Elsevier

Cognition

Volume 131, Issue 2, May 2014, Pages 284-299
Cognition

When children are better (or at least more open-minded) learners than adults: Developmental differences in learning the forms of causal relationships

https://doi.org/10.1016/j.cognition.2013.12.010Get rights and content

Highlights

  • We tested children’s and adults' ability to acquire abstract causal knowledge.

  • Participants saw and generalized from evidence consistent with 1 of 2 relationships.

  • Children efficiently learn abstract properties of causal relationships.

  • Children generalized better than adults when faced with an atypical relationship.

  • Our results match the predictions of a hierarchical Bayesian model.

Abstract

Children learn causal relationships quickly and make far-reaching causal inferences from what they observe. Acquiring abstract causal principles that allow generalization across different causal relationships could support these abilities. We examine children’s ability to acquire abstract knowledge about the forms of causal relationships and show that in some cases they learn better than adults. Adults and 4- and 5-year-old children saw events suggesting that a causal relationship took one of two different forms, and their generalization to a new set of objects was then tested. One form was a more typical disjunctive relationship; the other was a more unusual conjunctive relationship. Participants were asked to both judge the causal efficacy of the objects and to design actions to generate or prevent an effect. Our results show that children can learn the abstract properties of causal relationships using only a handful of events. Moreover, children were more likely than adults to generalize the unusual conjunctive relationship, suggesting that they are less biased by prior assumptions and pay more attention to current evidence. These results are consistent with the predictions of a hierarchical Bayesian model.

Introduction

In everyday life we reason about abstract and general causal principles as well as more concrete and specific causal relationships. For example, I not only know that I have to both put in the plug and push a button on my microwave to make it go, I know more generally that activating appliances has a characteristic causal structure. You both have to ensure that there is power coming to the appliance and that the on/off mechanism is set to “on”; neither cause is sufficient by itself. This sort of abstract principle has been referred to as an overhypothesis (Goodman, 1955, Kemp et al., 2007), that is, a hypothesis about the kinds of hypotheses that are likely to be true. Overhypotheses can shape subsequent inferences. If my new digital speaker fails to play, I’ll know to check both the power cord and the “play” button on my computer. These abstract principles thus constrain our hypotheses about specific causal relationships and help us learn more effectively (Kemp et al., 2007). They play a particularly important role in intuitive theories of biology, physics and psychology as “framework principles” (Gopnik and Wellman, 2012, Wellman and Gelman, 1992). So, where do these principles come from?

Recent work demonstrates that young children are remarkably skilled at learning specific causal relationships (e.g., Gopnik et al., 2004, Gweon and Schulz, 2011, Sobel and Kirkham, 2007, Kushnir and Gopnik, 2007, Schulz et al., 2007). For example, they can infer which blocks will activate a machine based on the contingencies between the blocks and the machine’s activation. But can children also learn more abstract causal principles, and use those principles to shape their subsequent inferences? There is one experiment showing that 4-year-olds can learn abstract causal categories of objects from data (Schulz, Goodman, Tenenbaum, & Jenkins, 2008) and one showing that they can learn abstract psychological categories (Seiver, Gopnik, & Goodman, 2012). There is also new evidence that in looking-time experiments, even infants can learn overhypotheses about properties of sets of objects (Dewar & Xu, 2010). There have also been studies examining the development of deductive reasoning and logical rules in children (e.g., Dias and Harris, 1990, Markovits and Vachon, 1990). But there have been no studies examining whether children can learn abstract principles about the logical form of causal relationships, or comparing children’s and adults’ abilities to do so. In this paper, we show that 4- and 5-year-old children can learn such principles, and can use them to design effective actions. In some circumstances, children learn these abstract causal principles more easily than adults do.

We contrast two abstract causal principles (overhypotheses) about the forms that relationships take in a causal system. One is that relationships have a disjunctive form, in which each cause has an independent probability of bringing about an effect. This form is pervasive in the literature on adult causal inference (e.g., Cheng, 1997, Griffiths and Tenenbaum, 2005). For example, a burglar alarm may be triggered by an intruder or the wind, and a fever may result from a virus or a bacterium. The other overhypothesis is that causal relationships have a conjunctive form in which individual causes are unable to produce an effect, but multiple causes in conjunction can do so. For example, a microwave turns on when both the plug is connected and a button is pressed, but not if either of these causes occurs by itself; likewise, a heart attack may only result if a person has both high blood cholesterol and a particular genetic susceptibility. Knowing when a machine or a disease has a conjunctive form or a disjunctive form helps us make the right inferences when we want to use the machine or cure patients.

