Causal Responsibility and Counterfactuals

David A. Lagnado, Tobias Gerstenberg, Ro'i ZultanCogn. sci 2013 · DOI

TL;DR

This paper proposes a novel framework for understanding how people assign causal responsibility, particularly in situations involving multiple agents. It argues that our judgments depend on both an agent's "pivotality" (whether they made a difference to the outcome) and their "criticality" (how important they were perceived to be before the outcome occurred), offering a more nuanced view than purely counterfactual accounts.

What Problem Does This Paper Address?

When multiple people contribute to an outcome, especially a negative one, how do we decide who is responsible? This is a complex problem, particularly when an outcome would have happened anyway due to the actions of others (over-determination). Traditional counterfactual theories of causation struggle here, as they often fail to assign responsibility when an agent's action wasn't strictly necessary for the outcome. This paper aims to bridge the gap between formal causal models and intuitive human judgments of responsibility, especially in multi-agent scenarios.

Methods

The authors build upon the structural model account of causality, which represents causal relationships as a system of equations. They extend this to responsibility by introducing two key concepts:

  • Pivotality: Whether an agent's action was a "but-for" cause, meaning the outcome would not have occurred without it, evaluated under specific contingencies. This is a retrospective measure.
  • Criticality: How important an agent's contribution was perceived to be before the outcome occurred. This is a prospective measure.

They propose a model where responsibility is a function of both these factors. The paper details experimental studies designed to test this criticality-pivotality model against simpler models and the structural model alone.

Datasets

Not discussed in the paper. The paper describes novel experimental paradigms (e.g., "the triangle game," "dot-clicking game") used to collect data on participants' responsibility attributions.

Key Insights and Ideas

  • Beyond Pivotality: Simply being pivotal isn't enough to explain all responsibility judgments. People consider how critical an agent was before the event, even if they were pivotal.
  • Criticality Matters: In situations where agents are equally pivotal, their perceived criticality significantly influences how much responsibility is assigned.
  • Prospective vs. Retrospective: Our judgments blend backward-looking assessments (pivotality) with forward-looking considerations (criticality), mirroring legal and moral concepts.
  • Structural Models as a Tool: The structural model framework provides a powerful way to formalize causal relationships and test hypotheses about intuitive judgments, even if it doesn't perfectly capture all nuances.
  • Teamwork and Responsibility: In group settings, the causal structure (e.g., conjunctive vs. disjunctive tasks) dramatically affects how responsibility is distributed.

Key Results

The experimental findings largely support the proposed criticality-pivotality model.

  • Heuristic Criticality Model: This model, which assigns criticality based on whether an agent's success was necessary for a team win, predicted participants' criticality judgments well.
  • Combined Model: A model combining the heuristic criticality model with the structural model of pivotality provided the best fit for participants' responsibility attributions.
  • Pivotality vs. Criticality Trade-off: In situations where criticality and pivotality were manipulated independently, both factors were shown to influence responsibility judgments. Participants attributed more responsibility when criticality was high, even if pivotality was lower, and vice versa.
  • Changes Away from Pivotality: The paper suggests that responsibility can also be graded based on how difficult it is to make an agent non-pivotal, indicating a notion of "robustness" in causal responsibility.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Model Dependence: Responsibility judgments are model-dependent, meaning different causal models of the same situation can lead to different attributions. Future work could explore how people construct these models.
  • Unit of Change: The definition of "pivotality" relies on the number of changes needed to alter an outcome. Defining the "unit of change" can be complex and affect responsibility assignments.
  • Hierarchical Structures: The current work focuses on simple group structures. Future research should investigate responsibility in more complex hierarchical settings (e.g., boss-employee relationships).
  • Shared vs. Group Responsibility: The paper primarily focuses on individual responsibility within a group. Further exploration of "group responsibility" as a distinct concept is needed.
  • Intentions and Epistemic States: While the model explains causal responsibility, it doesn't fully account for blame, which often depends on intentions, knowledge, and foresight. Future work could integrate these mental states into causal models.
  • Generalizability: The principles of pivotality and criticality need to be tested in more complex scenarios and with a wider range of individual differences.

Summary updated

BibTeX
@article{lagnado2013causal,
  author = {David A. Lagnado and Tobias Gerstenberg and Ro'i Zultan},
  title = {{Causal Responsibility and Counterfactuals}},
  journal = {Cogn. sci},
  year = {2013},
  doi = {10.1111/cogs.12054},
}
Figure