Why Inventing General Relativity Is the Final Test for AI - Adam Brown
Adam Brown argues that inventing general relativity from Newtonian physics may be the ultimate benchmark for AI intelligence. He suggests LLMs are interpolators operating at increasingly higher levels of abstraction, and that achieving this feat — possibly within 10 years — would signal AI has fully encompassed human intelligence.
Summary
In this short clip, Adam Brown presents a compelling framing for what the ultimate test of AI capability might look like: the ability to derive general relativity from the laws of Newtonian physics, mirroring Einstein's greatest intellectual leap.
Brown begins by characterizing LLMs as interpolators, but crucially notes that the level of abstraction at which they interpolate keeps rising. He suggests that from a sufficiently elevated conceptual vantage point, even the invention of general relativity could be understood as a form of interpolation — just at an extraordinarily high level of abstraction. This reframing challenges the common assumption that such a discovery is categorically beyond algorithmic reach.
He describes general relativity as 'the greatest leap that humanity ever made,' making it a natural capstone benchmark for machine intelligence. Brown tentatively estimates this capability could emerge within roughly 10 years, though he acknowledges uncertainty. He concludes by noting that while there are clear disanalogies between human cognition and LLMs, at the right level of abstraction, the distinction may become less meaningful — suggesting that the character of the achievement may differ from Einstein's, but the functional equivalence could still hold.
Key Insights
- Brown argues that LLMs may be able to 'invent' general relativity because from a sufficiently elevated level of abstraction, even that leap can be understood as interpolation.
- Brown characterizes LLMs as interpolators whose level of abstraction keeps rising, suggesting intelligence itself may be a function of abstraction height rather than a qualitatively different process.
- Brown identifies the ability to derive general relativity from Newtonian physics as the final and most demanding test for AI, calling it 'the greatest leap that humanity ever made.'
- Brown tentatively estimates that AI could achieve this capability in approximately 10 years, at which point he believes it would have 'fully encompassed human intelligence.'
- Brown acknowledges clear disanalogies between human intelligence and LLMs but suggests that at the right level of abstraction, the distinction may not fundamentally matter for evaluating AI capability.
Topics
Full transcript available for MurmurCast members
Sign Up to Access