Our main goal in the present set of studies was to re-visit the question whether people are capable of inducing causal models from covariation data alone without further cues, such as temporal order. In the literature there has been a debate between bottom-up and top-down learning theories in causal learning. Whereas top-down theorists claim that in structure induction, covariation information plays none or only a secondary role, bottom-up theories, such as causal Bayes net theory, assert that people are capable of inducing structure from conditional dependence and independence information alone. Our three experiments suggest that both positions are wrong. In simple three-variable domains people are indeed often capable of reliably picking the right model. However, this can be achieved by simple heuristics that do not require complex statistics.

Causal graphical models (CGMs) have become popular in numerous domains of psychological research for representing people’s causal knowledge. Unfortunately, however, the CGMs typically used in cognitive models prohibit representations of causal cycles. Building on work in machine learning, we propose an extension of CGMs that allows cycles and apply that representation to one real-world reasoning task, namely, classification. Our model’s predictions were assessed in experiments that tested both probabilistic and deterministic causal relations. The results were qualitatively consistent with the predictions of our model and inconsistent with those of an alternative model. We naturally reason about causally related events that occur in cycles. In economics, we expect that an increase in corporate hiring may increase consumers ’ income and thus their demand for products, leading to a further increase in hiring. In meteorology, we expect that melting tundra due to global warming may release the greenhouse gas methane, leading to yet further warming. In psychology, we expect that clinicians will affect (hopefully help) their clients but also recognize the clients often affect the clinicians. Many psychologists investigate causal reasoning using a formalism known as Bayesian networks or causal graphical models (hereafter, CGMs). CGMs are one hypothesis for how people reason with causal knowledge. There are claims that causal learning amounts to acquiring the structure and/or parameters of a CGM (Cheng, 1997; Gopnik et al.,

Bayesian models of human causal induction rely on assumptions about people’s priors that have not been extensively tested. We empirically estimated human priors on the strength of causal relationships using iterated learning, an experimental method where people make inferences from data generated based on their own responses in previous trials. This method produced a prior on causal strength that was quite different from priors previously proposed in the literature on causal induction. The predictions of Bayesian models using different priors were then compared against human judgments of strength of causal relationships. The empirical priors estimated via iterated learning resulted in the best predictions.

Conjunctive causes are causes that all need to be present for an effect to occur. They contrast with independent causes that by themselves can each bring about an effect. We extend existing “causal power ” representations of independent causes to include a representation of conjunctive causes. We then demonstrate how independent vs. conjunctive representations imply sharply different patterns of reasoning (e.g., explaining away effects for independent causes as compared to exoneration effects for conjunctive causes). An experiment testing how people reason with independent and conjunctive causes found that their inferences generally matched the model’s prediction, albeit with some important exceptions. Rather than operating in a vacuum, causes frequently interact with other factors to produce their effects. For example, the conjunction of two or more variables is often necessary for an outcome to occur. A spark may only produce fire if there is fuel to ignite, a virus may only cause disease if one’s immune system is suppressed, the motive to commit murder may result in death only if the means to carry out the crime are available. Sometimes, conjunctive causes take the form of enablers. For example, the presence of oxygen enables fire given spark and fuel. In contrast, disablers interact with existing causes by preventing normal outcomes. Although the eight ball’s path to the side pocket may appear inevitable, it may be interrupted by an earthquake, a falling ceiling tile, or a spilled beer. The last 20 years has seen a growing interest in the role of causal knowledge in numerous areas of cognition. Many studies have investigated how causal relations are learned from observed correlations (Cheng, 1997; Gopnik et al.,

We present an introduction to Bayesian inference as it is used in probabilistic models of cognitive development. Our goal is to provide an intuitive and accessible guide to the what, the how, and the why of the Bayesian approach: what sorts of problems and data the framework is most relevant for, and how and why it may be useful for developmentalists. We emphasize a qualitative understanding of Bayesian inference, but also include information about additional resources for those interested in the cognitive science applications, mathematical foundations, or machine learning details in more depth. In addition, we discuss some important interpretation issues that often arise when evaluating Bayesian models in cognitive science.

continuous causes

Agents situated in a dynamic environment with an initially unknown causal structure, which, moreover, links certain behavioral choices to rewards, must be able to learn such structure incrementally on the fly. We report an experimental study that characterizes human learning in a controlled dynamic game environment, and describe a computational model that is capable of similar learning. The model learns by building up a representation of the hypothesized causes and effects, including estimates of the strength of each causal interaction. It is driven initially by simple guesses regarding such interactions, inspired by events occurring in close temporal succession. The model maintains its structure dynamically (including omitting or even reversing the current best-guess dependencies, if warranted by new evidence), and estimates the projected probability of possible outcomes by performing inference on the resulting Bayesian network. The model reproduces the human performance in the present dynamical task.

studies of concept learning in adults address the learning of novel concepts, but much of learning involves the updating and restructuring of familiar concepts. Research on conceptual change explores this issue directly but differs greatly from the formal approach of the adult learning studies. This paper bridges these two areas to advance our knowledge of the mechanisms underlying concept restructuring. The main idea behind this approach is that concepts are built on causal-explanatory knowledge, and hence, models of causal induction may help to clarify the mechanisms of the restructuring process. A new paradigm is presented to study the learning and revising of causal networks. Experiments 1 and 2 showed that learners’ prior beliefs about the causal relations in a domain affected their hypotheses as they began to infer the correct causes. First, when the prior learning suggested evidence against some of the incorrect causes, this helped learners to focus on the correct causes later in learning. Second, the prior causal beliefs were difficult to give up, and they biased learners away from the correct causes that competed to explain the same effects. Experiment 3 showed that learning by intervention, as opposed to observation, affected the concept restructuring process in different ways, depending on what interventions were chosen and by whom. People choosing their own