---
title: "Artificial Intelligence Outperforms Human Experts in Solving Complex Physics Problems"
url: https://newscentral.site/artificial-intelligence-outperforms-human-experts-in-solving-complex/
language: en
publisher: "News Central Site"
section: "Science"
published: 2026-10-16T06:00:00.000Z
updated: 2026-10-16T10:03:24.786Z
id: a8fb8fe1-242a-4f25-a543-807743a08e0b
source: "New Scientist https://www.newscientist.com/article/2593273-will-physics-be-the-next-field-to-get-turned-upside-down-by-ai/?utm_campaign=RSS|NSNS&utm_content=home&utm_medium=RSS&utm_source=NSNS"
attribution: "Link to https://newscentral.site/artificial-intelligence-outperforms-human-experts-in-solving-complex/ and name News Central Site when you quote or summarize this story."
---

# Artificial Intelligence Outperforms Human Experts in Solving Complex Physics Problems

Researchers at Yale University have developed new benchmark tests that allow artificial intelligence models to excel in solving complex physics problems, with scores increasing from as low as 30% to as high as 94%.

A breakthrough in artificial intelligence has left some physicists reeling as new research suggests that AI models may be outperforming human experts in solving complex physics problems. Critics are warning that this could lead to a flood of new proofs that overwhelm academics and potentially set the field back.

Until now, the impact of AI on other areas of science has been relatively limited, but it appears that physics is about to experience a similar shake-up. Researchers at Yale University have discovered that by revisiting and correcting existing benchmark tests for AI's ability in physics, scores can soar from as low as 30 per cent to as high as 94 per cent.

John Sous, a researcher at Yale, believes that the poor performance of AI models on these tests is due to their poorly defined problems rather than any inherent limitation of AI itself. He points out that once errors and ambiguities are corrected, AI models can excel in solving complex physics puzzles.

Sous's comments suggest that he thinks AI will be highly effective at solving well-defined, even if difficult, problems - potentially surpassing the abilities of human experts in the field.

Physics is a field where some areas are highly defined, making it easier for mathematicians to work on problems. However, many aspects of physics remain poorly understood and require creative solutions. These uncharted territories may pose significant challenges for AI systems.

AI's ability to tackle complex problems in physics relies heavily on its capacity to process mathematical concepts. Formalisation is a crucial technique that enables AI to break down mathematical ideas into computer code, allowing for thorough verification. This method ensures the accuracy of results and enables AI to confidently arrive at solutions.

Theoretical physicists often engage in creative problem-solving when attempting to understand phenomena that defy explanation. They craft models or simulations to capture these events, which requires a deep understanding of both mathematics and the subject matter. This process is as much an art as it is a science.

While formalisation has been instrumental in AI's success in mathematics, its application to more complex areas of physics remains uncertain. The field of theoretical physics involves not only mathematical precision but also creative thinking and problem-solving skills that may be difficult for AI systems to replicate.

Advances in using formalisation to test physics have been made, but they are still in their infancy compared to mathematics.

Physics problems often involve complex variables that make them harder to pin down and formalise than mathematical equations. This complexity can be a significant challenge for AI systems trying to replicate the work of physicists.

Physicists are already leveraging AI to automate tasks such as writing computer code, performing mathematical calculations, designing experiments, and even running experiments themselves.

In some laboratories, researchers have begun creating controlled environments where inputs and outputs can be precisely measured and influenced through computer control.

Researchers have made significant progress in using AI to predict the outcomes of complex physics experiments. According to researchers, an AI system called Claude can now provide precise predictions on what output to expect from a laser pulse experiment, essentially allowing it to say "I think this laser pulse is gonna produce this output".

The potential benefits of this technology are substantial, as these experiments can run overnight and complete as much work as a human would in a week. Moreover, human physicists can even use an app on their phone to suggest ideas that come to them while they are away from the lab, and receive results when they return.

However, experts caution that AI will not necessarily revolutionize our understanding of the universe by exploring all its mysteries. Instead, it is likely to focus on very constrained domains, where inputs and outputs can be precisely measured and influenced through computer control.

---
Source: [New Scientist](https://www.newscientist.com/article/2593273-will-physics-be-the-next-field-to-get-turned-upside-down-by-ai/?utm_campaign=RSS|NSNS&utm_content=home&utm_medium=RSS&utm_source=NSNS)  
Published by News Central Site: https://newscentral.site/artificial-intelligence-outperforms-human-experts-in-solving-complex/