Abstract

1. Introduction

All of us, from the most sophisticated adults to the youngest children, often engage in what is commonly called “folk science,” that is, certain ways of understanding the natural and artificial world that arise more informally and not as direct reflections of formal instruction in scientific principles (Carey, 1988). There is now extensive work on how children and adults have developed folk psychologies (Wellman, 1990), folk physics (Proffitt, 1999; Vosniadou, 2001), and folk biologies (Inagaki & Hatano, 2002), as well as some indications of folk sciences in such areas as the behaviors of materials and substances (folk chemistry; Au, 1994), the behaviors of heavenly bodies (folk cosmology; Siegel, Butterworth & Newcombe, 2004), and the nature of value transactions (folk economics; Lakshminaryanan, Chen, & Santos, 2008). Without explicit instruction in such areas, people seem to develop domain-specific ways of thinking about relatively bounded sets of phenomena such as the behavior of solid objects, living kinds, and the minds of others.

These domain-specific understandings have been referred to as “intuitive theories” or “naïve theories,” on the assumption that they reflect sets of beliefs that cohere in a manner that resembles, in important respects, scientific theories (Carey, 1985; Carey & Spelke, 1996; Slaughter & Gopnik, 1996). In particular, intuitive theories are assumed to have coherence, consistency, and predictive value. In addition, such intuitive theories are thought to specify ontological kinds. Ontological kinds, in turn, are thought to be a product of causal explanatory systems that posit certain kinds as foundational entities within each system. Preschoolers, and perhaps even infants, are thought to have senses of ontological kinds that arise from their intuitive theories or folk sciences (Carey, 1985, 2009; Heyman, Phillips, & Gelman, 2003). Although researchers may disagree on just how early different theoretical domains appear, there appears to be considerable consensus on a folk physics and folk psychology having roots going back to infancy (Carey, 1985; Leslie, 1994). In addition, for 25 years, ever since Carey's pioneering work on conceptual change (Carey, 1985), researchers have asked if conceptual change in children might be analogous to the ways in which concepts change in the formal sciences as a consequence of theory change. If children's concepts are embedded in rich intuitive theories, perhaps it is useful to think of them as little scientists going through sequences of theoretical revolutions with corresponding conceptual changes.

At one level, this body of work is clearly right. From infancy onwards, humans seem to assess the domain of regularities they encounter and then reason in ways that reflect the causal regularities specific to that domain. For example, psychological entities seem to be governed by different causal principles (e.g., action at a distance) than the most typical physical mechanical ones and we all seem to constrain our beliefs in ways that honor these domain-specific principles. However, such domain specific constraints may not be sufficient for something to count as an intuitive theory, especially given that aspects of them may operate in other species as well (Carey & Spelke, 1996).

Naïve theories can also lead people astray. Indeed, a large subspecialty of cognitive science focusing on misconceptions has documented how laypeople and/or children make systematic mistakes in their reasoning about physical mechanics (e.g., Caramazza, McCloskey, & Green, 1981; Bertamini, Spooner & Hecht, 2004), biology (e.g., Inagaki & Hatano, 2002; Shtulman, 2006; Shtulman & Schulz, 2008) and psychology (e.g., Leslie, Friedman, & German, 2004; Malle, Knobe, & Nelson, 2007). Although such misconceptions may lead to misleading “deficit models” of cognitive development and instruction (Zimmerman, 2007), they certainly do exist and have been argued to serve as evidence for the coherence of folk science domains; coherence that drives people to make mistakes even in the face of real world counter evidence. Thus, misconceptions sometimes have been used as evidence for intuitive theories.

