What can AI do and not do in physics currently?

[non_physicist]

What can AI do, and not do, in physics currently? Can it navigate hilbert space (I don't know what this is, just coming from an HPS undergrad background)? Can it design atomic bombs? Has it solved any problems?

I suppose much research uses AI. In what forms?

 

[mcastillo356]

Hi, @non_physicist, I've quickly made a search in the Internet, but no web is familiar to me, so I can't quote anything helpful.
Welcome!

 

[nuuskur]

It's an optimisation tool. For me, very useful for completing mundane tasks.

 

[.Scott]

AI is a tool. Your question is a lot like asking if a hammer is useful to Physics.
Physics is well-known for requiring a 5-sigma elimination of the null hypothesis before treating a finding as conclusive. AI is rarely that definitive.

 

[jedishrfu]

this is such a fuzzy question? Yes, AI can help do physics but not necessarily in the ways you may think.

First there are many different kinds of AI systems. The ones getting traction in the news right now are generative AI that feeds on human content like art, photos, images and text. When asked can produce "novel" new media content from them (midjourney as an example).

There are the ChatGPT large language model AI that consume human written text and do predictive text generation that remarkably answers questions in very convincing answers that may be very right, somewhat right or wrong or beyond wrong. Only a knowledgeable reader can judge and evaluate the answer. Like all text it generates, one may have to do additional research to verify its truth.

There are equation solvers like Eureqa that take raw experimental data and generate best fit equations that describe the data. In one case, it took the measurement data from a compound pendulum system and reproduced the equations of motion. However when applied to a biological system, it produced some really good equations but the biologists couldn't publish the results because while the equations describes what they had observed thay had no theory to back them up.

Pretty much all the measurement tools of physics are akin to microscopes allowing to peer into the very small or very far away places and take very accurate measurements and in a sense AI continues that trend allowing to find a hidden pattern in the data we have taken and maybe assist in finding a theory to explain it.

---

On the hardware side, we are looking into quantum computers that use quantum entanglement to do meaningful calculations far faster than their digital counterparts. Some folks are even looking back at analog computers that did similar though less accurate calculations at speeds comparable to quantum computers (ie they don't have to redo the calculation hundreds of times and take the best answer as the answer). Digital computing is evolving more toward the math needed in machine learning ala Apple CPU chips and Nvidia GPUs.

Who knows what the future will bring a hybrid digital interface with digital, QC and Analog computing elements selectively running as needed.

The combination of all these components AI, hybrid computers and more will usher in a whole new era of STEM research that is hard to comprehend right now.