Lucas and Griffiths (2010) showed that adults can learn these overhypotheses about the forms of causal relationships and explained this process in terms of a hierarchical Bayesian model. In a hierarchical Bayesian model, the prior probability of an abstract causal principle is combined with observed data via Bayes’ rule. This determines the posterior probability of the principle. The process is hierarchical: evidence can inform both a lower-level hypothesis, such as one about which events are causes of an effect, and an overhypothesis that constrains or leads to that lower-level hypothesis, such as one about how likely causal events are in general, and what kinds of causal relationships apply in a domain. If young children can also learn and then exploit causal overhypotheses, this might help explain the swiftness and generality of early causal learning.

We can also ask whether there are developmental differences between children and adults. Adults appear to be biased towards expecting disjunctive relationships and learn these relationships more easily than conjunctive relationships (Lucas & Griffiths, 2010), a pattern that is consistent with the prevalence of disjunctive relationships in the literature in general (Cheng, 1997, Griffiths and Tenenbaum, 2005, Lu et al., 2008).

Intuitively, we might expect that children would find it more difficult to learn overhypotheses than adults, particularly unusual overhypotheses. After all, dating back to Piaget and Vygotsky, researchers have often assumed that children move from more concrete to more general knowledge. Moreover, adults have both more knowledge and more developed information-processing capacities than children.

However, many developmentalists have recently argued for a Bayesian approach to cognitive development and particularly causal learning (e.g., Gopnik and Schulz, 2007, Gopnik and Wellman, 2012, Tenenbaum et al., 2011, Xu and Kushnir, 2013). The Bayesian approach suggests an alternative and somewhat counterintuitive developmental hypothesis. According to the Bayesian view, learning a new hypothesis involves combining the prior probability of that hypothesis with the observed data. Since children have less experience than adults, their “priors” will be different. In particular, they might be less biased towards hypotheses that are consistent with prior experience and more likely to accept hypotheses – including overhypotheses – that are consistent with new evidence. So children might actually be better at learning an unusual abstract causal principle than adults. In particular, a Bayesian approach suggests that children might find it easier to learn that causal relationships take a conjunctive form.

Differences in the prior expectations of adults and children could take different forms. Adults and children might simply assign high prior probabilities to different hypotheses so that adults have a strong a priori commitment to one kind of relationship, and children to another. Alternatively, adults and children might just differ in the strengths of their commitments, with children holding more diffuse beliefs. This latter possibility is consistent with the difference between a “low temperature” and a “high temperature” system, to borrow an analogy from statistical physics: the adults have congealed in their beliefs and are hard to shift, while the children are more fluid and consequently more willing to entertain new ideas.

If children are more flexible when learning about causal relationships, does this make them better learners than adults? The answer to this question depends on how we define learning. If we say that better learners are people who make correct inferences more often when they are faced with common and familiar situations, then the advantage goes to adults. On the other hand, if we focus on how quickly learners assimilate new information and update their beliefs, and come to understand novel situations, then flexibility – whether it is due to less-entrenched ideas about the structure of the world or an exploratory approach to updating beliefs – is a marker of better learning.

There is some reason to believe that children show this sort of superior flexibility in other domains. In cognitive neuroscience, researchers have suggested that young brains, with less top-down control, may be more flexible and plastic than older brains (Thompson-Schill, Ramscar, & Chrysikou, 2009). Moreover, young children are able to learn a wider variety of language sounds more easily than adults (Kuhl, 2004), are better than adults at discriminating between faces of non-human primates (Pascalis et al., 2005), and are more likely to look beyond the conventional uses of tools in order to solve problems (German & Defeyter, 2000). However, we do not know whether an analogous effect applies to children’s causal learning and their development of intuitive theories.

