Effective Actions

Abstract

Three studies look at whether the assumption of causal determinism (the assumption that all else being equal, causes generate effects deterministically) affects children’s imitation of modeled actions. We show that, even when the frequency of an effect is matched, both preschoolers (N = 60; mean: 56 months) and toddlers (N = 48; mean: 18 months) imitate actions more faithfully when modeled actions are deterministically rather than probabilistically effective. A third study suggests that preschoolers’ (N = 32; mean: 58 months) imitation is affected, not just by whether the agent’s goal is satisfied but also by whether the action is a reliable means to the goal. Children’s tendency to generate variable responses to probabilistically effective modeled actions could support causal learning. DIFFERENTIAL IMITATION 3 Imagine that every time your Uncle Robbie makes a soufflé it rises perfectly, but when your Uncle Sam makes a soufflé, sometimes it rises and sometimes it falls. Although you might learn to cook both by observing Uncle Sam’s failures and by observing Uncle Robbie’s successes, if you were learning from Uncle Robbie, you would probably imitate his technique faithfully, while if you were learning from Uncle Sam, you might be inclined to vary the recipe. That is, the precision with which you imitate an observed action might be affected by your beliefs about the efficacy of the action; optimal learning might depend on knowing when to imitate and when to explore. In this study, we look at whether a similar proposition is true for young children: do children differentially imitate deterministically and probabilistically effective actions? Previous research on children’s imitation raises a puzzle. On the one hand, children are very good at reproducing modeled actions. Indeed, in some contexts, children will faithfully copy even arbitrary, unnecessary actions. For instance, children will imitate elaborate, causally irrelevant routines to open a box even when the mechanisms that could be used to open the box directly are obvious (Horner & Whiten, 2005; Lyons, Young, & Keil, 2006; Whiten, Custance, Gomez, Teixidor, & Bard, 1996). Similarly, children will copy an actor who, for no apparent reason, activates a toy with her head, even though the children can (and often do) also activate the toy with their hands (Geregely, Bekkering, Kiraly, 2002; Meltzoff 1992). On the other hand, children will sometimes override modeled actions in order to generate their own means to inferred ends. Thus for instance, toddlers do not copy actions that fail to achieve the agent’s intended outcome (Meltzoff, 1995). If an adult pulls on a barbell toy but does not pull it apart, 18-month-olds do not imitate the ‘failed’ DIFFERENTIAL IMITATION 4 action. Instead they act to achieve the inferred goal of the action: they pull apart the toy. Critically children do not ‘read through the goals’ of the action if the action is performed by a machine rather than a person; the physical action by itself is not sufficient to lead children to compete the causal sequence. However, it is equally striking, and to our knowledge, less often noted, that children’s only cue that the action failed to fulfill the adult’s intention was the failure of the expected causal relationships (e.g., between pulling and separating). That is, the children were not given any linguistic or affective cues suggesting that the action failed to achieve the agent’s goals (the actor did not say ‘whoops’ or frown); the children simply saw that the actor pulled on the toy and the toy remained intact. Thus the only cue that the action “failed” to achieve the agent’s goals was the absence of a causal relationship that the child expected (e.g. between pulling and separating). In the context of intentional action, children seem to be able to use their knowledge of the causal structure of an event to infer the intentional structure of the event. The puzzle then concerns the role of causal knowledge in children’s imitative learning. Why do children sometimes seem to suspend their own causal knowledge in order to copy modeled actions faithfully (even when there are simpler means to the end) but at other times use their causal knowledge to override modeled actions in favor of novel means to inferred ends? What predicts the fidelity with which young children reproduce modeled actions? Gergely and colleagues have proposed that some differences in children’s imitation can be explained by assuming that children respect a principle of rational action (Csibra & Gergely, 1998; Gergely & Csibra, 1997; Gergely & Csibra, 2003; DIFFERENTIAL IMITATION 5 Gergely, Nadasdy, Csibra, & Biro, 1995). That is, children may assume that intentional actions performed by rational agents are optimal within the constraints of the situation. If children observe actions that appear to be sub-optimal (e.g., an adult using her head to activate a toy or using arbitrary routines to open a box), they may