Imagine a world where artificial intelligence isn’t just a buzzword but a genuine partner in groundbreaking research. But here’s where it gets controversial: what happens when AI doesn’t just assist but becomes a critical step in solving complex problems? This is the reality Professor Scott Aaronson, the Schlumberger Centennial Chair of Computer Science and director of the Quantum Information Center at UT, recently experienced. On October 6, 2025, Aaronson shared in his office how he leveraged AI to tackle a key technical challenge in his latest research paper, co-authored with Freek Witteveen of Centrum Wiskunde & Informatica in the Netherlands. The paper, posted on arXiv (a non-peer-reviewed platform), builds on Aaronson’s 2008 work in quantum physics, and its recent circulation among students, researchers, and colleagues was fueled by one intriguing word: ‘AI.’
In a September 27 blog post on Shtetl-Optimized, his long-running platform for sharing research updates, Aaronson revealed he used OpenAI’s latest model, GPT-5, to assist with a complex mathematical step in his proof. ‘It was the first time in my career I’ve seen AI genuinely help in the research process,’ Aaronson noted. While GPT-5 initially provided an incorrect answer, Aaronson’s iterative dialogue with the model eventually led to a breakthrough. ‘The key,’ he emphasized, ‘is knowing enough about the subject to spot when AI goes wrong. That’s non-negotiable.’
And this is the part most people miss: AI’s role in research isn’t about replacing human expertise but augmenting it. Phillip Harris, a postdoctoral researcher at the University of Bonn, chimed in on Aaronson’s blog, offering an even better solution than GPT-5’s. ‘There’s this urge to nitpick AI’s output,’ Harris observed. ‘Everyone’s a bit on edge, looking for flaws.’ Yet, Harris argues that AI has advanced to the point where mathematicians should embrace it—at least to some degree. Focusing on pure math, which explores abstract concepts rather than practical applications, Harris highlights AI’s surprising prowess in this domain. ‘Pure math is so detached from real life,’ he explained, ‘and that’s exactly why AI excels at it.’
Aaronson’s experience isn’t isolated. Last year, he experimented with AI for similar problems but found the results less accurate. This time, GPT-5 not only provided a viable solution but saved significant time—time he and his colleague would have otherwise spent solving the problem manually. ‘AI is undeniably useful,’ remarked Andrew McAlinden, a senior in computer science and mathematics and a student in Aaronson’s Introduction to Quantum Information Science class. ‘If I’m stuck, my options are office hours, a friend, or AI. Either way, I’m consulting someone—AI just speeds it up.’
This shift isn’t lost on UT’s computer science department. Aaronson shared that a committee will convene this semester to discuss how AI should reshape departmental operations. His vision? Offer courses where students are encouraged to use AI alongside traditional learning methods. ‘We need to guide this toward better outcomes,’ Aaronson urged. ‘Even if AI can perform foundational tasks, we must still teach those skills.’
Here’s the thought-provoking question: As AI becomes more integrated into research and education, are we risking over-reliance, or is this the natural evolution of how we solve problems? Should students be required to use AI, or does that undermine the value of independent learning? Share your thoughts in the comments—let’s spark a debate!
