Bayesian Expected Uncertainty Reduction (B-EUR) Model: A Computational Account of What Makes Design Options Worth Trying

Understanding Design Exploration

Design processes are inherently uncertain, requiring designers to perform trials to learn how specific actions lead to desired outcomes. While prior research has identified uncertainty as a driver of exploration, it has remained unclear exactly how environmental properties—such as the predictability of design outcomes or the scope of knowledge gained from a single prototype—influence the perceived value of these trials. This paper introduces the Bayesian Expected Uncertainty Reduction (B-EUR) model to provide a computational account of how designers evaluate candidate actions based on their potential to reduce epistemic uncertainty.

The B-EUR Model

The authors formalize the value of a design action as the expected reduction in entropy regarding action-outcome relations. They define two key environmental properties:

• Generalizability (l): The extent to which knowledge from one trial applies to neighboring design options.
• Outcome Discriminability (σ): The clarity with which differences between design outcomes can be distinguished.

Using a graph-shape guessing task, the researchers simulated how these properties affect the 'epistemic value' of a trial. The model predicts that when generalizability is too low, trials are isolated and provide little broader insight; when it is too high, trials provide redundant information. Consequently, epistemic value follows an inverted-U-shaped curve relative to generalizability. Conversely, higher outcome discriminability consistently increases the information gained from each trial, making them more valuable.

Experimental Validation

To test the model, the authors conducted human experiments using a graph-guessing game that isolates learning from pragmatic reward-seeking. Participants' subjective ratings of the 'value of trying' and their enjoyment of the exploration process mirrored the model’s predictions, showing an inverted-U-shaped relationship with generalizability. These findings suggest that designers are intuitively sensitive to the information-theoretic value of their actions, and that the B-EUR model can help practitioners frame design problems and organize prototype sets to maximize learning efficiency.