In short, there seems to be an emerging consensus about the existence of many folk sciences across all cultures that lead both to real successes at understanding the world and to misconceptions. Yet, one critical problem seems largely unaddressed. If we look in more detail at the nature of intuitive theories, how feasible is it that they should suffice as vehicles for understanding? Are they sufficiently well articulated, and in the right ways, to enable laypeople and children to make sense of the world around them? Do they meet reasonable levels of coherence, consistency, and predictive power to count as theories? Here, I argue that the answer is not straightforward. If one looks at intuitive theories and folk science through one lens, namely their sufficiency as stand alone explanatory systems, the matter seems quite hopeless and a real mystery as to how anyone gains explanatory insight into the workings of the world. That lens raises similar worries, however, when refocused on more formal science and, as a result, suggests a different, more socially embedded analysis of how formal science succeeds. This analysis, and the reasons that formal science succeeds, point toward a new way of looking at folk science as well, and a series of studies is then described showing that folk science is not only feasible but also that the foundations making it feasible are present at surprisingly early ages.

More specifically, this article considers the nature of the intuitive understandings in the folk sciences and documents their severe limitations when considered as mechanistic theories. They are remarkably incomplete and often contain inconsistencies and contradictions, and most lay people grossly underestimate these limitations as well. Although these shortcomings may seem to cripple any lay ability to make sense of the world, a different alternative emerges if we consider more recent views of the formal sciences in which gaps and inconsistencies and other kinds of gross simplifications are also shown to be commonplace. Formal sciences surmount those limitations by developing and navigating a social infrastructure of knowledge in which there are divisions of cognitive labor and sophisticated mechanisms for recognizing appropriate experts and knowing when and how to defer to them.

This alternative view of formal science suggests a reanalysis of folk science as well, in which roughly comparable mechanisms enable lay people to have a sense of relevant areas of expertise and how to defer to them. Despite having very crude internal models of how the world works, lay people may have sophisticated ways of navigating the terrain of expert clusters in their culture. Moreover, this ability may emerge early in development and may be how causal understanding is framed even by young children, that is, as embedded in larger social networks of knowledge resources. To illustrate this idea with one kind of example, three studies are described that document how children, even as they know very little about specific mechanisms, come to sense fertile vs. infertile areas of expertise. Finally, the implications of this alternative view of folk science are discussed both in terms of the development of explanatory understandings and in terms of the challenges that remain in understanding how folk science is used on a daily basis.

There have been many debates about the meanings of “success” or “progress” in the formal sciences and whether such meanings require a realist philosophy of science (Psillos, 1999, 2009). Here it is simply assumed that the formal sciences are successful because they enable their practitioners to characterize aspects of the world in ways that allow them to intervene on elements of the world so as to come closer to desired goals (Woodward, 2003) and to test hypotheses in more rigorous ways (including cases where intervention is not possible). Sciences make progress when they become more successful over extended time periods (allowing for shorter term dead ends and detours).

A folk science would be considered feasible to the extent that it enables laypeople to characterize the world in ways that enable them to intervene more effectively and confidently to achieve goals and that enable them to test and justify hypotheses. One key question asks whether folk science could be feasible in this sense if it is primarily understood in terms having internal mechanistic representations of the world. That is, are people's internal representations sufficiently articulated to directly enable them to have more effective and confident interventions and more powerful ways of testing and justifying hypotheses? I will argue that, in general they are not, but that there are other facets to folk science that give it much more power.

1.3. Science Communities

The story is more intricate with communities of scientists rather than lone individuals. Scientists, of course, almost invariably rely on each other in intricate and powerful ways. The idea of a lone wolf making great discoveries may have romantic appeal, but is not true in the history of the modern sciences (Dunbar & Blanchette, 1995). A central part of science as a group enterprise involves the two facets of the divisions of labor that distinguish all mature sciences: patterns of deference and understanding the distribution and relations among areas of expertise. Although the empirical studies in this article focus on the emerging ability to sense causal patterns above the level of mechanism and not specifically on folk science communities, the ability to sense rich causal patterns is an essential part of functioning within those communities. In particular, the sensing of such causal patterns is critical to determining plausible and fruitful areas of expertise in a community and to choosing between competing experts.

