Neural Correlates of Causal Confounding

As scientists, we are keenly aware that if putative causes perfectly covary, the independent influence of neither can be discerned—a “no confounding” constraint on inference, fundamental to philosophical and statistical perspectives on causation. Intriguingly, a substantial behavioral literature suggests that naïve human reasoners, adults and children, are tacitly sensitive to causal confounding. Here, a combination of fMRI and computational cognitive modeling was used to investigate neural substrates mediating such sensitivity. While being scanned, participants observed and judged the influences of various putative causes with confounded or nonconfounded, deterministic or stochastic, influences. During judgments requiring generalization of causal knowledge from a feedback-based learning context to a transfer probe, activity in the dorsomedial pFC was better accounted for by a Bayesian causal model, sensitive to both confounding and stochasticity, than a purely error-driven algorithm, sensitive only to stochasticity. Implications for the detection and estimation of distinct forms of uncertainty, and for a neural mediation of domain-general constraints on causal induction, are discussed.

Extended Bayesian inference incorporating symmetry bias

In this study, we start by proposing a causal induction model that incorporates symmetry bias. This model has two parameters that control the strength of symmetry bias and includes conditional probability and conventional models of causal induction as special cases. It can reproduce causal induction of human judgment with high accuracy. We further propose a human-like Bayesian inference method to replace the conditional probability in Bayesian inference with the aforementioned causal induction model. In this method, two components coexist: the component of Bayesian inference, which updates the degree of confidence for each hypothesis, and the component of inverse Bayesian inference that modifies the model of each hypothesis. In other words, this method allows not only inference but also simultaneous learning. Our study demonstrates that the method with both Bayesian inference and inverse Bayesian inference enables us to deal flexibly with unsteady situations where the target of inference changes occasionally.

What Is Transferred in Causal Generalization Across Contexts?

The covariation and causal power account for causal induction make different predictions for what is transferred in causal generalization across contexts. Two experiments tested these predictions using hypothetical scenarios in which the effect of an intervention was evaluated between (Experiment 1) or within (Experiment 2) groups. Each experiment contained a manipulation of ΔP, power and their combination. Both experiments found that causal transfer was determined by ΔP rather than causal power. The overall transfer pattern supports ΔP transfer account rather than the other transfer accounts. Causal transfers based on ΔP are irrational, violating the coherence criterion of the causal power framework. The ΔP transfer is consistent with previous findings that ΔP is a main mental non-normative measure of causal strength in causal induction.

Correlation Detection with and without the Theories of Conditionals: A model update of Hattori & Oaksford (2007)

We view observational causal induction as a statistical independence test under rarity assumption. This paper complements the two-stage theory of causal induction proposed by Hattori and Oaksford (2007) with a computational analysis. We show that their dual-factor heuristic (DFH) model has a rational account as the square root of the index of (non-)independence under extreme rarity assumption, contrary to the criticism that the DFH model is non-normative (e.g., Lu et al., 2008). We introduce a model that considers the proportion of assumed-to-be rare instances (pARIs), which is the probability of biconditionals (according to several theories of compound conditionals) and can be seen as a simplified version of the DFH model. While being a single conditional probability, pARIs approximates the non-independence measure, the square of DFH. In reproducing the meta-analysis in Hattori and Oaksford (2007), we confirm that pARIs and DFH have the same level of descriptive adequacy, and that the two models have the highest fit among more than 40 models. Then, we critically examine the computer simulations which were central to the rational analysis in Hattori and Oaksford (2007). We point out two problems in their simluations: samples in some of the simulations being restricted to generative ones, and in-definite values of models because of the small samples. In the light of especially the latter problem of definability, pARIs shows higher applicability.

Formalizing Prior Knowledge in Causal Induction

Prior knowledge plays a central role in causal induction, helping to explain how people are capable of identifying causal relationships from small amounts of data. Bayesian inference provides a way to characterize the influence that prior knowledge should have on causal induction, as well as an explanation for how that knowledge could itself be acquired. Using the theory-based causal induction framework of Griffiths and Tenenbaum (2009), this chapter reviews recent work exploring the relationship between prior knowledge and causal induction, highlighting some of the ways in which people’s expectations about causal relationships differ from approaches to causal learning in statistics and computer science.