AI’s Next Frontier: From Math to the Cosmos
The scientific community is bracing for a seismic shift. Just as artificial intelligence continues to redefine the landscape of mathematics, a groundbreaking study from MIT suggests that its transformative power is poised to ripple through an array of other theoretical sciences. The message is clear: AI isn’t just coming for math; it’s already knocking on the doors of probability, quantum physics, and even cosmology.
Last week, at a compelling talk at MIT, Neil Thompson, a leading researcher in computing and AI progress, laid out a stark vision for the future. In a mere thirty minutes, Thompson presented a compelling case, underscored by his latest research, that several scientific disciplines are on the cusp of profound disruption by advanced AI models.
The Unstoppable Ascent of AI in Theoretical Fields
A Leap in Problem-Solving Capability
While mathematicians grapple with the automation sweeping their field, Thompson’s study, conducted with Gabriel Manso and Claudio Mayrink Verdun, points to the next wave of impacted areas: probability, statistics theory, economic theory, high-energy theory, gravitation and cosmology, and quantum physics. The rapid evolution of frontier AI models is the driving force behind this impending transformation.
The data is striking: between mid-2024 and mid-2026, the most sophisticated AI models witnessed an astonishing leap in their ability to correctly solve theoretical problems within these fields. Their success rate surged from a mere 2 percent to an impressive 35 percent. Crucially, this trajectory shows no signs of deceleration.
“Within just a few years, we expect a lot of theory questions to be totally transformed in terms of AI’s ability to do them,” Thompson predicts, signaling a future where AI plays an indispensable role in scientific inquiry.
Rigorous Testing, Unbiased Results
Thompson’s lab, renowned for investigating the societal and innovative impacts of computer science advancements, employed a meticulous methodology. The team sourced problems from research papers published between December 2025 and May 2026. These problems were then presented to various OpenAI models—o1, o3, GPT-5, GPT-5.1, GPT-5.4, and GPT-5.5—to assess their problem-solving prowess.
A critical aspect of the study was ensuring the models’ independent reasoning. Solutions to these problems were deliberately chosen from papers published after the training cutoff for GPT-5.5 and its predecessors. This meant the AI models had to devise solutions autonomously, rather than recalling pre-fed training data. Furthermore, internet access was blocked, preventing any online lookup of answers, solidifying the integrity of the results.
While the study’s scope was focused on fields where AI could operate without needing physical simulations or experiments, its implications are vast. The ‘shock waves’ currently reverberating through mathematics are a harbinger of what’s to come for other theory-heavy scientific domains.
Lessons from the Math Revolution: Navier-Stokes and Beyond
The mathematical community recently experienced its own ‘AI moment’ when OpenAI announced that an unreleased model had cracked the long-standing Navier-Stokes equations, fundamental to understanding fluid motion. This revelation sparked considerable debate, particularly after a mathematician from NYU accused OpenAI of ‘gazumping’ their own parallel progress.
Since then, OpenAI and other entities have unveiled hundreds more solutions to complex mathematical problems, prompting a profound introspection within the field. Some mathematicians advocate for limits on AI usage, emphasizing that the journey of problem-solving often holds as much value as the final answer itself.
The Future of Science: Collaboration, Not Replacement
Thompson posits that, much like in mathematics, AI’s advancements in other scientific areas will bring both opportunities and challenges. He highlights that theoretical breakthroughs frequently catalyze practical scientific progress, suggesting that increased AI integration could accelerate this cycle.
Already, prominent AI researchers are establishing new labs dedicated to ‘AI for science,’ and major AI companies are significantly scaling up their scientific research divisions. These initiatives aim to empower AI not just for theoretical analysis, but also for conducting real-world experiments.
While AI may alleviate the burden on human theorists to solve every key problem, experts remain crucial for interpreting solutions and ensuring their validity. “We are still going to need people to generate good ideas to connect the dots,” emphasizes Gabriel Manso, a co-author of the study. He cautions that AI, particularly with highly complex problems, is not infallible. Indeed, OpenAI itself retracted several solutions post-publication from its latest dataset.
“We also need ways to verify what was done, because it’s very easy for a model to tell you ‘this is the right answer,’ and when you look at the entire solution, you start to see maybe it is not,” Manso concludes, underscoring the enduring need for human expertise and critical oversight in the age of AI-driven discovery.
For more details, visit our website.
Source: Link
Leave a comment