Prompts shape how AI models follow default choices in group games
Prompt framing governs LLM default following in collective-action
Computer Science and Game Theory
Summary
When AI language models are given a pre-filled choice, like a suggested answer in a game, how much they follow that choice depends on how it's worded and what options they have. The authors tested this with games where people share resources or work together. They found the AI is more likely to pick the default if it's presented as a suggestion rather than just information. This means the way choices are shown affects AI decisions, which is important when these models help make or recommend decisions.
What this means in practice
- •For ai interface designers: Design prompts and interfaces that consider wording effects to better control AI model choice biases in decision support tools.
- •For game developers: Use findings on default influence to fine-tune AI agents’ behavior in social and economic simulation games.
Authors
Eladio Montero-Porras, Axel Abels, Tom Lenaerts
Abstract
Large language models are increasingly deployed as agents that make or recommend decisions on behalf of users, often operating through interfaces that pre-fill suggested values or default options. Whether models treat such defaults as merely informational or as suggestions that systematically alter their choices remains unclear. We study default deference in two one-shot social dilemmas: a common-pool resource (CPR) extraction game and a threshold public-good (TPG) contribution game. We measure how defaults shift each model's choice distribution relative to its no-default baseline, across default values, wordings, and action-space granularities. We find that pre-filled defaults pull probability mass on the default value in both games, but the magnitude depends strongly on wording: the same model can show high pull under one formulation and near-zero pull under another. Permission-style wording reduces default pull in both games, more strongly in CPR than in TPG. Default pull is weaker in coarse action spaces, and conflict defaults attract more mass than agreement ones. These results indicate that default deference depends on the model, the wording of the interface, and the structure of available choices. For agentic systems, evaluating model behaviour without controlling the surrounding choice architecture can miss an important source of behavioural variation.