nonetheless assume that “there must be a good reason” (Gergeley et al., 2002) for the agent’s choice of action. In the absence of an obvious ‘good reason’ for the action, children might assume the action has an unobserved causal relationship to the effect or they might revise their understanding of the agent’s goals (e.g., they might assume the goal was to demonstrate a convention or ritual). In either case, children should imitate the modeled action faithfully. However, situational constraints can provide an obvious ‘good reason’ for the modeled action. As elegantly demonstrated by an extension of Meltzoff’s (1988) lightactivation paradigm, if the situational constraints on the actor and the child are different, the child may not imitate the modeled action. In the new paradigm, the actor again used her head to activate a toy, however, this time the actor’s hands were occupied holding a blanket. In this case, children did not imitate the head action; instead they activated the toy with their hands (Gergely et al., 2002). The situational constraint (holding the blanket) provided an explanation for the actor’s unusual action, thus we suggest, screening-off (Reichenbach, 1956) a causal role for the particular means used to achieve the dominant goal (activating the light). Since the child was not also imitating the ancillary goal of holding a blanket, this analysis freed the child to (a) infer the causal structure of the main event (depressing the button makes the light go on), and then (b) achieve the actor’s dominant goal by novel but simpler means (using their hands). DIFFERENTIAL IMITATION 6 We are sympathetic to the idea that children adopt a principle of rational action, but we note that this proposal does not resolve our original puzzle. The claim that children assume that rational agents act optimally with respect to situational constraints is prima facie incompatible with the fact that children recognize that adults’ actions are sometime ineffective. As researchers have noted, “judgments about the ‘rationality’ of means always translate into judgments of ‘efficacy’” (Gergely & Csibra, 2003, p. 290). If children assume that agents perform the most rational action available given the constraints of the situation, it is difficult to understand how children might construe modeled intentional actions as ‘failed’ actions. Why would a child assume that an adult who activates a toy with her head (instead of her free hands) is acting optimally, but that an adult who pulls on a toy but fails to separate it is not? It is tempting to conclude that children assume that modeled actions are optimal when the actions achieve the agent’s goal and not when they fail. Note however, that this presumes that children can simply ‘read off’ the success or failure of the agents’ goal from the sequence of events. This may indeed be the case when the agent provides explicit linguistic and affective cues about whether or not her goal has been achieved (e.g., “There!” or “Whoops!”; Carpenter, Akhtar, & Tomasello, 1998). However if the agent pulls on a toy, it is possible to infer that the agent failed in her goal to separate the toy, but it is also possible to infer that the agent succeeded in her goal to pull on the toy. Critically, if children always assume that agents act optimally, the inference that the adult succeeded should be the preferred inference. That is, under the assumption that adults always take the most rational action given the situational constraints, children should not infer that they could improve upon the observed action. DIFFERENTIAL IMITATION 7 Here we suggest that although children do assume that agents act rationally with respect to their goals, they do not make this assumption uncritically. We suggest that children analyze goal-directed actions in the context of their broader causal knowledge. As discussed, there is considerable evidence that, given common situational constraints, children faithfully imitate arbitrary, causally irrelevant actions (activating a toy with their heads, engaging in elaborate rituals to open a box). We suggest that this is because arbitrary actions are, by definition, actions about which children have few prior expectations. If children do not have sufficient prior causal knowledge to evaluate the efficacy of the modeled actions, we expect that children will adopt a principle of rational action and assume the adult actions are optimal. Provided the modeled actions are not screened-off by a known relationship to an ancillary goal, children should imitate such actions faithfully. Because children’s tendency to imitate arbitrary actions has been well established by previous research, we will not replicate that aspect of our analysis here. However, if children do have sufficient prior knowledge to evaluate the relationship of the modeled action to the goal, we predict that children will imitate the modeled action faithfully only if they construe the action as an optimal means to the inferred en