On the Limits of AI-driven Discovery in Science

Value of the Information & Strength of the Argument

The talk provides a valuable philosophical analysis of the epistemic limits of AI in scientific discovery. Schindler’s argument is well-structured and clear, using a dilemma to frame the discussion. He effectively uses concrete examples like AlphaFold and Halicin to illustrate his points, and he acknowledges counterarguments, which strengthens his position. The argument is solid, though it relies on a particular definition of ‘discovery’ as the discovery of new phenomena, which may be contested. The thought experiment with binoculars is particularly effective in conveying the idea of epistemic narrowness.

Scientific Rigor, Source Quality, Title Accuracy

Schindler demonstrates scientific rigor by referencing key literature in philosophy of science, such as Kuhn’s work on scientific revolutions, and by discussing recent AI developments like AlphaFold and Halicin. He also cites philosophers working on opacity, such as Katherine Creel. The sources are appropriate for the argument, though they are not exhaustively listed. The title accurately reflects the content, and the talk is well-organized. The speaker’s expertise is evident, and he presents a balanced view, acknowledging the successes of AI while highlighting its limitations.

The title accurately reflects the content, which focuses on the limitations of AI in scientific discovery, particularly the discovery of new phenomena.

8/10

The talk is a well-structured philosophical argument by an expert in philosophy of science, drawing on established literature and concrete examples. The argument is nuanced and acknowledges counterexamples, but it is primarily an opinion piece rather than an empirical study.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction and overview of the talk's agenda.
• 10:00Explanation of deep neural networks and supervised learning.
• 20:00Introduction of the dilemma for DNN discovery.
• 30:00Discussion of the discovery of new phenomena and the concept of epistemic openness.
• 40:00Analysis of AlphaFold and the thought experiment about binoculars.
• 50:00Discussion of Halicin as a potential counterexample.
• 1:00:00Introduction of the aggravated artifact problem and the opacity of DNNs.
• 1:10:00Conclusion and summary of the argument.

• Samuel Schindler's personal website — Speaker's academic profile and publications.

• Kuhn, T. (1962). The Structure of Scientific Revolutions — Referenced for the discussion of scientific discovery and paradigm shifts.

• Duede, E. (2023). AI and the context of discovery — Argues that AI can be useful for discovery of ideas, which contrasts with Schindler's focus on phenomena.