Scientists routinely defer to colleagues for data, theories, or results that are essential to their own enterprise. They do so with partial, sometimes quite minimal, understandings of their colleagues' areas of expertise. Most people doing cognitive science research with functional magnetic resonance imagery (fMRI) have very modest technical understandings of how signals are processed so as to yield the final results. They do believe, however, that they have good reason to trust those who are experts in the physical and computational aspects of fMRI. Many molecular biologists may have to defer in a massively parallel manner to experts on x-ray crystallography, bioinformatics, bioassay chips, and organic chemistry, among many other areas, in order to conduct their own very focused research project. In turn, all those to whom they defer also have their own networks of deference, which can create a branching pattern of the sort shown in Fig. 1, in which a researcher on cell movement must rely on an ever widening web of experts. (See Nersessian, in press, for other examples of collaborations across disciplines).

Open in a separate window

Fig. 1

Patterns of deference in the sciences. Any practicing scientist, or lab group, is part of a richly interconnected massive net of deference relations. Here, an expert on cell movement depends critically on experts in topics ranging from microscopy, to gene regulation, to protein synthesis to lattice structures. Each of these experts in turn defers to other experts in both asymmetrical and symmetrical patterns in an ever more branching net of which this figure only shows a small fragment.

The setting up of chains of deference draws on sophisticated cognitive processing. One certainly does not defer automatically to any one who makes a claim relevant to one's research. Cognitive neuroscientists, for example, will not generally pay attention to astrologists' claims about how the alignment of heavenly bodies influences the brain, no matter how earnest and persistent the astrologer is. In addition, scientists have to know when there is no need to defer. Some phenomena seem so transparent and self-evident that it is silly and wasteful to defer to “experts.” Thus, the cell biologist does not feel a need to defer to an expert on whether a cell is moving.

In other cases, scientists know that a domain does not have enough internal substance to require experts; it wears all its information “on its sleeve.” For example, if a dozen people contracted a new strain of influenza and all happened to be wearing striped shirts when they became ill, no one would seek out experts on people who wear striped shirts to understand the disease. However, if all the people who contracted the disease had diabetes, one might consult a diabetes expert to see if there was a causal pattern that linked diabetics to influenza (in fact there is, see Valdez, Narayan, Geiss, & Engelgau, 1999). Scientists make many decisions along these lines every day, decisions that critically influence how they amplify their own understandings. Even a scientist with a very modest knowledge of influenza would see the potentially greater usefulness of diabetes experts than alleged experts on people who wear striped shirts.

A related ability concerns skill at knowing something about knowledge clusters in areas other than one's own. Practicing molecular biologists, for example, not only need to know how to defer, they also need to know the layout of knowledge communities adjacent to their own domain. At least some of these intuitions seem to be driven by assumptions about the world as being causally heterogeneous (Cartwright, 1999; Keil, Stein, Webb, Billings & Rozenblit, 2008); namely, that there are unique causal patterns associated with various domains and that one can sense such differences without learning many details. For example, the casual patterns and principles associated with biochemistry might be seen as distinct from those associated with evolution. If a domain's principles are those governing physiological pathways and reactions, one might assume that an expert on one particular set of pathways and reactions will have much greater than average insight into other pathways and reactions than, say, an expert on natural selection. We assume there are strong similarities between the principles governing one pathway and those of another. Such an assumption follows from the idea of local property clusters as governed by a common set of causal relations and principles.

Scientists may also discover legitimate experts by finding or creating common grounds through physical models where they and other experts have aspects of models that “interlock” and thereby form a kind of conduit between two areas of expertise (Nersessian, in press). Thus, in one lab, a method was developed to record from multiple neurons in vivo such that the model system of neurons in culture “interlocks concepts, methods, and materials from biology, chemistry, and electrical engineering” (Nersessian, p.19). In many cases, chains of deference may spread out from such model systems.

