Generative Responses and Language Mastery

To what extent is artificial intelligence today actually intelligent? Can we say that so-called large language models (LLMs) like ChatGPT possess genuine expertise or wisdom? This fall, join us as we discuss the distinction between genuine and apparent wisdom and expertise in the context of ancient debates about the same topic, and see how these ancient debates could apply to our current questions about artificial intelligences. This four-week reading group will synthesize classical theories of human understanding and communication—namely, those of Plato in Ion and Gorgias—with present-day research on LLMs and other "intelligent" programs.

We will be discussing the capacities and limitations of Large Language Models, particularly their role in communication and persuasion. We will then look at a specific case of generating persuasive medical advice, which will be considered from the Platonic perspective in the next session.

Readings:

“Talking about Large Language Models” by Murray Shanahan

“Working with AI to Persuade: Examining a Large Language Model’s Ability to Generate Pro-Vaccination Messages” by Karinshak et al. (excerpts)