Computational Modeling of Backwards-Blocking Reasoning in Human Adults

Abstract

Causal reasoning is a fundamental cognitive ability that enables humans to learn about the complex interactions in the world around them. However, it remains unknown whether causal reasoning is underpinned by a Bayesian mechanism or an associative one. For example, some maintain that a Bayesian mechanism underpins human causal reasoning because it can better account for backward-blocking (BB) and indirect screening-off (IS) findings than certain associative models. However, the evidence is mixed about the extent to which learners engage in both kinds of reasoning. Here, we report an experiment and several computational models that examine to what extent adults engage in BB and IS reasoning using the blicket-detector design. The results revealed that adults’ causal ratings in a backwards-blocking and indirect screening-off condition were consistent with associative rather than a Bayesian computational model. These results are interpreted to mean that adults use associative processes to reason about causal events.

When Naïve Pedagogy Breaks Down: Adults Rationally Decide How to Teach, but Misrepresent Learners’ Beliefs

“…In this paradigm, knowledgeable agents can choose different combinations of objects to place on the blicket detector, with the goal of getting a naïve learner to infer the underlying rule. Blicket‐detector paradigms had their origins in developmental science (Gopnik, 2012; Griffiths, Sobel, Tenenbaum, & Gopnik, 2011; Kushnir, Gopnik, Lucas, & Schulz, 2010) but have since been used extensively with adults as well (e.g., Benton & Rakison, 2020; Gelpi, Prystawski, Lucas, & Buchsbaum, 2020; Griffiths et al., 2011; Herbst, Lucas, & Buchsbaum, 2017; Tenenbaum & Griffiths, 2003), as they provide a simple experimental paradigm to test causal learning without any interference from people's world knowledge (a potential limitation that we return to in the discussion). We implemented a causal structure we expected to be learnable but not trivially obvious, drawing on prior research showing that a causal relationship based upon the logical operator “and”—where two blocks in conjunction are required to activate a machine—is not immediately obvious to adults (see Gopnik et al., 2017; Lucas, Bridgers, Griffiths, & Gopnik, 2014).…”