Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks

By Tomer D. Ullman. 2023

388 citations recorded by the source corpus.

Intuitive psychology is a pillar of common-sense reasoning. The replication of this reasoning in machine intelligence is an important stepping-stone on the way to human-like artificial intelligence. Several recent tasks and benchmarks for examining this reasoning in Large-Large Models have focused in particular on belief attribution in Theory-of-Mind tasks. These tasks have shown both successes and failures. We consider in particular a recent purported success case, and show that small variations that maintain the principles of ToM turn the results on their head. We argue that in general, the zero-hypothesis for model evaluation in intuitive psychology should be skeptical, and that outlying failure cases should outweigh average success rates. We also consider what possible future successes on Theory-of-Mind tasks by more powerful LLMs would mean for ToM tasks with people.

Subfields: Artificial Intelligence, Large Language Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Large Language Models

In Safety, Robustness & Interpretability

Citations in the corpus

Testing theory of mind in large language models and humans (2024, methodology): “However, the mixed success of these models 23 , along with their vulnerability to small perturbations to the provided prompts, including simple changes in characters’ perceptual access 25 , raises concerns about the robustness and interpretability of the observed successes.”
Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies (2022, background): “Concurrent work by Ullman (2023) shows that although GPT-3 previously showed success on Theory of Mind psychology tasks (Kosinski, 2023), it fails on prompts with several types of directed variations, illustrating the necessity of evaluating the robustness of observed eﬀects using alternative…”
Playing repeated games with large language models (2023, neutral): “Experiments were performed in accordance with the relevant guidelines and regulations approved by the ethics committee of the University of Tübingen (protocol nr. 701/2020BO).”