We examine this developmental hypothesis through head-to-head comparison of children and adults in a causal learning task. Specifically, we explore how children and adults learn that causal relationships follow a conjunctive or a disjunctive form. We examine whether children can form appropriate abstract generalizations, whether they use these abstract principles to shape more specific causal hypotheses and, finally, whether they are more willing to make these generalizations than adults.

Section snippets

Experiment 1: Learning the forms of causal relationships

Young children often have difficulty explicitly articulating causal hypotheses, so we designed a modified version of the experiment in Lucas and Griffiths (2010) that only required yes/no judgments. The experiment had two phases. First, in the training phase, children saw a set of events involving prospective causes (“blickets”) and an effect (activation of a “blicketness machine”). One set of events indicated that the machine worked disjunctively – each object individually did or did not

Experiment 2: Interventions and baselines

Experiment 2 made several changes to the design of Experiment 1 in order to rule out alternative hypotheses.

In Experiment 1, we asked participants if objects were blickets. Children and adults might treat this question differently. However, if someone genuinely believes that an event X causes an event Y, then she should produce X to bring about Y. Therefore, in Experiment 2 we asked participants which objects they would use to activate the machine.

As we noted above, one possibility is that

Experiment 3: Domain and language controls for adults

We designed a new experimental condition to test whether the syntactic and semantic details of our cover story were responsible for the adult disjunctive bias. Our first goal was to determine whether the blicket cover story – and electrical devices in general – caused the bias. Our second goal was to determine whether picking out causes using nouns (“blickets”) rather than adjectives (“blicket blocks”) caused the bias. With those goals in mind, we repeated the conjunctive condition of

General discussion

Our experiments were designed to explore two questions: whether children could learn high-level generalizations about the form of causal relationships at all, and, if they could, how they differed from adults. Our results show that children can learn the forms of causal relationships, and also that they can be more sensitive to evidence than adults are.

This result might seem surprising. After all, adults have more working memory and attentional resources than children and have seen a wider

Acknowledgements

This research was supported by the James S. McDonnell Foundation’s Causality Collaborative Initiative, Air Force Office of Scientific Research, Grants FA9550-07-1-0351 and FA9550-10-1-0232 to T.L.G., and National Science Foundation Grant (BCS-1023875) to A.G. and T.L.G. A preliminary version of this work was presented at the 32nd Annual Meeting of the Cognitive Science Society and appears in the proceedings of the conference (Lucas, Gopnik, & Griffiths, 2010).

References (32)

  • A. Gopnik et al.

    Mechanisms of theory formation in young children

    Trends in Cognitive Sciences

    (2004)
  • L.E. Schulz et al.

    Going beyond the evidence: Abstract laws and preschoolers’ responses to anomalous data

    Cognition

    (2008)
  • S. Carey

    Conceptual change in childhood

    (1985)
  • S. Carey

    The origin of concepts

    (2009)
  • P. Cheng

    From covariation to causation: A causal power theory

    Psychological Review

    (1997)
  • K.M. Dewar et al.

    Induction, overhypothesis, and the origin of abstract knowledge evidence from 9-month-old infants

    Psychological Science

    (2010)
  • M.G. Dias et al.

    The influence of the imagination on reasoning

    British Journal of Developmental Psychology

    (1990)
  • T.P. German et al.

    Immunity to functional fixedness in young children

    Psychonomic Bulletin and Review

    (2000)
  • N. Goodman

    Fact, fiction, and forecast

    (1955)
  • A. Gopnik et al.

    A theory of causal learning in children: Causal maps and Bayes nets

    Psychological Review

    (2004)
  • A. Gopnik et al.

    Words, thoughts, and theories

    (1997)
  • A. Gopnik et al.

    Causal learning: Psychology, philosophy, and computation

    (2007)
  • A. Gopnik et al.

    Reconstructing constructivism: Causal models, Bayesian learning mechanisms, and the theory

    Psychological Bulletin

    (2012)
  • T.L. Griffiths et al.

    Structure and strength in causal induction

    Cognitive Psychology

    (2005)
  • H. Gweon et al.

    16-month-olds rationally infer causes of failed actions

    Science

    (2011)
  • C. Kemp et al.

    Learning overhypotheses with hierarchical Bayesian models

    Developmental Science

    (2007)
  • Cited by (0)

    View full text