Quantum Physics and Deep Learning, What are possibilities?

Question

I would like to ask to you a question that I have been wondering about for a while. What would be if we could record all the quantum behaviors (momentum, rotation, translation, etc) of atomic (or subatomic) particles over a certain period of time (long enough) and the values of the interactions between each other (attraction, repulsion, etc) and what would be if we could create a model with machine (or deep) learning? Can such a model make the uncertainty in the behavior of subatomic particles certain? And could it describe what is being studied and known as the "Theory of Everything"?

When I asked the question to a gpt model (ChatGPT) for a quick insight, he asserted the uncertainty principle and claimed that the model could not go beyond it. The answer seemed reasonable, but I'm still curious about the valued opinions ones who working on quantum physics and/or machine learning. (Any tips or advice (article, conference, etc) to determine research direction would also be valuable (if any)). (I am not sure whether here is correct place to ask the question)

Answer 1

What would be if we could record all the quantum behaviors (momentum, rotation, translation, etc) of atomic (or subatomic) particles over a certain period of time (long enough) and the values of the interactions between each other (attraction, repulsion, etc)

This is already not possible. Not just in practice, in principle. What would you be recording them WITH? What would you be measuring them WITH? Well... presumably, other things made up of quantum objects. So are you recording the quantum things in the environment, AND ALSO recording the quantum things in the measuring devices? I propose to you that you cannot do all of that, in any world, no matter what. You cannot record EVERYTHING that happens physically inside this universe using only other stuff inside this universe.

Answer 2

You can already pose your question for the simple case of tossing a fair dice again and again for a certain duration. You will get that the result converges to 50:50 and that no individual pattern can be predicted for the next elementary event.

According to quantum mechanics, in your setting there is no law which determines the individual event, not only that we do not know the law. It is obvious that this result from physics does not depend on the method for analyzing the data. Hence combining the question with deep learning is not relevant.

Answer 3

I ran a similar experiment recently, typing the contents of all of last years' editions of The Racing Post into a machine-readable file, which I uploaded to ChatGPT in the hope of getting a firm prediction for the 2.30 at Haydock Park. Sadly, the hopeless bot was only able to give me poor odds and some made-up gossip from one of the stables. However, the experience put me in a good position to develop fresh insights, which I can share with you as follows.

There are several obstacle that would prevent your suggested scheme from leading to a theory of everything.

The uncertainty principle does not tell us that nature is certain and our understanding of it is not- it tells us that nature is uncertain. Studying the uncertain behaviour of trillions of particles in the hope of divining a theory of everything is like studying a trillion roulette wheels in the hope that will tell you the result of the next spin of the wheel.

Another challenge is that you would need to have observations from a representative range of interactions between particles, and there the problem is that we are not able to recreate particle interactions at very high energy ranges. You might be aware of the circular 'Large Hadron Collider' at CERN, which has a diameter of about five miles. To explore some theories of physics it would be necessary to have a particle accelerator with a diameter large enough to enclose the Solar System, which does not appear to be a feasible proposition.

To further the difficulties facing your suggestion, a theory of everything would have to take into account gravity, and gravitational effects between subatomic particles are so small that they would not be discernible in the observational data. Were you to attempt to overcome that problem by feeding the AI model with data about the stars etc, you would be thwarted by the fact that we can no longer observe the entire Universe, so there may be important large scale phenomena that we are prevented ever from observing. Also, there are aspects of nature we have never observed directly- such as dark matter, or the behaviour of matter inside black holes- and they might be forever out of our observational reach.

If those obstacles were not a sufficient deterrent, you might also want to bear in mind that ChatGPT works essentially by trawling existing sources of information and regurgitating the patterns within it, subject to a system of weightings. Its inherent capabilities are therefore woefully inadequate for the sort of task you have in mind.

Answer 4

I think there are people more qualified than philosophers to answer the question. But prima facie if quantum fluctuations are reliably probabilistic then no amount of deep learning will help you.

Answer 5

I am afraid ChatGPT was right about this one. The uncertainty principle indeed is a fundamental property of the Universe. According to this principle, nothing, not even the Universe itself, can predict where we are going to find the particle the next time we decide to look for it. With non-zero probability it could be anywhere in the Universe.

Therefore, even if we recorded every single interaction that ever happened, the required information will not be there -- nor it will be deductible, through deep learning or otherwise.

Worse still, the advantage of deep learning models in particular is in their speed -- but this comes at the cost of precision. And using an imprecise model can only increase the uncertainty of our predictions.

Finally, it pays to remember that on macro level the quantum uncertainty quickly diminishes and the Universe becomes effectively deterministic.