Scientists Start to Use AI to Answer Complex Physics Problems
While AI might be fun to use to make a person in an old photograph sing merrily, or to glam-up your Facebook posts, it is far more powerful than that and has the potential to answer questions that are beyond the brightest minds.
Physics is filled with phenomena and processes that are simple enough for us to predict the outcome off, but that we don’t really understand the mechanics of. Take, for instance, the simple phase change in materials. We know that certain circumstances such as temperature and pressure make a material modify its form, but the driving forces behind the change, and the rate at which it happens may not be so well understood. We know that water will change from its solid form (ice) to its liquid form at zero degrees centigrade and one atmosphere, and will then transform once more to its gaseous form (steam) at 100° centigrade and one atmosphere.
Developing Exotic Materials
But water is a very simple case and it can only yield one product at each step – ice, water, or steam. Consider the same phase changes in a steel alloy, with a range of different elements within it, which may act differently according to the type and amounts of elements and changes to the pressures and even environmental gases. This is a much bigger deal and the chemistry becomes so complex that it hasn’t been possible for scientists and researchers to predict what might happen.
Given the desire for increasingly complex and exotic materials, the need to be able to understand what will happen without having to try different combinations and system parameters becomes important. Imagine a very lightweight material that can withstand the temperatures, pressures and thermal loading of a jet engine; we don’t know if such a material exists, but to be able to identify if one were possible would be a leap forward in materials engineering. We just don’t have the brain power to compute the myriad possibilities.
But AI does, and there is increasing interest in using its power to help drastically reduce the time and effort needed to pinpoint such exotic materials.
Worlds Away
The James Webb telescope has delivered the hardware that could identify either potentially habitable planets outside of our local solar system, or the tell-tale signs of intelligent life in the stars around us. Unfortunately, what it has produced so far is such a mass of information that it would take researchers hundreds of years to assess what we already have, and with more data being added on a daily basis, it becomes a mammoth task. Once again, recent advances in AI have given astrophysicists a tool that will automatically sift and look for candidates amongst the huge dataset with increasing accuracy.
There are hundreds of physics and science-related fields that AI could become a major tool in not only searching for information, but also predicts the outcome of chemical reactions and can be used to identify new and exotic alloys. But this potential has become a whole lot more likely with generative AI.
Generative AI is a form of algorithmic artificial intelligence that generates a range of new output based on the type of data they have been trained on. This type of AI is capable of generating text, images, videos, or other data using generative models, often in response to fairly specific prompts. Unlike traditional AI systems that are designed to recognize patterns and make predictions, generative AI creates new content in the form of images, text, and audio inputs. Plainly, the use of this in technical fields means that researchers can have their AI systems more focused on specifics rather than simply spending time and effort on unnecessary information. That makes both learning and searching much faster and results more accurate.
Creating Accurate Systems
This kind of system can help scientists quickly investigate the thermodynamic properties of new and exciting materials or detect entanglement in quantum systems – a new and very promising area of physics. Ultimately, this technique could make it possible for scientists to discover unknown phases of matter autonomously, which could then be synthesised in the laboratory to prove their existence and to test their properties.
Frank Schäfer, a postdoctoral researcher in the Julia Lab in the Computer Science and Artificial Intelligence Laboratory (CSAIL), said “If you have a new system with fully unknown properties, how would you choose which observable quantity to study? The hope, at least with data-driven tools, is that you could scan large new systems in an automated way, and it will point you to important changes in the system. This might be a tool in the pipeline of automated scientific discovery of new, exotic properties of phases”. The generative AI system has also attracted the attention of researchers at MIT, and they are starting to get good results.
Phase transitions in materials are an obvious place to start with introducing generative AI into the mix, because the potential for many easy wins with new materials is tantalising. Materials also offer a well understood key to the process in that they have an “order parameter”. In terms of materials, an order parameter is a physical attribute that is significant and is anticipated to undergo change, and it can be used to identify these transitions.
Water, for example, undergoes a phase transition from liquid to solid (ice) when its temperature falls below 0° Celsius. In this scenario, an appropriate order parameter could be established by comparing the proportion of water molecules that are a part of the crystalline lattice to those that are in a disordered state. By getting generative AI to understand how these parameters impact a material, and how the development of a new material is driven by them, understanding the outcomes of different alloying elements becomes significantly easier.
Generative models, such as those that underpin ChatGPT and Dall-E, typically operate by estimating the probability distribution of a set of data. They then use this estimate to produce new data points that align with the range. By having a known function within the generative model, a great amount of computer guess work can be removed, making the process far faster.
Simplified Game Development
All of these steps are making AI not only more accurate but faster too, and that can only be a good thing for not only the furtherment of science, but could become a permanent feature in many other computer-driven areas. Take, gaming, for instance; Generative AI could be used to hunt bugs in the many millions of lines of code in a game in much the same way as it uses known features in chemical reactions to predict outcomes.
We at Unity Developers are keen to understand the potential for generative AI, and how it will positively impact our projects. Keep checking back for updates.
