Physics + AI: Revolutionising Physics Through Tech

4 min readAug 28, 2024

--

We’ve all known that AI (artificial intelligence) is an invaluable asset to our society since its entrance into our lives, mainly ChatGPT in November 2022, a type of generative AI that can answer your homework assignments and teach you how to bake the world’s best brownies. But how can AI and ML (machine learning) help physicists answer fundamental questions about ourselves and the universe we live in?

ML and AI have history in physics research since the earliest beginnings of AI itself, appearing as a key word in papers at the time. A key reason is the prevalence of large data sets in physics. LLMs (Large Language Models) are of great use in processing big data, a prominent example being the Large Hadron Collider and its associated high energy experiments at CERN. According to its own statistics, the CERN Data Centre processes over 30TB of data from the LHC a year, most of which does not consist of phenomena of interest to physics but nevertheless requires mass computing power to handle. This has been hugely benefited by development in machine learning. The CMS and ATLAS experiments, for example, in 2023, trained AI models to recognise the characteristics of jets, essentially a stream of particles, originating from known particles in order to find the atypical jets which could be key to new physics. Alternatively, they also taught AI to consider whole collision events and look for anomalous signals, the blips which could hint to something new.

ML being used to revolutionise the understanding of particle jets

Astronomy, one of the oldest disciplines in the world, is another field in physics which, like particle physics, has seen immense growth with machine learning. Since 1970, there have been more than 90 active telescopes, of varying scope and accuracy; however, the James Webb Space Telescope, the premier telescope of the decade, provides 235 GB of data daily, handled by hundreds of observatories worldwide to investigate the universe. This is one telescope of dozens active today. This process of finding the desired image out of thousands of our celestial backdrops can be vastly improved by training AI models using available images and simulations. An example discussed by Fermilab astrophysicist Brian Nord is finding objects exhibiting gravitational lensing, which identified the largest black hole ever found in March 2023. Gravitational lenses are notoriously difficult to identify by eye but the trained AI models are able to do this with ease and speed. Although, it is significant that there are issues with accuracy, as noted by Nord, that interpretable error bars cannot be generated from neural networks easily. Calibrated error bars would allow scientists to better identify the cause for error and train AI models to avoid this.

AI being used to discover gravitational lenses

Not only is AI used to expedite the processing of massive databases, neural networks, the fundamental concept of AI, has been used to model complex theories in physics. An example is SciNet, a neural network modified by scientists at ETH Zurich into a few parameter layers so each neuron contains separate information about the problem. Thus, not only can this neural network solve a given problem, it can also provide additional, fresh insight and conceptual knowledge at unprobed angles. The team of scientists behind SciNet expressed interest in looking at the issue of galaxy rotation curves, which spin faster than they should with calculations of visible matter. Using SciNet, various masses of the unseen hypothetical dark matter could be added or adapt theories of relativity, which astrophysicists could then interpret. This particular study shows the potential in neural networks as a guidance tool to narrow down theoretical physicists’ guesses.

Regardless of the form, the use of AI/ML will become a routine tool for every professional physicist, just like vector calculus and group theory. Like the pontifications in the 1940s, famously by Caltech physicist Richard Feynman, of how computers would be used in physics, this discussion will become void: tasks such as classifying images or finding outliers in data will be bygone and the daily tasks of physicists will be of a different form, just as computers meant no more hand drawn graphs. AI will be integrated into society and physics research as seamlessly as modern technology.

But could AI be used to solve some of the biggest questions in physics? Could an AI model come up with a unified theory of everything? Resolve Einstein’s general theory of relativity and Quantum Mechanics? Many scientists think not, that LLMs would not be capable of the intuitive links in current theory to create new ideas, hypothesis generation. Nevertheless, in the end, how the field of artificial intelligence will change physics will inevitably depend on the future of AI.

This article was written by Yuna Choi.

If you would like to contact her, reach out to her on LinkedIn: https://www.linkedin.com/in/yuna-choi-b99782283/

If you would also like to write an article, please contact us @youthinai.official@gmail.com

You can find our website for more information here: https://youth-in-ai.github.io/#

References:

Physics + AI: Revolutionising Physics Through Tech

Written by Youth in AI