When correlation equals causation: A behavioral and computational account of second-order correlation learning in children

Abstract

We examined 2-and 3-year-old children's ability to use second-order correlation learning-in which a learned correlation between two pairs of features (e.g., A and B, A and C) is generalized to the noncontiguous features (i.e., B and C)-to make causal inferences. Previous findings showed that 20-and 26-month-olds can use second-order correlation learning to learn about static and dynamic features in category and non-category contexts. The present behavioral study and computational model extend these findings to show that 2-and 3-year-old children can detect the second-order correlation between an object's surface feature and its capacity to activate a novel machine but only if the children had encoded the first-order correlations on which the second-order correlation was based. These results have implications for children's developing information-processing capacities on their ability to use second-order correlations to infer causal relations in the world.

Exploring the foundations of early scientific literacy: Children’s causal stance.

“…Coming to understand the world in ways that support effective reasoning, accurate prediction, and adaptive behavior requires keen attunement to causality. Accordingly, perception of physical causality emerges early (Mascalzoni et al, 2013; Oakes & Cohen, 1995), and an extensive literature describes children’s rapidly-developing abilities to infer causality more broadly from patterns of covariation and the outcomes of active interventions (e.g., Benton et al, 2021; Goddu & Gopnik, 2020; Goddu et al, 2021; Gopnik et al, 2001). Not only are young children capable of perceiving and reasoning about causality, but they appear to have a strong intrinsic motivation to acquire causal understanding of the world around them.…”

Infants’ Attributions of Insides and Animacy in Causal Interactions

“…The third account of a sole ontogenetic origin to causal representations is that causality is a counterfactual-supporting dependence relation inferred from patterns of covariation (Chaput & Cohen, 2001). Such accounts are usually examined more in the context of adult causal reasoning (Cheng, 1997;Pearl, 2000) or causal learning in 2-4-year-old children (Benton, Rakison, & Sobel, 2021;Gopnik, Sobel, Schulz, & Glymour, 2001;Gopnik et al, 2004).…”

‘But what is the mechanism?’: Demystifying the ever elusive ‘developmental mechanism’

“…Note this entire description would be the mechanism of how second-order correlation learning might unfold, from the start stage to the final stage, to enable children to know that the test object with the purple diamond on it is causally efficacious. Benton et al (2021) found that children activated the machine with the test object that possessed the purple diamond but not with the test object that possessed the yellow circle. Given that there was no other basis on which to choose between the two test objects other than the embodied second-order correlations, this finding supported the idea that children's causal inferences were based on second-order correlation learning.…”

“…To illustrate how this definition can be used to construct complete definitions of developmental mechanisms of cognitive change, consider second‐order correlation learning as one example of such a mechanism (Benton et al, 2021; Rakison & Benton, 2019; Yermolayeva & Rakison, 2016). This is a domain‐general, associative‐learning mechanism that enables learners to detect the relation between non‐contiguously presented features and cues.…”

“…The second aim is to provide a concrete example of a developmental mechanism of cognitive change, discuss how it works in detail, and show how it fits the definition of mechanism that is offfered. The mechanism that I will focus on is second‐order correlation learning (e.g., Benton et al, 2021; Rakison & Benton, 2019; Yermolayeva & Rakison, 2016). My hope is that by describing how this mechanism works and unfolds in detail, this description can serve as a useful model of what a complete description of a developmental mechanism looks like.…”