In summary, there are many social dimensions to science that allow it, as a collective enterprise, to vastly surpass the powers of any one mind. Practicing scientists know how to tap into social networks of knowledge so as support their own gaps. Formal science works, and is feasible, just because of this aggregate community activity. Just as the division of physical labor added much more efficiency to economic systems (Durkheim, 1893/1997; Smith, 1776/1904), the division of cognitive labor in the sciences has enabled more efficient ways of doing scientific research. However, such divisions of cognitive labor depend on mechanisms of deference, evaluation, collaboration, and even competition. One major part of such mechanisms is having a sense of causal patterns above the level of mechanism and using that sense as a guide to the right areas of expertise.

2. A Preliminary Study on Plausible Areas of Expertise across Domains

To develop a method for studying plausible areas, a preliminary study asked if children come to distinguish causally empty categories from those having a causal richness worthy of expertise. It examined three different domains, living natural kinds, non-living natural kinds, and artifacts to see if the ability to distinguish plausible from implausible areas of expertise varied across domains. These three domains were chosen because they represent one of the largest cuts of distinct causal patterns in the natural and artificial worlds (Keil, 1979). Twenty five adults and 77 children distributed across grades K, 2, and 4 were asked to rate, on a four point scale, the extent to which each of twelve phenomena required an expert. Half the phenomena represented causally empty categories (e.g., “dogs with red collars”) and half represented causally rich ones (e.g., “dogs that are good at hunting”). A second design constraint was that there should be ample members in both plausible and implausible categories so that participants would not distinguish them on the basis that some categories were populated by more instances than others (there are just as many dogs with red collars as there are dogs that are good at hunting). The two living kind pairs referred to “…dogs that are good at hunting” vs. “…just those dogs with red collars” and “… just those flowers that bloom every year” vs. “… just those flowers planted in purple pots.” The two artifact pairs referred to “… just those cars that have electric (hybrid) engines” vs. “…just those cars with more bugs on the right side than the left side of the windshield” and “… just those books that are bought by teenagers” vs. “… just those books that have green covers.” The two non-living natural kind pairs referred to “… just those clouds that are in the sky during thunderstorms” vs. “…just those clouds that are shaped like teddy bears” and “… just on forest fires” vs. “… just those fires used to roast marshmallows.”

In this preliminary study, the youngest children first detected the plausible vs. implausible category difference for artifacts, but not till second grade and older did children also detect the plausible vs. implausible difference for living kinds and non-living natural kinds (all differences were significant at p < .01 in repeated measures ANOVAS and t-tests with corrections for multiple tests). Fourth graders were close to adult levels at detecting the contrast between the plausible and implausible items. The ability to perceive plausible expertise categories therefore appears to be quite fragile in young children with dramatic improvements in the ability between kindergarten and 4^th grade.

There were, however, several limitations to this preliminary study. Some implausible items may have been rejected because they seemed unlikely as opposed to causally empty (e.g., some children felt that it was implausible for clouds to be shaped like teddy bears). Also, the youngest children might have performed at a higher level with more simplified instructions and with materials that reduced unnecessary processing loads. Finally, although a cross-domain difference was found, with only two items per domain, that pattern was only suggestive.

3. Study 1: Domains and Plausible Areas of Expertise

The preliminary study suggested that the youngest school children have only a faint sense of how to tell plausible expertise domains apart from implausible ones in terms of the density of underlying causal structure. In addition, it suggested that the ability to make such a distinction might first appear for artifacts. Given that kindergartners were able to perform at above chance levels for artifacts, the question arises as to whether more simplified stimuli and procedures might make it possible for higher levels of success in younger children for other categories as well. To correct these limitations, Study 1 screened (through consensus judgments of three adults) all items for plausibility such that the existence of members of a category would not be judged implausible or silly even if the category was not legitimate for expertise. Richer pictorial diagrams were used to make sure that children understood the situations being described. Finally, Study 1 focused on just two domains, with new items and more instances in the artifact and living kinds categories, so as to better assess the generality and robustness of the domain difference found in the preliminary study.

