Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations
The paper explores the impact of improving Theory of Mind (ToM) capabilities in Large Language Models (LLMs) on human-AI interactions. It highlights the limitations of existing benchmarks that do not adequately reflect the dynamic nature of these interactions. The authors propose a new evaluation paradigm and present findings that suggest static benchmark improvements do not necessarily enhance real-world HAI performance.
- ▪The study introduces a new paradigm for evaluating Theory of Mind in AI interactions.
- ▪Existing benchmarks often overlook the dynamic and open-ended aspects of human-AI interactions.
- ▪Improvements in static benchmarks do not always correlate with better performance in real-world applications.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.15205 (cs) [Submitted on 28 Apr 2026] Title:Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations Authors:Nanxu Gong, Zixin Chen, Haotian Li, Zishu Zhao, Jianxun Lian, Huamin Qu, Yanjie Fu, Xing Xie View a PDF of the paper titled Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations, by Nanxu Gong and 7 other authors View PDF HTML (experimental) Abstract:Improving the Theory of Mind (ToM) capability of Large Language Models (LLMs) is crucial for effective social interactions between these AI models and humans.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
