MIT/Harvard model treats design as a cooperative signaling game

The paper, accepted at CogSci 2026, frames design as a cooperative game where the designer (assistant) makes choices that communicate an artifact’s underlying affordances and causal structure, while the user (principal) infers the true model by recursively reasoning about the designer’s intent. This approach extends Rational Speech Act (RSA) theory—traditionally used for language—to the domain of physical design. The model balances informativeness against efficiency: designers place visually identical keys on trays to help users deduce which key unlocks which door, without relying on explicit labels or text.

In experiments with grid-world layouts, the researchers compared their pragmatic model against literal (non-mentalizing) baselines. The pragmatic model—where both designer and user recursively simulate each other’s reasoning—better matched actual human judgments about which keys go with which doors. The findings suggest that even minimal, unlabeled design choices can effectively communicate complex functional relationships, offering a computational foundation for why people can understand novel artifacts after just a few interactions. This has implications for UX design, explainable AI, and human-robot interaction.