Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Act or Clarify? Modeling Sensitivity to Uncertainty and Cost in Communication

Polina Tsvilodub, Karl Mulligan, Todd Snider, Robert D. Hawkins, Michael Franke

TL;DR

The paper investigates how humans decide between acting under uncertainty and seeking information through clarification questions (CQs) in social communication. It develops a layered, expected-regret-based model predicting when to ask CQs versus act, and tests it with two experiments: Experiment 1 examines linguistic responses to questions under varying uncertainty and option costs, while Experiment 2 extends to directives and non-linguistic actions to probe binarized versus gradient uncertainty effects. Results show people are more likely to ask CQs when uncertainty is present and the cost of wrong actions is high, with a distinct two-stage pattern where CQ use is influenced by uncertainty and action costs, and direct actions vary gradually with uncertainty. A Bayesian model-fitting approach supports the expected-regret account as a plausible mechanism for the observed tradeoffs, offering a quantitative framework for rational communication under risk and informing the design of more human-like AI systems capable of cost-aware clarification.

Abstract

When deciding how to act under uncertainty, agents may choose to act to reduce uncertainty or they may act despite that uncertainty.In communicative settings, an important way of reducing uncertainty is by asking clarification questions (CQs). We predict that the decision to ask a CQ depends on both contextual uncertainty and the cost of alternative actions, and that these factors interact: uncertainty should matter most when acting incorrectly is costly. We formalize this interaction in a computational model based on expected regret: how much an agent stands to lose by acting now rather than with full information. We test these predictions in two experiments, one examining purely linguistic responses to questions and another extending to choices between clarification and non-linguistic action. Taken together, our results suggest a rational tradeoff: humans tend to seek clarification proportional to the risk of substantial loss when acting under uncertainty.

Act or Clarify? Modeling Sensitivity to Uncertainty and Cost in Communication

TL;DR

The paper investigates how humans decide between acting under uncertainty and seeking information through clarification questions (CQs) in social communication. It develops a layered, expected-regret-based model predicting when to ask CQs versus act, and tests it with two experiments: Experiment 1 examines linguistic responses to questions under varying uncertainty and option costs, while Experiment 2 extends to directives and non-linguistic actions to probe binarized versus gradient uncertainty effects. Results show people are more likely to ask CQs when uncertainty is present and the cost of wrong actions is high, with a distinct two-stage pattern where CQ use is influenced by uncertainty and action costs, and direct actions vary gradually with uncertainty. A Bayesian model-fitting approach supports the expected-regret account as a plausible mechanism for the observed tradeoffs, offering a quantitative framework for rational communication under risk and informing the design of more human-like AI systems capable of cost-aware clarification.

Abstract

When deciding how to act under uncertainty, agents may choose to act to reduce uncertainty or they may act despite that uncertainty.In communicative settings, an important way of reducing uncertainty is by asking clarification questions (CQs). We predict that the decision to ask a CQ depends on both contextual uncertainty and the cost of alternative actions, and that these factors interact: uncertainty should matter most when acting incorrectly is costly. We formalize this interaction in a computational model based on expected regret: how much an agent stands to lose by acting now rather than with full information. We test these predictions in two experiments, one examining purely linguistic responses to questions and another extending to choices between clarification and non-linguistic action. Taken together, our results suggest a rational tradeoff: humans tend to seek clarification proportional to the risk of substantial loss when acting under uncertainty.
Paper Structure (11 sections, 4 equations, 3 figures)

This paper contains 11 sections, 4 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Experiments
  3. Experiment 1: Reacting to Questions
  4. Methods & Materials
  5. Results
  6. Experiment 2: Reacting to Directives
  7. Methods & Materials
  8. Results
  9. Computational Model
  10. Bayesian Model Fitting
  11. Discussion

Figures (3)

  • Figure 1: Left: Colored bars: Results of Experiment 1, showing proportions and 95% bootstrapped CIs of selected responses (colors) in different uncertainty conditions (x-axis) under different costs (facets). Under high uncertainty, no preference is mentioned, so the mention-some labels were assigned to responses randomly. Black dots: posterior predictive means (dots) and 95% credible intervals (black lines) of the computational model , under parameters fitted to human data via Bayesian data analysis. Right: Example item from Experiment 1 in the small option space, low uncertainty condition.
  • Figure 2: Results of Experiment 2: unnormalized mean likelihood rating of different reaction options, grouped by uncertainty and costliness. Different direct actions were condensed into a single label for analysis. Error bars indicate 95% bootstrapped CIs, and points represent individual trials.
  • Figure 3: Estimates of expected log-likelihoods (dots) with standard errors (bars) for each model. Higher values are better.