3.2. Results

Because there were no significant effects of order or gender, all data was collapsed across these categories for all age groups. Average scores were then computed for the plausible and implausible scores for living kinds and artifacts, creating 4 summary scores for each child. The mean overall scores and standard deviations are shown in Table 2.

Table 2

Study 1 means (on a 4 point scale) and standard deviations (in parentheses).

	Living Kinds		Artifacts
Grade	Plausible	Implausible	Plausible	Implausible
K	2.49 (0.61)	2.48 (0.65)	2.77 (0.68)	2.08 (0.64)
2	3.12 (0.64)	1.66 (0.43)	2.68 (0.45)	1.51 (0.58)
4	3.37 (0.43)	1.52 (0.45)	3.14 (0.38)	1.35 (0.40)
A	3.56 (0.61)	1.36 (0.47)	3.60 (0.30)	1.31 (0.54)

Open in a separate window

A 4 × 4 repeated measures ANOVA was used to analyze the likelihood judgments of categories, with item type (plausible living kind, implausible living kind, plausible artifact, implausible artifact) as the within-subject factor and age (K, 2, 4, Adults) as the between-subject factor. Effect size estimates were computed using partial eta squared (η²). The analyses indicated a main effect of item type, F(3, 98) = 310.38, p < .001, η² = .76, primarily because participants responded differently as a function of whether items were plausible or implausible. There was also a main effect of grade, F(3, 98)=3.44, p < .02, η² = .10, reflecting a somewhat larger drop with age of implausible judgments relative to plausible ones. There was also a significant item type by grade interaction, F(9, 98) = 27.91, p < .001, η² = .46, indicating a greater contrast between the two implausible and the two plausible categories in older children and adults.

The pattern of developmental change is more easily visualized when difference scores are calculated by subtracting the implausible scores from the plausible scores for each category type (e.g., the plausible dog item minus the implausible dog item). To determine if there were contrasts between the broad categories of living kinds and artifacts, new mean difference scores were created for each of these two aggregate categories and are shown in Figure 2.

Open in a separate window

Fig. 2

Ratings of difference scores for likelihood of experts for different domains in Study 1 with standard errors shown. Younger children sensed the plausible vs. implausible domain difference earlier for the aggregate artifact category than living kind category. There was a dramatic rise from Kindergarten to Second Grade in the ability to sense appropriate expert categories for living kinds.

A 2 × 4 repeated measures ANOVA was used to analyze the likelihood judgment difference scores, with aggregate category type (living kinds, artifacts) as the within-subject factor and age (K, 2, 4, Adults) as the between-subject factor. The analyses found no main effect of category type, F(1,98) = .60, n.s., η² = .006. There was a main effect of grade, F(3, 98) = 36.54, p < .001, η² = .61, reflecting larger difference scores in older children and adults. There was also a category type by grade interaction, F(3, 98) = 7.78, p < .001, η² = .19. In addition, t tests with Bonferroni corrections for multiple tests showed that the difference scores for each age were significantly different from those for all other ages (all at p < .02).

When difference scores were compared through t-tests to a score of 0 (chance level), all scores but one were significantly different from 0 at the p < .01 level (with Bonferroni corrections for multiple t-tests). The one non-significant case was the living kind difference score for kindergartners. This pattern held for every individual item as well. Thus, separate t-tests revealed that each of the four artifact pair differences was significantly different from a population mean of 0 in the kindergarten age group (all p's < .04), while none of the of living kind pair differences was significantly different from 0 in the kindergarten age group. All items were significantly different from 0 in the 2^nd and 4^th graders and adults (all p's < .01). Similarly, all plausible scores showed significant rises across ages and all implausible scores showed significant drops across ages, as revealed by separate one way ANOVA's (all p's < .001).

