More data, skills and compute remain compelling hypotheses for advancing AI-driven discovery. But as a blanket prescription, this set may not capture the highest-leverage interventions across the range of scientific domains.
Five domains, chosen for leverage
To investigate, Renaissance Philanthropy, with support from Google.org, is studying where AI is becoming a structural enabler of scientific discovery — and where it isn’t. Specifically, we've selected five domains for a deeper look:
Structural Biology — a relatively mature, AI-native field with established infrastructure, a large community, and demonstrated impact
Mathematics — a recent wave of breakthrough results in formal methods, theorem proving and verification, with fast-building momentum
Neurodegenerative Diseases — immense disease burden with few effective therapies despite AI-enabled progress in related scientific domains
Climate Science — among the defining challenges of our time, yet disproportionately underexplored, despite AI/ML approaches establishing state-of-the-art results
Autonomous or Self-Driving Labs — the frontier of autonomy, with the promise of closed-loop experimentation testing how far AI can drive discovery itself
We chose these to span the spectrum — varying data regime, maturity, methodology — while focusing on areas where targeted progress could have outsized impact in the coming years. This study also builds on our broader AI for Science work across the ecosystem, including our AI for Science Datasets call, the AI for Math Fund, the Open Source for Science Fund, the call-to-action for Autonomous Scientific Instruments, and our Advanced Research for Climate Emergencies portfolio.
Who we want to hear from
If your work touches one of these five domains, we want to hear from you. Our goal is to pinpoint where interventions stand to have the most impact, so for each domain we’re seeking people across a variety of roles:
Researchers or engineers, building or using AI/ML in their science
Infrastructure and tooling builders, working on datasets, benchmarks, platforms
Funders, deciding upon capital flows and allocation
Policy and governance, shaping protocols and standards
Importantly, we’re interested in hearing from voices across the adoption curve. We’re not just looking for AI proponents. Some of the most valuable signals will come from those who are skeptical of these tools and their utility: understanding where AI is not creating value in a given domain matters as much as understanding where it does.
Two ways to help
If you’re working in these fields, you can help shape this work in two ways:
Sign up to engage in this study. This will either be an interview or a survey — ongoing now throughout September 2026. No preparation needed. aixscience.renphil.org/engagement
Refer someone whose view would sharpen the picture. Introductions go a long way! aixscience.renphil.org/referral
The findings from this study will directly inform funding recommendations to major institutions around the world. The landscape report will be published in late 2026 as a public good: a sharper, shared map of where AI can move science forward — and what it will take to get there. Help us build that map.
Thoughts? Questions? Suggestions? Please reach out at aixscience@renphil.org