That was close! A counterfactual simulation model of causal judgments about decisions

Abstract

How do people make causal judgments about other's decisions? Prior work has argued that judging causation requires going beyond what actually happened and simulating what would have happened in a relevant counterfactual situation. Here, we extend the counterfactual simulation model of causal judgments for physical events, to explain judgments about other agents' decisions. In our experiments, an agent chooses what path to take to reach a goal. In Experiment 1, participants either made hypothetical judgments about whether the agent would succeed were it to take a certain path, or counterfactual judgments about whether the agent would have succeeded had it taken a different path. In Experiment 2, participants made causal judgments about whether the agent succeeded or failed because of the path that it took. Our computational model accurately captured participants' judgments in both experiments and we find that causal judgments are better explained by counterfactuals rather than hypotheticals.

Mental jenga: A counterfactual simulation model of causal judgments about physical support.

“…The CSM captures people's causal judgments about dynamic physical events (Gerstenberg, Goodman, et al, 2021), omissions (Gerstenberg & Stephan, 2021), and static scenes involving physical support. While most existing causal theories only apply to event causation, the CSM provides a unifying framework that explains people's causal judgments across a variety of different kinds of causal relationships (Sosa et al, 2021;Wu et al, 2022). In this paper, we investigated people's judgments about block towers as a case study.…”

Towards a computational model of responsibility judgments in sequential human-AI collaboration

“…Existing theories about the cognitive process of responsibility attribution have established strong ties with causality (Pearl, 2009) and counterfactual reasoning (Byrne, 2016;Kahneman et al, 1982;Roese, 1997). Humans tend to consider an object, event, action or agent as (causally) responsible for an outcome if they can mentally simulate an alternative reality where that outcome would have been different if the candidate cause had not existed or occurred in the first place (Beckers, 2023;Chockler & Halpern, 2004;Gerstenberg et al, 2018;Halpern & Kleiman-Weiner, 2018;Lagnado et al, 2013;Langenhoff et al, 2021;Triantafyllou et al, 2022;Wu & Gerstenberg, 2024;Wu et al, 2023;Xiang et al, 2023;Zultan et al, 2012). In that context, Gerstenberg et al (2021) have developed the counterfactual simulation model (CSM), a computational model that accurately predicts the extent to which people perceive an object (e.g., a moving billiard ball) as a cause of an observed outcome (e.g., potting another ball).…”

“…More recently, Wu et al (2022Wu et al ( , 2023 have explored extensions of the CSM in social settings using Markov decision processes (MDPs) (Sutton & Barto, 2018) as generative models of agent behavior. Reminiscent of the results in the phys- Obstacles that are unknown to the agent in control but known to the other agent appear faded.…”

“…ical domain, they have shown that the CSM predicts people's judgments about the extent that a decision of a psychological agent caused an outcome based on counterfactual simulations where that agent has made a different decision (Wu et al, 2022). However, in the context of responsibility attribution, the shift of focus from physical objects to agents introduces additional complexity, since an agent's actions are conditioned on their epistemic state (Beckers, 2023;Franklin et al, 2022;Halpern & Kleiman-Weiner, 2018;Kirfel & Lagnado, 2021).…”

“…However, in the context of responsibility attribution, the shift of focus from physical objects to agents introduces additional complexity, since an agent's actions are conditioned on their epistemic state (Beckers, 2023;Franklin et al, 2022;Halpern & Kleiman-Weiner, 2018;Kirfel & Lagnado, 2021). To explore this further, Wu et al (2023) have experimented with a gridworld environment where an agent is trying to achieve an outcome in the presence of a second (potentially adversarial) agent. They have proposed an extension of the CSM that additionally models the first agent's belief about the second agent's intention and explains responsibility judgments by combining counterfactual simulations with intention inferences (Kleiman-Weiner et al, 2015).…”