AI Could Speed Scientific Discovery, Says Anthropic Leader

Scientists might soon have a powerful new research partner that never gets tired of reading papers or analyzing data.

Jack Clark, co-founder of AI company Anthropic, told economic policymakers in Bangkok that artificial intelligence could transform how quickly researchers make discoveries. Instead of just helping people write emails or summarize meetings, AI could help scientists spot patterns worth investigating and explore ideas that might otherwise go unnoticed.

The difference matters because scientific progress shapes everything from the medicines we have to the problems we can solve. Faster discovery could mean faster answers to questions that affect millions of lives.

Anthropic already has a concrete example. In September, the company's AI system Claude examined DNA data and flagged an unusual enzyme system that resembled CRISPR arrangements. Scientists had never characterized this system before, and while nobody knows yet what it does, the discovery gives researchers something new to investigate.

That's exactly the kind of help AI could provide. Research requires endless supporting work like reading papers, cleaning datasets, and comparing methods. An AI assistant could explore multiple approaches while scientists focus on the most promising leads, making their limited time go further.

The Ripple Effect

The potential reaches beyond individual labs. Stanford's latest AI Index reports that researchers published over 80,000 AI-related science papers in 2025, a 26% jump from the previous year. That explosion of interest suggests many scientists see real possibilities.

But AI isn't ready to run experiments alone. The same Stanford report shows these systems excel at answering chemistry questions while struggling with complete scientific tasks. They're helpful teammates, not replacements for human judgment.

The verification process remains crucial. Before any AI-generated insight becomes accepted science, researchers must trace it back to the original data, check for errors, and ensure others can reproduce the results. A convincing explanation means nothing if it falls apart under scrutiny.

Physical limitations matter too. Even brilliant digital assistance can't make lab equipment work faster or manufacture materials that don't exist yet. Real-world testing still takes real-world time.

Still, helping scientists explore more ideas efficiently could accelerate the timeline from question to answer, and in fields like medicine or climate science, faster answers could change lives.