Higher-level cognition depends on the ability to learn models of the world. We can characterize this at the computational level as a structure-learning problem with the goal of best identifying the prevailing causal relationships among a set of relata. However, the computational cost of performing exact Bayesian inference over causal models grows rapidly as the number of relata increases. This implies that the cognitive processes underlying causal learning must be substantially approximate. A powerful class of approximations that focuses on the sequential absorption of successive inputs is captured by the Neurath's ship metaphor in philosophy of science, where theory change is cast as a stochastic and gradual process shaped as much by people's limited willingness to abandon their current theory when considering alternatives as by the ground truth they hope to approach. Inspired by this metaphor and by algorithms for approximating Bayesian inference in machine learning, we propose an algorithmic-level model of causal structure learning under which learners represent only a single global hypothesis that they update locally as they gather evidence. We propose a related scheme for understanding how, under these limitations, learners choose informative interventions that manipulate the causal system to help elucidate its workings. We find support for our approach in the analysis of 3 experiments. (PsycINFO Database Record

Children between 5 and 8 years of age freely intervened on a three-variable causal system, with their task being to discover whether it was a common cause structure or one of two causal chains. From 6 or 7 years of age, children were able to use information from their interventions to correctly disambiguate the structure of a causal chain. We used a Bayesian model to examine children's interventions on the system; this showed that with development children became more efficient in producing the interventions needed to disambiguate the causal structure and that the quality of interventions, as measured by their informativeness, improved developmentally. The latter measure was a significant predictor of children's correct inferences about the causal structure. A second experiment showed that levels of performance were not reduced in a task where children did not select and carry out interventions themselves, indicating no advantage for self-directed learning. However, children's performance was not related to intervention quality in these circumstances, suggesting that children learn in a different way when they carry out interventions themselves.

To tackle a hard problem, it is often wise to reuse and recombine existing knowledge. Such an ability to bootstrap enables us to grow rich mental concepts despite limited cognitive resources. Here we present a computational model of conceptual bootstrapping. This model uses a dynamic conceptual repertoire that can cache and later reuse elements of earlier insights in principled ways, modelling learning as a series of compositional generalizations. This model predicts systematically different learned concepts when the same evidence is processed in different orders, without any extra assumptions about previous beliefs or background knowledge. Across four behavioural experiments (total n = 570), we demonstrate strong curriculum-order and conceptual garden-pathing effects that closely resemble our model predictions and differ from those of alternative accounts. Taken together, this work offers a computational account of how past experiences shape future conceptual discoveries and showcases the importance of curriculum design in human inductive concept inferences.

Interacting with a system is key to uncovering its causal structure. A computational framework for interventional causal learning has been developed over the last decade, but how real causal learners might achieve or approximate the computations entailed by this framework is still poorly understood. Here we describe an interactive computer task in which participants were incentivized to learn the structure of probabilistic causal systems through free selection of multiple interventions. We develop models of participants' intervention choices and online structure judgments, using expected utility gain, probability gain, and information gain and introducing plausible memory and processing constraints. We find that successful participants are best described by a model that acts to maximize information (rather than expected score or probability of being correct); that forgets much of the evidence received in earlier trials; but that mitigates this by being conservative, preferring structures consistent with earlier stated beliefs. We explore 2 heuristics that partly explain how participants might be approximating these models without explicitly representing or updating a hypothesis space.