This is a special post for quick takes by Lokesh. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Lokesh's Shortform
7Lokesh
4dscft
1Lokesh
2dscft
1Lokesh
3dscft
2james oofou
1Lokesh
One reason advanced AI might not be able to figure out "all the physics" is that the observation and analysis of "laws of universe" may not be constrained by intelligence but rather time. It could be the case that there are some regular patterns periodic in billions of years. Intelligence would not be able to do much, in this case.
1Physics is not something that you solve by “acquiring more data”. The underlying laws are incredibly simple. Famously Einstein derived both the special and general theory of relativity almost only from pure thought. Experimental data is needed as confirmation. As a theoretical physicist myself, I have no doubt that ASI will be able to essentially figure out “all of physics”.
Thanks for replying. I think I can see where our disagreement is. I'm afraid the way science has worked so far may not be the case in the future. Somehow we have arrived at *almost right laws only from pure thought. Reality may be way more complicated which we may only be able to figure out by acquiring more data.
The progress in fundamental physics has always been towards increasing simplicity and this is not going to change. The universe is an incredibly constrained system and it is almost impossible to come up with a consistent unifying model (we are still looking for it). We already have more than enough data for a sufficiently intelligent entity to figure out "the theory of everything", it is only a lack of imagination.
I would be *very surprised if it turned out to be true, but I grant that you could be correct, although I would assign a very low probability to it.
You are entitled to your opinion but I'm a tenured theoretical physicist and the view I expressed is not controversial and shared by most people in the field. Most theoretical physicists are really scared or excited by AI. I believe it is an almost consensus in the community that AI will eventually "solve physics", and many believe that this could happen fairly soon.
I think yes. I am assuming only, let's say, an oscillatory behaviour with a time period T = a billion years. Observing this phenomenon itself should take us that much time, and it is distinct from the problem of induction, at least in this case.