AI is pulling off increasingly impressive intellectual feats. It can write sophisticated software, solve difficult scientific problems, and make discoveries that would have taken enormous human effort just a few years ago.
The achievements in mathematics are especially striking. AIs are tackling problems that have stumped mathematicians for decades, sometimes finding proofs or solutions no human had managed to discover.
But so far, we’ve seen little comparable progress in the world of grand unifying theories in science.
I’ve discussed elsewhere why we should expect such breakthroughs to be radically more difficult. A big part of it is that, unlike in mathematics, the problem isn’t necessarily sitting there as a conjecture waiting to be proved or disproved.
In science, some of the greatest discoveries concern phenomena the field already thinks it has explained. When I proposed that primate color vision evolved as an empathic sense, for example, the field thought it already knew why we evolved color vision.
And sometimes the discoveries concern phenomena nobody even realized needed explaining. When I began investigating why letters are shaped the way they are, practically nobody had even thought to ask the question.
Doing groundbreaking unifying science, then, often means being asked to solve a problem the field thinks is already solved - or doesn’t even recognize as a problem in the first place.
That’s a very different intellectual challenge from being handed the Riemann Hypothesis and asked to prove it.
I’d like to put this as a fairly concrete challenge to AI. Surely AIs will get there eventually. And perhaps stating the challenge explicitly will even help accelerate things.
Here it is.
THE CHALLENGE
You have access to the entirety of scientific knowledge, across every discipline.
You are not being given a particular field, an unsolved problem, a phenomenon to investigate, or even a hint about what kind of explanation to pursue.
Your task is to make a major, original, unifying scientific discovery.
1. PHENOMENON
Recognize an important phenomenon that deserves a deeper explanation.
Perhaps the field already has explanations, but you see that they’re inadequate, conceptually or empirically.
Or perhaps the field hasn’t even noticed there’s a phenomenon requiring explanation at all.
2. NEW PRINCIPLE
Discover a new, elegant explanatory principle that changes our understanding of the phenomenon, perhaps overturning the prevailing explanation altogether.
The principle should bring together facts that previously seemed unrelated, revealing an underlying order nobody had recognized.
3. PREDICTIONS
Derive surprising, specific predictions from the principle.
These should be genuine consequences of the theory, not just known facts fitted into it after the fact.
4. EVIDENCE
Find or generate empirical evidence that distinguishes your theory from existing alternatives.
Show that your new principle doesn’t just fit the evidence, but is simpler than existing explanations while explaining more - accounting for more phenomena with fewer independent assumptions, and making predictions that competing theories don’t equally explain.
⸻
That’s it.
No assigned problem. No hints. The entirety of science is your playground.
Find something important that science doesn’t yet understand, even if it thinks it does. And explain it better than anyone has before.
That’s the AI Grand Theory Challenge.
Related video
Why science won’t fall as easily to AI as math. Moment 652



Do you follow the German theoretical physicist Sabine Hossebfelder on Youtube? She explores many of the same areas you are: theoretical physics, applied physics, math, energy, neuroscience, AI, and more. I'd love to see one of you interview the other.