Samuel Schindler - On the Limits of AI-driven Discovery in Science

Samuel Schindler (Aarhus University)

The Center for Philosophy of Science invites you to our Featured Former Fellow online lecture presented by:   

Samuel K. Schindler – Philosopher of Science, Aarhus University

Friday, November 21st  @ 12:00 pm - 1:30 pm EST

Title: On the Limits of AI-driven Discovery in Science

Abstract: 

There has been much enthusiasm among the public general about recent developments in AI. The enthusiasm is echoed among scientists, culminating in the Nobel Prize for AlphaFold. In this paper, I want to dampen this enthusiasm slightly and articulate limits on attempts to make AI-driven scientific discoveries – understood as discoveries of new phenomena rather than ideas. To get a sense of the limitations of AI-driven discovery, I compare machine learning to standard scientific instrumentation and argue that the much-discussed opacity of machine learning restricts our ability to distinguish signal from noise. This constitutes a problem for both “target-specific” and “target-unspecific” learning regimes, but especially for the latter – arguably the more promising regime for the discovery of yet unknown phenomena – and gives rise to what I call the Aggravated Artifact Problem.

This talk will be available online only:

Zoom: https://pitt.zoom.us/j/97096992508

 YouTube at https://www.youtube.com/channel/UCrRp47ZMXD7NXO3a9Gyh2sg.