Rigidity in LLM Bandits with Implications for Human-AI Dyads
This research identifies robust decision biases in LLMs, which is significant for practitioners and researchers designing human-AI interaction systems, as these biases could shape how humans interact with AI.
This paper investigates decision biases in LLMs using a two-arm bandit setup, running 20,000 trials per condition. They found that LLMs amplify positional order into stubborn one-arm policies under symmetric rewards and exploit rigidly but underperform an oracle under asymmetric rewards. These behaviors are robust to common decoding configurations like temperature and top-p.
We test whether LLMs show robust decision biases. Treating models as participants in two-arm bandits, we ran 20000 trials per condition across four decoding configurations. Under symmetric rewards, models amplified positional order into stubborn one-arm policies. Under asymmetric rewards, they exploited rigidly yet underperformed an oracle and rarely re-checked. The observed patterns were consistent across manipulations of temperature and top-p, with top-k held at the provider default, indicating that the qualitative behaviours are robust to the two decoding knobs typically available to practitioners. Crucially, moving beyond descriptive metrics to computational modelling, a hierarchical Rescorla-Wagner-softmax fit revealed the underlying strategies: low learning rates and very high inverse temperatures, which together explain both noise-to-bias amplification and rigid exploitation. These results position minimal bandits as a tractable probe of LLM decision tendencies and motivate hypotheses about how such biases could shape human-AI interaction.