Artificial intelligence has solved math problems that went unanswered for decades, according to a report from Northeastern Global News published on October 22, 2026.
Just a few years ago, AI tools often got basic math wrong. Studies showed they gave the wrong answer more than half the time when adding two four-digit numbers.
That changed quickly. In 2024, Google DeepMind’s AI systems scored high enough for a silver medal at an elite high school math competition. In 2025, OpenAI earned a gold.
AI Tackles Problems Older Than Most Mathematicians
In May 2026, AI made progress on a question posed by mathematician Paul Erdős in 1946. The problem asks how many pairs of dots on a flat surface can sit exactly one unit apart.
Experts had a long-held prediction about the answer. The AI overturned it by applying number theory techniques to geometry, finding a link that specialists had missed.
In September, OpenAI announced that its AI agents had found a solution to the Navier-Stokes problem, which describes how fluids move. The answer could help scientists understand when computer models of air or blood flow stop being reliable.
The work reportedly took 88 hours and used around 10,000 AI agents searching through decades of research. People had worked on the question for roughly 90 years.
Ken Duffy, chair of the mathematics department at Northeastern University, said well-defined problems like these make useful test cases for AI.
Experts Say AI Still Struggles to Show Its Work
Duffy said research math is about more than finding answers. “The currency of research is (finding) a good question to ask,” he said.
Mathematicians are recognized for explaining how they reached a solution. A good proof shows the reasoning and gives others a path to learn from.
AI often falls short here. Its proofs can be hard for mathematicians to follow, and some AI results have been retracted after they did not hold up in Lean, a proof-checking system.
“When we mark students in exams, we don’t give them full points for getting the correct answer if they don’t explain how they got there,” Duffy said.
AI has proven useful in other fields. In biology, it predicted the shapes of more than 200 million proteins, helping scientists learn what they do.
Duffy’s team used AI to map cells in bone marrow to study changes caused by leukemia. The task once required graduate students to examine images of up to 100 million cells by hand.
Duffy said the AI picked out individual cells accurately with a fraction of the human effort.
James Boyd, co-founder of the Wolfram Institute, said AI is challenging assumptions about math as a whole. “It would be healthy for the field to broaden beyond proving theorems alone,” he said.
Boyd pointed to experimental mathematics, which draws conclusions from patterns. In this approach, computers handle the number crunching and AI tests whether the patterns hold up.
He calls this “high-performance human mathematics.” Boyd said keeping humans in charge of each step makes it unlikely they will be pushed aside in favor of the results.
