AI in Physics: A Promising Future for Academia?

[kelly0303]

Hello! I see more and more papers that use AI for physics analysis or to find approximate solutions that would otherwise be impossible or very slow to find. What do you think is the future for AI in physics on long term. Do you think it would become a field on its own within physics? Would it help you to get a job in academia?

 

[StatGuy2000]

To the OP:

I'm not by any means an expert in AI, but have been keeping in touch from a distance on developments in that field. My own take is that AI, and more specifically machine learning, has transitioned from a primarily academic field of research within computer science into a much more applied field, with wide applicability in a variety of areas, including physics.

I don't foresee AI becoming its own field within physics, but I do see much greater scope for collaboration between AI researchers and physicists, and the collaboration will go both ways -- AI researchers working with physicists on, say, large scale data analysis from high-energy particle physics or astrophysics, as one example, and statistical physicists applying their understanding of interactions at the macroscopic level (and the math behind it) into problems related to AI (as already have been done in the area of neural networks).

 

[Päällikkö]

It depends on what you classify as physics. Finding fundamental laws of nature and formulating theories, I don't see AI or ML doing. However, viewed as a powerful tool of statistics and prediction, I can certainly see these tools being used to supplement in computational physics, finding new minima and predicting new structures, and replacing complicated (e.g. PDE) solvers with something that can quickly and accurately guess the solution.

For example, until relatively recently, the field of protein folding was dominated by tools built by physicists and computational chemists, whereas with the advent of FoldIt! and beyond, I see approaches not having to do much at all with physics taking an increasingly important role, and indeed I think physics and the type of bottom up approach inherent to the field is going to be giving quite a bit of way to the magic black boxes of machine learning in the near future.

 

[gleem]

From The American Physical Society New , June 2018 (Volume 27, Number 6)

"AI Makes Inroads in Physics "
https://www.aps.org/publications/apsnews/201806/artificial.cfm

applications to LIGO, LHC and astronomy

 

[kelly0303]

From The American Physical Society New , June 2018 (Volume 27, Number 6)

"AI Makes Inroads in Physics "
https://www.aps.org/publications/apsnews/201806/artificial.cfm

applications to LIGO, LHC and astronomy

Thank you for this! So my main concern is this: If I join during my PhD a group who works in applying AI techniques to, say, condensed matter physics (I saw many papers in that directions) and I do a good job, find better approximations, maybe even deeper insights in understanding more complex systems, will I have a chance in academia, later, with this kind of PhD? Will this kind of research prove my knowledge in the field (which in this case would be condensed matter) or will the universities think that I am good in AI, but will be reluctant about my knowledge of the physics behind and prefer to choose for a given position someone who doesn't know much AI, but his PhD was more directly related to the field (e.g. doing some condensed matter experiments in the lab).

 

[Vanadium 50]

It is difficult to determine how people will react to an as-yet undone piece of work.

 

[CrysPhys]

Thank you for this! So my main concern is this: If I join during my PhD a group who works in applying AI techniques to, say, condensed matter physics (I saw many papers in that directions) and I do a good job, find better approximations, maybe even deeper insights in understanding more complex systems, will I have a chance in academia, later, with this kind of PhD? Will this kind of research prove my knowledge in the field (which in this case would be condensed matter) or will the universities think that I am good in AI, but will be reluctant about my knowledge of the physics behind and prefer to choose for a given position someone who doesn't know much AI, but his PhD was more directly related to the field (e.g. doing some condensed matter experiments in the lab).

Another major unknown is what initiatives the major funding organizations (such as NSF and DARPA) will be supporting at the time you've completed your PhD and postdocs and are applying for faculty positions. If there is money allocated for AI/Physics programs, you'll be in a good position. If not, ...