The authors outline a cognitive and computational account of causal learning in children. They propose that children use specialized cognitive systems that allow them to recover an accurate “causal map ” of the world: an abstract, coherent, learned representation of the causal relations among events. This kind of knowledge can be perspicuously understood in terms of the formalism of directed graphical causal models, or Bayes nets. Children’s causal learning and inference may involve computations similar to those for learning causal Bayes nets and for predicting with them. Experimental results suggest that 2to 4-year-old children construct new causal maps and that their learning is consistent with the Bayes net formalism.

A scheme is described for locally Bayesian parameter updating in models structured as successions of component functions. The essential idea is to back-propagate the target data to interior modules, such that an interior component’s target is the input to the next component that maximizes the probability of the next component’s target. Each layer then does locally Bayesian learning. The approach assumes online trial-by-trial learning. The resulting parameter updating is not globally Bayesian but can better capture human behavior. The approach is implemented for an associative learning model that first maps inputs to attentionally filtered inputs and then maps attentionally filtered inputs to outputs. The Bayesian updating allows the associative model to exhibit retrospective revaluation effects such as backward blocking and unovershadowing, which have been challenging for associative learning models. The back-propagation of target values to attention allows the model to show trial-order effects, including highlighting and differences in magnitude of forward and backward blocking, which have been challenging for Bayesian learning models.

The dynamics model, which is based on L. Talmy’s (1988) theory of force dynamics, characterizes causation as a pattern of forces and a position vector. In contrast to counterfactual and probabilistic models, the dynamics model naturally distinguishes between different cause-related concepts and explains the induction of causal relationships from single observations. Support for the model is provided in experiments in which participants categorized 3-D animations of realistically rendered objects with trajectories that were wholly determined by the force vectors entered into a physics simulator. Experiments 1–3 showed that causal judgments are based on several forces, not just one. Experiment 4 demonstrated that people compute the resultant of forces using a qualitative decision rule. Experiments 5 and 6 showed that a dynamics approach extends to the representation of social causation. Implications for the relationship between causation and time are discussed.

Five studies investigated (a) children’s ability to use the dependent and independent probabilities of events to make causal inferences and (b) the interaction between such inferences and domain-specific knowledge. In Experiment 1, preschoolers used patterns of dependence and independence to make accurate causal inferences in the domains of biology and psychology. Experiment 2 replicated the results in the domain of biology with a more complex pattern of conditional dependencies. In Experiment 3, children used evidence about patterns of dependence and independence to craft novel interventions across domains. In Experiments 4 and 5, children’s sensitivity to patterns of dependence was pitted against their domain-specific knowledge. Children used conditional probabilities to make accurate causal inferences even when asked to violate domain boundaries. The past two decades of research have demonstrated that young children understand cause and effect in a wide range of contexts. By the age of 4, children’s folk physics includes knowledge about the causal relationship between object properties and object motion

computation. In preparation. Address for correspondence:

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.

Attempts to characterize people's causal knowledge of a domain in terms of causal network structures miss a key level of abstraction: the laws that allow people to formulate meaningful causal network hypotheses, and thereby learn and reason about novel causal systems so e#ectively. We outline a preliminary framework for modeling causal laws in terms of generative grammars for causal networks. We then present an experiment showing that causal grammars can be learned rapidly in a novel domain and used to support one-shot inferences about the unobserved causal properties of new objects. Finally, we give a Bayesian analysis explaining how causal grammars may be induced from the limited data available in our experiments.

Abstract We propose Markov random fields (MRFs) as a probabilistic mathematical model for incorporating the internal states of other agents, both human and robotic, into robot decision making. By using estimates of Theory of Mind (ToM), the “mental ” states of other agents can be incorporated into decision making through statistical inference, allowing robots to balance their own goals and internal objectives with those of other collaborating agents. The MRF model is wellsuited to domains in which the joint probability over latent (action) and observed (perceived) variables can be factored into pairwise interactions between these variables. Specifically, these interactions occur through functions that evaluate “local evidence ” between an observed and latent variable and “compatibility ” between a pair of latent variables. We will model some experimental findings from ToM research and show how the MRF model can be applied to a social robotics task. We will also describe how to use belief propagation on a multi-robot MRF as a novel approach to

In multiple‐cue learning people acquire information about cue‐outcome relations and combine these into predictions or judgments. Previous studies claim that people can achieve high levels of performance without explicit knowledge of the task structure or insight into their own judgment policies. It has also been argued that people use a variety of suboptimal strategies to solve such tasks. In two experiments we re‐examined these conclusions by introducing novel measures of task knowledge and self‐insight, and using ‘rolling regression ’ methods to analyze individual learning. Participants successfully learned a four‐cue probabilistic environment and showed accurate knowledge of both the task structure and their own judgment processes. Learning analyses suggested that the apparent use of suboptimal strategies emerges from the incremental tracking of statistical contingencies in the environment. These findings have wide repercussions for the study of multicue learning in both normal and patient populations. Insight 3

How do humans learn contingencies between events? Both pathway-strengthening and inference-based process models have been proposed to explain contingency learning. We propose that each of these processes is used in different conditions. Participants viewed displays that contained single or paired objects and learned which displays were usually followed by the appearance of a dot. Some participants predicted whether the dot would appear before seeing the outcome, whereas other participants were required to respond quickly if the dot appeared shortly after the display. In the prediction task, instructions guiding participants to infer which objects caused the dot to appear were necessary in order for contingencies associated with one object to influence participants ’ predictions about the object with which it had been paired. In the response task, contingencies associated with one object affected responses to its pair mate irrespective of whether or not participants were given causal instructions. Our results challenge single-mechanism accounts of contingency learning and suggest that the mechanisms underlying performance in the two tasks are distinct.