Knobe's “person as moralist” view contests two related claims about human cognition: that it is clustered by discipline, much as university departments are, and that cognition in two “scientific” disciplines – folk psychology and causal inference – is analogous to scientific inquiry. Knobe then presents evidence that the psychology of intention and causation are “suffused with moral considerations” (sect. 5.3, para. 3), by which he means to show that there is neither a separation between disciplines, nor can reasoning about scientific topics be considered “scientific.”

We suggest another perspective on these moral asymmetries: that they are, at least in part, the consequences of early links between causal learning and social learning. Specifically, they are the result of how, as children, we gather evidence for such learning by observing and interacting with people. The adult moralist recruits knowledge gained from years of social evidence gathering – years spent learning from people and about people simultaneously. Therefore, to understand the adult moralist we must first understand her predecessor – the child scientist.

For a long time, developmental psychologists studied children's knowledge separately, according to domain. Some research examined early causal reasoning – intuitions about spatio-temporal relations (Leslie & Keeble Reference Leslie and Keeble1987; Oakes & Cohen Reference Oakes and Cohen1990), causal mechanisms (Bullock et al. Reference Bullock, Gelman, Baillargeon and Friedman1982; Schulz Reference Shultz1982), and the use of statistical cues in causal judgments (Gopnik et al. Reference Gopnik, Sobel, Schulz and Glymour2001; Sobel & Kirkham Reference Sobel and Kirkham2006). Other research focused on children's “mind-reading” abilities – what they knew about the intentions, desires, beliefs, and knowledge states underlying human actions (e.g., Lutz & Keil Reference Lutz and Keil2002; Repacholi & Gopnik Reference Repacholi and Gopnik1997; Wellman Reference Wellman1990; Woodward Reference Woodward1998). Others sought to understand children's knowledge of social categories (Bigler & Liben Reference Bigler and Liben2007; Heyman & Gelman Reference Heyman and Gelman2000), and still others focused on developing moral and conventional knowledge (e.g., Turiel Reference Turiel1983). The picture that emerged from these separate subfields is a lot like the mental university described by Knobe – separate departments for separate knowledge structures.

The domain-specific approach has lead to important discoveries about the content of early physical, biological, psychological, and social and moral knowledge. However, trying to apply this approach wholesale to learning processes has been less fruitful. Take causal learning: spatio-temporal cues and mechanism knowledge are useful, but are often unavailable. Statistical cues are also useful, but cannot help distinguish between causes and spurious correlations. Most often, ordinary causal learning depends on social interaction; evidence comes from doing things and watching others do things. Human actions are a child scientist's natural causal experiments (Gopnik et al. Reference Gopnik, Glymour, Sobel, Schulz, Kushnir and Danks2004; Schulz et al. Reference Schulz, Kushnir, Gopnik, Gopnik and Schulz2007).

Importantly, along with physical evidence (e.g., toys making noise, milk spilling, sticks breaking), causal actions contain valuable social evidence (a knowing glance at the right button, a cry of “oops!”, a desire for two short sticks). To evaluate the quality of causal evidence, children take knowledge, ability, and intention into account. For example, infants and preschoolers distinguish intentional actions from accidental ones, and this leads them to make different causal inferences (Carpenter et al. Reference Carpenter, Akhtar and Tomasello1998; Meltzoff Reference Meltzoff1995). Preschoolers prefer to learn new causal relations from knowledgeable rather than ignorant causal agents (Kushnir et al. Reference Kushnir, Wellman and Gelman2007). Children also treat causal evidence differently when a demonstrator is explicitly teaching them (Bonawitz et al. Reference Bonawitz, Shafto, Gweon, Chang, Katz, Schulz, Taatgen and van Rijn2009; Rhodes et al., in press). This evidentiary link is not limited to passive observations – it influences and interacts with the evidence children generate themselves through play. Thus, when children get ambiguous evidence from another person, they privilege evidence from their own past actions (Kushnir et al. Reference Kushnir, Wellman and Gelman2009), or are motivated to explore further to generate new evidence (Schulz & Bonawitz Reference Schulz and Bonawitz2007).

Other research suggests that children break traditional domain boundaries when learning about people, as well. For example, infants use contingency detection (Shimizu & Johnson Reference Shimizu and Johnson2004) or violations of contiguity (Saxe et al. Reference Saxe, Tzelnic and Carey2007; Spelke et al. Reference Spelke, Phillips and Woodward1995) to infer the presence of a psychological agent when other cues to agency are absent. Toddlers and preschoolers infer other people's preferences based on violations of random sampling, not merely positive regard and enthusiasm (Kushnir et al. Reference Kushnir, Xu and Wellman2010). Children may use statistical cues to track other individual regularities, such as personality traits (Siever et al, under review). They also readily track social regularities, such as norms and group characteristics (Kalish Reference Kalish2002; Rhodes & Gelman Reference Rhodes and Gelman2008).

From her earliest social experiences, the child scientist is engaged in a dynamic process of hypotheses formation, evidence-gathering, and theory change. The adult moralist, on the other hand, is asked to reason about a single instance of human behavior. The adult must therefore rely on her existing knowledge – knowledge acquired through this early learning process. We now have a better sense of where this knowledge begins; recent studies show early understandings of empathy, fairness, help, harm, and a host of moral precursors (e.g., Hamlin et al. Reference Hamlin, Wynn and Bloom2007). Knobe's analysis encourages us not to stop with domain-specific characterizations of knowledge. Instead, we should broaden how we view evidence from human actions to include their moral and normative dimensions, and investigate how these early evidential links give rise to later moral asymmetries in reasoning. This approach leads to interesting questions for research with adults, so long as we carefully distinguish between reasoning based on existing knowledge and the process of learning something new. When adults learn, for example, how do moral asymmetries change in response to further evidence? Is the evidence itself evaluated asymmetrically?

To conclude, while it may be wise at times to abandon the separation of disciplines, it seems premature to draw conclusions from Knobe's experimental data about the process by which they are integrated. To better understand this process, we need to look at learning at all ages, and continue research connecting moral development to both causal learning and social cognition.