Source
Quillien (2026), “Luck and Intentional Action: A Causal Account”, LIT-132 (Open Mind 10, 857–883), read in full in NOTE-092. The supplementary information was not read.
What was actually shown
Participants rated four statements per scenario on 1–7 scales: that the agent did X intentionally, that X happened because the agent wanted X, that the agent knew how to do X, and that the action raised the probability of X.
- Study 1 (Knobe’s marksman; skill × morality, within-subjects; N = 198; pre-registered). Low skill lowered both intentionality (6.65 vs 5.19, F(1, 591) = 240) and causation (6.56 vs 5.35, F = 194). Standardized β: causation .46, know-how .33, probability-raising .14.
- Study 2 (Kraemer means/ends × morality, between-subjects; N = 673; not itself pre-registered, following pre-registered pilots). β: causation .56, know-how .20, probability-raising .16. In the morally bad vignette causation tracked the effect (5.39 vs 2.68, t(315) = 11.6). In the neutral vignette intentionality differed (2.70 vs 2.20, p = .009) and causation did not (2.22 vs 2.20, p = .91).
- Study 3 (gumball machine; knowledge × process, within-subjects; N = 142 after 56 of 198 were excluded; pre-registered). The process (normal or a mix-up after the choice) mattered more when the agent knew how the machine worked: Knowledge × Process F(1, 423) = 32 for intentionality and 36 for causation. Know-how and probability-raising showed no such interaction (p = .12, .79), as their accounts are “agnostic” about it. β: causation .53, know-how .27, probability-raising .10. A causation-only model fit better than a know-how × probability-raising model (conditional R² .57 vs .44, difference CI [.07, .19]).
- Across the 12 condition means, intentionality correlates with causation r(10) = .98, with know-how .82 and with probability-raising .88.
- Mediation (exploratory, not pre-registered): causation carries 64%, 45% and 49% of the luck effects.
What could have come out otherwise: causal judgements could have stayed flat under the luck manipulations, or the Study 3 interaction could have been absent or shown up in know-how instead. Neither happened. The neutral Kraemer case is where the prediction failed.
What this does not say
- That luck lowers intentionality by way of causal judgement. The mediation is exploratory and covers about half the effect, and the author calls the evidence “mostly correlational”. Know-how and probability-raising stay weakly or moderately associated with intentionality when causation is controlled (General Discussion).
- That the causal account explains the Kraemer effect. It does in the morally bad vignette only.
- That a computational model was fitted. The account is verbal: necessity and robustness are taken informally from counterfactual theories of causal judgement, and robustness is inferred from the manipulations, never measured (NOTE-092, corrections).
- That the causation item measures causal strength. “Because [she] wanted” is close in wording to the intentionality item, which invites consistency effects the author himself raises.
- That the pattern is general. One scenario family per study; the morality contrast is confounded with difficulty; US Prolific workers only. The r = .98 rests on 12 condition means.
- Anything about what agents are. This is the folk concept of intentional action. The paper’s analogy with knowledge in Gettier cases, that the concept is “externalist”, is argued by analogy only. It does not study attributions to machines.
Connections
- Goldstein & Lederman (THEORY-616) concern whether and which desires to attribute to a language-model instance. This account concerns how, once a desire is attributed, people decide an outcome was done intentionally: the desire–outcome link must be robust.
- Pernu (LIT-144) shares the necessity and sufficiency vocabulary of counterfactual causation, but as metaphysics, not folk psychology.