Inducing causal relationships from observations is a classic problem in scientific inference, statistics, and machine learning. It is also a central part of human learning, and a task that people perform remarkably well given its notorious difficulties. People can learn causal structure in various settings, from diverse forms of data: observations of the co-occurrence frequencies between causes and effects, interactions between physical objects, or patterns of spatial or temporal coincidence. These different modes of learning are typically thought of as distinct psychological processes and are rarely studied together, but at heart they present the same inductive challenge—identifying the unobservable mechanisms that generate observable relations between variables, objects, or events, given only sparse and limited data. We present a computational-level analysis of this inductive problem and a framework for its solution, which allows us to model all these forms of causal learning in a common language. In this framework, causal induction is the product of domain-general statistical inference guided by domain-specific prior knowledge, in the form of an abstract causal theory. We identify 3 key aspects of abstract prior knowledge—the ontology of entities, properties, and relations that organizes a domain; the plausibility of specific causal relationships; and the functional form of those relationships—and show how they provide the constraints that people need to induce useful causal models from sparse data.

The ability to derive predictions for the outcomes of potential actions from observational data is one of the hallmarks of true causal reasoning. We present four learning experiments with deterministic and probabilistic data showing that people indeed make different predictions from causal models, whose parameters were learned in a purely observational learning phase, depending on whether learners believe that an event within the model has been merely observed (“seeing”) or was actively manipulated (“doing”). The predictions reflect sensitivity both to the structure of the causal models and to the size of their parameters. This competency is remarkable because the predictions for potential interventions were very different from the patterns that had actually been observed. Whereas associative and probabilistic theories fail, recent developments of causal Bayes net theories provide tools for modeling this competency. Causal knowledge underlies our ability to predict future events, to explain the occurrence of present events, and to achieve goals by means of actions. Thus, causal knowledge belongs to one of our most central cognitive competencies. However, the nature of causal knowledge has been debated. A number of philosophers and

How do humans discover causal relations when the effect is not immediately observable? Previous experiments have uniformly demonstrated detrimental effects of outcome delays on causal induction. These findings seem to conflict with everyday causal cognition, where humans can apparently identify long-term causal relations with relative ease. Three experiments investigated whether the influence of delay on adult human causal judgments is mediated by experimentally induced assumptions about the timeframe of the causal relation in question, as suggested by Einhorn & Hogarth (1986). Causal judgments generally decreased when a delay separated cause and effect. This decrease was less pronounced when the thematic context of the causal relation induced participants to expect a delay. Experiment 3 ruled out an alternative explanation of the effect based on variations of cue and outcome saliencies, and showed that detrimental effects of delay are reduced even more when instructions explicitly mentioned the timeframe of the causal relation in question. Knowledge thus mediates the impact of

computation. In preparation. Address for correspondence:

www.elsevier.com/locate/cogpsych

Associative learning theory postulates two main determinants for human causal learning: contingency and contiguity. In line with such an account, participants in Shanks, Pearson, and Dickinson (1989) failed to discover causal relations involving delays of more than two seconds. More recent research has shown that the impact of contiguity and delay is mediated by prior knowledge about the timeframe of the causal relation in question. Buehner and May (2002, 2003) demonstrated that the detrimental effect of delay can be significantly reduced if reasoners are aware of potential delays. Here we demonstrate for the first time that the negative influence of delay can be abolished completely by a subtle change in the experimental instructions. Temporal contiguity is thus not essential for human causal learning. An associative learning analysis of human causal learning postulates two main determinants of judged causal strength: the contingency and the contiguity between the potential cause (cue) and the effect (outcome) (e.g., see Shanks & Dickinson, 1987). Empirical research in the last decades has largely focused on the congruency between cue–outcome contingency and judged causal strength. Early reports suggested that human causal judgements closely track variations in cue-outcome contingency (e.g., Jenkins & Ward, 1965), while more recent studies revealed a more complex picture (e.g., Chapman & Robbins, 1990). In fact, the theoretical and empirical relations between contingency and judged causality are still the subject of a hot debate

Leading accounts of judgment under uncertainty evaluate performance within purely statistical frameworks, holding people to the standards of classical Bayesian (Tversky & Kahneman, 1974) or frequentist (Gigerenzer & Hoffrage, 1995) norms. We argue that these frameworks have limited ability to explain the success and flexibility of people's real-world judgments, and propose an alternative normative framework based on Bayesian inferences over causal models. Deviations from traditional norms of judgment, such as "base-rate neglect", may then be explained in terms of a mismatch between the statistics given to people and the causal models they intuitively construct to support probabilistic reasoning. Four experiments show that when a clear mapping can be established from given statistics to the parameters of an intuitive causal model, people are more likely to use the statistics appropriately, and that when the classical and causal Bayesian norms differ in their prescriptions, people's judgments are more consistent with causal Bayesian norms.

People learn quickly when reasoning about causal relationships, making inferences from limited data and avoiding spurious inferences. Efficient learning depends on abstract knowledge, which is often domain or context specific, and much of it must be learned. While such knowledge effects are well documented, little is known about exactly how we acquire knowledge that constrains learning. This work focuses on knowledge of the functional form of causal relationships; there are many kinds of relationships that can apply between causes and their effects, and knowledge of the form such a relationship takes is important in order to quickly identify the real causes of an observed effect. We developed a hierarchical Bayesian model of the acquisition of knowledge of the functional form of causal relationships and tested it in five experimental studies, considering disjunctive and conjunctive relationships, failure rates, and cross-domain effects. The Bayesian model accurately predicted human judgments and outperformed several alternative models.

Three experiments examined human processing of stimuli as predictors and causes. In Experiments 1A and 1B, two serial events that both preceded a third were assessed as predictors and as causes of the third event. Instructions successfully provided scenarios in which one of the serial (target) stimuli was viewed as a strong predictor but as a weak cause of the third event. In Experiment 2, participants ’ preexperimental knowledge was drawn upon in such a way that two simultaneous antecedent events were processed as predictors or causes, which strongly influenced the occurrence of overshadowing between the antecedent events. Although a tendency toward overshadowing was found between predictors, reliable overshadowing was observed only between causes, and then only when the test question was causal. Together with other evidence in the human learning literature, the present results suggest that predictive and causal learning obey similar laws, but there is a greater susceptibility to cue competition in causal than predictive attribution. This paper examines differences between predictive and causal learning in humans. Events often occur in our environment according to a consistent temporal distribution. Some events occur simultaneously (e.g., the sound and sight of water running out of the tap), whereas other events occur sequentially (e.g., hunger dissipates after the intake of food). When the events repeatedly take place following a sequential distribution in time, the first event (i.e., the antecedent event) can become a signal for the occurrence of the second event (i.e., the subsequent event). Learning to predict the occurrence of an event on O.P. was supported by a postdoctoral fellowship from the Spanish

Causal knowledge enables us to predict future events, to choose the right actions to achieve our goals, and to envision what would have happened if things had been different. Thus, it allows us to reason about observations, interventions and counterfactual possibilities.