AI companies want to become frontier scientific researchers
Part of the cover package on AI and science. Reported facts: Google DeepMind’s work has made Nature’s cover nine times and AlphaFold won Demis Hassabis a share of the 2024 Nobel; OpenAI this autumn claimed a Millennium Prize problem and solutions to 300+ open questions; Dario Amodei said in September Claude found a potential CRISPR successor; Tokyo’s Sakana AI reported in Nature that its “AI scientist” paper passed first-round review at a top AI conference; DeepMind spin-out Isomorphic Labs has 17 drug candidates in development. The Economist’s interpretation is that labs chase science mainly to improve their models, because benchmarks are saturating and verifiable maths suits reinforcement learning; some speculate OpenAI’s paper dump was a by-product of a training run. Caveats: superhuman maths may make AI ability more “jagged”, and frontier-scale science may be out of reach for all but the richest firms. Anthropic’s Jack Clark predicts AI will be able to fully improve AI by 2028—a forecast, not a result.
Why it matters
AI-for-science is becoming a frontier-lab battleground, which squeezes or validates AI-for-science startups (Sakana, Isomorphic and many Dealroom-tracked biotech/materials companies). The cover leader (“What if the machines take over science?”) argues the same theme editorially and isn’t counted separately.
