[00:00] Artificial intelligence is about to hit science,  big time. After it has taken over mathematics,   it’s now coming for physics. And that  is having a very interesting start. [00:12] There have now been so many AI-derived proofs  that I have lost track. At this point no one can   deny any more that AI can deliver new and  correct mathematical proofs. The arXiv is   being flooded with AI generated maths papers.  Mathematicians are complaining that AI proofs   [00:29] are incomprehensible and inelegant slop while  they weep into their notepads. Then, last week,   OpenAI released a new batch of proofs,  one that makes steps towards resolving   a long-standing physics problem:  whether Yang-Mills theories exist. [00:46] How relevant is this? Well, Yang-Mills  theories are a type of theory that we   use in the standard model of particle  physics. So you may reasonably ask, OK,   how can they possibly not exist given  that the standard model works so well? [01:01] The reason is that what mathematicians mean by  “existence” is not the same that scientists,   or indeed everyday people mean. For a scientist  something exists if it is a useful description   [01:14] of nature. For a mathematician it means to  prove that an object with certain mathematical   properties can consistently be defined. It might  totally be that the mathematical object of a   [01:26] Yang-Mills theory does not exist, but that the  calculations that physicists have derived from   the idea of Yang-Mills theories still work  in the sense of making correct predictions.   You would then expect that the standard model is  technically something close to Yang-Mills theory. [01:43] OpenAI didn’t actually settle the case, at  least that is my understanding of the situation,   as they haven’t told us just exactly  what they did. But physicists went   and complained immediately that it’s got  no relevance for physics. And just like   [01:57] mathematicians reacted to the Navier-Stokes  proof, I think they aren’t quite honest.   Because if this result had been put forward  by a human being, they’d have celebrated it. Meanwhile physicists themselves have made good  headway on dismantling their own discipline. You   [02:15] see, for decades, those physicists who work in  the foundations have been publishing bullshit   papers. By this I mean not papers that are  wrong, though some of them are. But I mean   papers that are irrelevant, mathematical fiction  that teaches us nothing about reality. They   [02:30] invent new particles or forces or, indeed,  entire universes, and pretend it’s science. it is simple and you can get it published and  that’s a job. This nonsense paper production has   [02:46] completely stalled progress in the field.  But because it’s mathematically simple   and doesn’t require much creativity,  it is something that AI can do now. It’s no longer hypothetical. One physicist  has used AI to write 100 papers about the   [03:03] one detection of a potential dark matter  candidate, just for fun. One group has put   forward a custom AI to cook up new particle  physics theories. Another group did the same   for modified gravity and dark matter. New  theories on demand. And yet someone else has   [03:19] used AI to write a new physics paper a day. And  the thing is, these papers are now by technical   standards basically indistinguishable from  all the other nonsense that gets published. What happens next, then, is going to be  interesting. This sudden oversupply of   [03:36] irrelevant papers will force physicists to  acknowledge that most of their own work is   also irrelevant. They will then have to  come up with new quality standards. And   this could finally, finally move the  field forward. Maybe the first sign of   [03:51] physical superintelligence will be an AI that  refuses to generate another dark matter model. What quality standards? Well, I explained in my  book Lost in Math what would be a useful strategy,   [04:03] namely, to focus on resolving  mathematical inconsistencies.   This is also why I think that  the OpenAI Yang Mills result,   whatever exactly that is, might eventually  turn out to be useful for physics as well. [04:16] But this also makes clear where the current  bottleneck is. It’s that in contrast to   mathematics, physicists don’t have clearly  defined problems of inconsistency that you   [04:28] can feed into an AI. They instead have  a lot of ill-defined pseudo-problems.   Like for example the question of why the  fine structure constant is what it is. We just recently saw an AI-generated paper  about this problem. It puts forward a super   [04:44] contrived numerical construction. The thing  is though, in terms of explanatory power it   is very difficult to beat a constant of nature.  Yes, the fine structure constant is actually   energy-dependent but that doesn’t matter for  my point. My point is that if your explanation   [05:00] is more complicated than the thing you are  trying to explain, it is not good science. This paper about the fine structure  constant is interesting nevertheless,   not because of its content but because of its  origin. It comes from Alexander Wissner-Gross,   [05:16] that’s the guy who coined the phrase  math is cooked. “Math is cooked”  He has a PhD in physics and with two dozen  other physicists, he is running an institute   for physical superintelligence. Their mission  is to bring us the second golden era of physics. [05:33] Nice idea. The problem is though,  that AI has been trained on decades   of mistakes in the foundations of physics.  And it’s being prompted by people who have   made these mistakes. That’s not  a good starting point. However,   [05:47] I also know that Claude has been trained on my  book “Lost in Math” because Anthropic admitted   to violating my copyright. This is why I remain  hopeful and on standby for the second golden   [05:59] era of physics. Maybe they should start by  figuring out why the first golden era ended. Mathematics is beautiful and that should be true  not just for the meaning but also for the type   [06:12] setting. If you want beautiful equations for your  papers, slides or videos, I recommend you use   Radical Pie from today's sponsor. Radical Pie is  an editor for equations where what you see is what   [06:25] you get. It produces instant professional results  that rival and often exceed the output quality   of LaTeX. But it's so much easier to use. Every  feature is available through the user interface   [06:37] where you find a huge variety of mathematical  and scientific symbols. And you can adjust all   the fine details like spacing, line thickness, the  shapes of arrows, and even how much an integral   leans forward. But wait, there's more. Radical  Pie also has tools that let you draw lines and   [06:54] shapes directly inside your equations. It's an  amazingly simple to use yet powerful tool. And   the best thing, it's not a subscription model. You  buy it once and it's yours. Go to radicalpie.com   [07:07] to download a free trial. And if you use my  promo code SABINE, you'll get 15% off your   personal copy that works forever. So go and check  this out. Thanks for watching. See you tomorrow.