Artificial intelligence, having already disrupted the field of mathematics by solving longstanding challenges, is now poised to revolutionize other theoretical sciences such as economics, high-energy physics, and quantum physics, according to new MIT research. The study tracks a remarkable leap in AI’s problem-solving capabilities and anticipates significant shifts in scientific inquiry driven by AI models.
- AI models solved up to 35% of new theoretical problems between 2024-2026.
- Study covers economics, physics, cosmology, and related theory-heavy fields.
- Researchers caution human experts remain essential to validate AI solutions.
What happened
MIT researchers led by Neil Thompson conducted a study testing advanced AI models on unresolved theoretical problems drawn from recent research papers between December 2025 and May 2026. The models, including iterations of OpenAI’s GPT-5, were tasked with independently generating solutions without internet access or exposure to the solutions during training.
The findings showed an extraordinary leap in AI’s problem-solving capacity, increasing from about 2% success in mid-2024 to 35% by mid-2026 across fields such as economic theory, quantum physics, and cosmology. This progress suggests AI is beginning to disrupt traditional scientific inquiry beyond just mathematics.
Why it matters
These advancements imply AI could become a primary driver of theoretical breakthroughs in multiple scientific disciplines. As AI autonomously tackles complex problems, it may accelerate the pace of fundamental discoveries and influence how theoretical insights translate into practical innovations.
However, this surge in AI capability also raises important concerns, including the need for expert oversight to verify solutions due to potential errors and oversights in AI-generated answers. Additionally, the traditional scientific process, which values the journey to discovery, might be reshaped by AI’s ability to produce final answers quickly.
What to watch next
The scientific community should monitor how AI-integrated research evolves, particularly efforts to combine AI-driven theoretical analysis with real-world experimentation. New labs and initiatives focused on AI for science are emerging, signaling growing investment in this approach across major AI and scientific institutions.
Equally important will be frameworks for validation and interpretation to ensure AI outputs are both correct and meaningful. As AI continues to advance, collaborative models between human experts and AI will likely become crucial to harness its full potential while mitigating risks of reliance on unverified solutions.