Science is (partly) over. Will politics be? Reading Leviathan and the Air-Pump

I recently read Leviathan and the Air-Pump[1]. The dispute over scientific experiments described in this book can be matched up with our situation with AI today. In the book, Hobbes criticizes Boyle's air-pump experiments as flawed, not just in their results, but also in the whole design of the experiments and in their credibility. He thinks the scientific group Boyle defended is just a small group, with entry barriers, qualification checks, and all kinds of conditions, which interprets experiments to get the results it wants, so any experimental result can be made to fit. Hobbes thinks that if the different groups have no unified position, this will lead to division and civil war. Hobbes thinks that in both science and politics there needs to be an absolute authority to judge, to eliminate disputes, so that everything is absolutely right or wrong, with no room for argument. He thinks the best science is like geometry, where you start from some assumptions and can derive any conclusion, with no room for dispute.

Over the past few hundred years, this system of peer review in science has worked well, and we think that Boyle's side won. This also happens to match our democratic and liberal values. But what has happened recently in mathematics is that OpenAI, as well as Anthropic, has released a series of mathematical results[2] one after another, and in this field, AI systems have surpassed all purely human mathematicians. In mathematics, this relies on a mathematical language called Lean, which makes mathematics formalized and verifiable, plus large-scale training and agents running in parallel. Within this field, large-scale AI systems have already reached Hobbes's ideal vision of geometry, a supreme authority whose word is final and which anyone can verify.

For now, this supreme authority has moved out of the small circle of mathematicians, but who is the sovereign? On the surface, there is a mathematical verification system, a formal verification language. But in reality, the key is the speed of production. Although other humans can in theory also produce these formally verified results, and on the surface it looks like a public standard, in front of OpenAI's massive volume of proofs and large-scale agents, results published by individuals become trivial. The authority has become whoever controls the largest-scale AI.

Now let's go back to politics. Since in science authority has already won out over the small circle, what reason do we have to think politics won't go the same way? Of course we can say there are all kinds of value claims in politics, but I won't discuss specific claims. As far as what the authors of this book want to say is concerned, the norms in science can to a large extent be carried over into politics. They are similar things. So if you defend some value of freedom in politics, that defense can also be carried over into science, but within science it has now already failed. The question is, facing rapidly advancing AI systems, to what extent will we still keep our independence in serious political judgment and decision-making?

  1. Shapin, Steven, and Simon Schaffer. Leviathan and the Air-Pump: Hobbes, Boyle, and the Experimental Life. Princeton: Princeton University Press, 2011. First published 1985. ↩
  2. OpenAI. "Working with Academic Communities to Build Scientific Understanding." OpenAI, October 26, 2026. https://openai.com/index/building-scientific-understanding-with-academic-communities/ ↩