AI Agents Now Reconstruct Complex Scientific Papers Without Access to Code or the Papers Themselves
Ethan Mollick demonstrates AI agents independently reproducing results from complex research papers given only methods descriptions and data — a capability that could transform scientific reproducibility or upend academic publishing norms.
Wharton professor Ethan Mollick, one of the most closely watched voices on AI's intersection with knowledge work, demonstrated this weekend that current AI agents can independently reconstruct the results of complex scientific papers when given only the methods section and underlying data — no access to the original code, figures, or conclusions. As @emollick noted, the agents are "now good enough" to perform this task reliably, a threshold that has significant implications for how science is conducted and verified.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.