AI's Scientific Breakthroughs Under Scrutiny Amid Concerns of Undermining Research

Artificial intelligence is accelerating scientific discovery, but a recent dispute involving a mathematical breakthrough has ignited concerns that AI could ultimately undermine the scientific process itself.

The controversy emerged last month when OpenAI, the creator of ChatGPT, announced that its AI agents had solved one of mathematics' long-standing unsolved problems: the Navier–Stokes equations. The company stated that its AI completed the task in 88 hours. The organization responsible for a prize associated with the problem indicated that it had "apparently been settled," but a formal review would require time.

However, hours before OpenAI's announcement, Tristan Buckmaster, a mathematician at New York University, publicly voiced his concerns. Buckmaster, along with Levent Alpöge, a mathematician employed by rival company Anthropic, had been working independently towards the same goal. They had stored their research drafts in OpenAI's coding tool, Codex. Buckmaster noted that his co-researcher was alerted that news of their progress had "reached" OpenAI, though he did not specify the source of this information. He questioned whether his own research, stored within Codex, might have inadvertently contributed to OpenAI's work, though he clarified that he was "not accusing anyone of anything."

Beyond specific disputes, AI poses a broader threat to scientific journals, which are responsible for upholding scientific integrity. Critics argue that these publications are struggling to manage a surge of AI-generated distortions. Dalmeet Singh Chawla, a science journalist with a decade of experience reporting on research misconduct, stated that generative AI has "basically disrupted the whole scientific process."

Chawla told the BBC that publishers' integrity staff would need to "drastically increase [in number and effort] if they're anywhere near the fight," adding, "Right now, it seems like they're not."

Despite these concerns, AI's potential to power scientific research remains significant, given the vast amounts of data generated by scientific instruments. Telescopes and particle accelerators, for instance, collect billions of data points that are beyond human comprehension. AI is adept at identifying crucial scientific insights within such large datasets.

A prominent example of AI in scientific advancement is Google DeepMind's AlphaFold. In 2020, this AI model accurately predicted the structures of over 200 million proteins in just one year. Proteins are fundamental to life, acting as the molecular machines that carry out most functions within cells. Their folded shapes dictate their capabilities, from forming tissues and organs to fighting infections.

Previously, determining protein structures required extensive and time-consuming laboratory experiments. AlphaFold's ability to predict these structures so rapidly represented a major leap forward. The computational work that would have taken hundreds of millions of years in laboratories was accomplished in a single year by AlphaFold. Google DeepMind made these predictions freely available to researchers. The head of DeepMind and the lead scientist on the AlphaFold project were recognized with the Nobel Prize in Chemistry for their work.

Dr. Pushmeet Kohli, who leads DeepMind's science initiatives, described AlphaFold as a "watershed moment" in scientific AI. Meanwhile, firms like Basecamp Research are collecting vast amounts of genetic data, aiming to build AI systems capable of understanding biological processes, drawing parallels to how large tech companies have used online text to develop Large Language Models.