Do Large Language Models know what humans know?

TL;DR

This study investigates whether large language models like GPT-3 can understand others' mental states, finding they show some sensitivity but do not fully replicate human belief attribution, implying additional mechanisms are involved.

Contribution

The paper provides empirical evidence that language exposure alone partially explains theory of mind development, highlighting the need for other mechanisms in humans.

Findings

GPT-3 exceeds chance in false belief tasks

Humans outperform GPT-3 in belief attribution

Language exposure alone does not fully account for human theory of mind

Abstract

Humans can attribute beliefs to others. However, it is unknown to what extent this ability results from an innate biological endowment or from experience accrued through child development, particularly exposure to language describing others' mental states. We test the viability of the language exposure hypothesis by assessing whether models exposed to large quantities of human language display sensitivity to the implied knowledge states of characters in written passages. In pre-registered analyses, we present a linguistic version of the False Belief Task to both human participants and a Large Language Model, GPT-3. Both are sensitive to others' beliefs, but while the language model significantly exceeds chance behavior, it does not perform as well as the humans, nor does it explain the full extent of their behavior -- despite being exposed to more language than a human would in a lifetime. This suggests that while statistical learning from language exposure may in part explain how humans develop the ability to reason about the mental states of others, other mechanisms are also responsible.