Intentionality All-Stars Redux: Do language models know what they are talking about?

The rise of large language models (LLMs) has brought with it a robust debate about whether these machines know what they are talking about, or whether they are just bullshitting. Answering this question requires wrestling with the problem of intentionality: what are the conditions for our thoughts, beliefs, and judgments to be about the world? How can we differentiate a model that says the right thing from one that understands what it is saying? This article addresses the intentionality of LLMs by revisiting Haugeland’s storied team, “the Intentionality All-Stars.” I lay out four major approaches to intentionality that are regularly applied to explain LLMs behavior—as well as their failures. Spelling out the different positions, in turn, shows that there are multiple conditions that must be met for an LLM to know what it is talking about—and the numerous ways current models fall short.