Neural Theory-of-Mind? On the Limits of Social Intelligence in Large LMs

By Maarten Sap, Ronan Le Bras, Daniel Fried, Yejin Choi. 2022

333 citations recorded by the source corpus.

Social intelligence and Theory of Mind (TOM), i.e., the ability to reason about the different mental states, intents, and reactions of all people involved, allows humans to effectively navigate and understand everyday social interactions. As NLP systems are used in increasingly complex social situations, their ability to grasp social dynamics becomes crucial.In this work, we examine the open question of social intelligence and Theory of Mind in modern NLP systems from an empirical and theorybased perspective. We show that one of today’s largest language models (GPT-3; Brown et al., 2020) lacks this kind of social intelligence out-of-the box, using two tasks: SocialIQa (Sap et al., 2019), which measure models’ ability to understand intents and reactions of participants of social interactions, and ToMi (Le, Boureau, and Nickel, 2019), which measures whether models can infer mental states and realities of participants of situations.Our results show that models struggle substantially at these Theory of Mind tasks, with well-below-human accuracies of 55% and 60% on SocialIQa and ToMi, respectively. To conclude, we draw on theories from pragmatics to contextualize this shortcoming of large language models, by examining the limitations stemming from their data, neural architecture, and training paradigms. Challenging the prevalent narrative that only scale is needed, we posit that person-centric NLP approaches might be more effective towards neural Theory of Mind.

Subfields: Artificial Intelligence, Social Psychology, Large Language Models, Cognitive Science & Psychology

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 Social Psychology

In Large Language Models

In Cognitive Science & Psychology

Citations in the corpus

QLoRA: Efficient Finetuning of Quantized LLMs (2023, background): “Theory of Mind Guanaco displays surprisingly strong Theory of Mind capabilities [41, 52].”
Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks (2023, background): “14)) suggests the premise of the dilemma is unfounded – current LLMs do not pass ToM tests.”
Evaluating large language models in theory of mind tasks (2023, background): “58 and 59 ), others found that LLMs’ performance was inconsistent and brittle ( 26 , 60 , 61 ).”
Understanding Natural Language Understanding Systems. A Critical Analysis (2023, background): “Sap, LeBras, Fried and Choi (2022) also report that the ability of neural language models in tasks designed to test their social intelligence is very far from the human one.”
Faith and Fate: Limits of Transformers on Compositionality (2023, background): “However, these models have also been shown to struggle across multiple domains, including algorithmic reasoning [59], commonsense reasoning [44, 34], theory of mind [47], planning [58], logical reasoning [52], and ethical reasoning [30].”
Mirages: On Anthropomorphism in Dialogue Systems (2023, background): “Recent work (e.g., Sap et al., 2022) has investigated models for theory of mind , i.e., an entity’s ability to infer other people’s ‘mental states, such as beliefs, desires, and intentions, and to understand how mental states feature in everyday explanations and predictions of people’s behaviour’…”
Reasoning with Language Model Prompting: A Survey (2022, background): “Besides, (Sap et al., 2022) show that one of today’s largest language models (GPT-3 (Brown et al.”
Testing theory of mind in large language models and humans (2024, methodology): “One example is the false belief test, one of the most widely used tools so far for testing the performance of LLMs on social cognitive tasks 19,21–23, 25, 51 , 52 .”