4. Study 2: Reducing the Need to Understand Expertise

The preliminary study and Study 1 document a strong developmental shift between kindergarten and second grade in the ability to tell plausible expertise domains apart from implausible ones in terms of the density of underlying causal structure. In addition, they both pointed towards the ability to make such a distinction first with artifacts. Yet both studies have levels of complexity that may have created special difficulties for younger children and which may be unrelated to the ability to see implausible and causally empty domains, per se. In particular, both involved using a rating scale, which while certainly feasible with young children and even preschoolers (Mills & Keil, 2005), could pose additional cognitive loads. In addition, although younger children can be taught about experts (Lutz & Keil, 2002), it might reduce cognitive load to not have a training sequence on experts. To that end the procedure was changed in Study 2, but the same stimulus items as in Study 1 were used.

4.2. Results

Each choice was scored as either a 0 (picking the implausible domain person) or a 1 (picking the plausible domain person). These scores were then added together and divided by the number of items in each category (4 artifact pairs and 6 living kind pairs consisting of 4 animals and 2 plants). A 2 × 4 repeated measures ANOVA was used to analyze the likelihood judgments of categories, with category type as the within-subject factor and age (K, 2, 4, Adults) as the between-subject factor. Effect size estimates were computed using partial eta squared (η²). There was a main effect of category type (F(1, 102) = 4.11, p < .05, η² = .039). The grade by category type interaction was not significant (F(3, 102) = 1.45, η² = .041). There was a main effect for grade (F(3, 102) = 6.70, p < .001, η² = .165).

As seen in Fig. 3, participants tended to do better on the artifacts, especially the kindergarteners. Two tailed t-tests against a mean of .50 revealed that kindergartners were above chance levels for artifacts (t = 3.39, p < .003, df = 24) and for living kinds (t = 2.32, p < .03, df = 24). If a Bonferroni correction is made for multiple t-tests, the kindergartners remain well above chance levels on the artifacts and are not significantly above chance levels for living kinds.

Open in a separate window

Fig. 3

Results from Study 2 in which children were asked to pick which of two people would be more useful expert on the same category. Shown here are the average scores for all artifacts and for all living kinds (on a scale ranging from .5 (representing picking at random (50%) between the plausible and implausible domains) to a 1.0 (representing 100% choice of the plausible domain)), with standard errors shown. As in the preliminary study and Study 1, kindergarteners were more able to discern the plausible domains for artifacts than for living kinds.

5. Study 3: A Different Sense of Appropriate Experts

Study 1 and the preliminary study consistently demonstrate that kindergarteners have considerable difficulty evaluating the plausibility of an expertise domain on the basis of causal density and structure. Although other studies do document an ability of even preschoolers to reason about experts, the possibility remains that younger children might generally have a much harder time thinking about the plausibility of someone being an expert and that a problem with thinking about expert plausibility is causing drops in performance rather than a problem in seeing causal density and structure. Study 2 addressed this issue by using a task that did not explicitly ask about who was the expert. Another manipulation check would be to use a different kind of task that did ask about plausible experts and on which kindergarteners were able to do quite well. Study 3 therefore used a different criterion from causal density, namely causal relevance, to see if younger children might be able to choose between plausible and implausible experts.

Prior work has shown that kindergartners are clearly sensitive to a dimension of causal structure known as “causal relevance,” namely knowing that certain types of properties tend to be causally important in specific domains (Keil, Smith, Simons & Levin, 1998), a phenomenon that is well documented in adults (Medin & Shoben, 1988) and even in other primates, who weigh different property types as more central for tools vs. foods (Santos, Hauser, & Spelke, 2002). Adults and children alike tend to see color, surface markings and precise number of parts as less causally relevant to understanding artifacts than shape, size, and fragility. For living kinds, these properties tend to have more equal levels of importance and often color, surface markings, and precise number of parts can outweigh overall shape, size, and fragility. Thus, a hot pink, striped hammer with a two-piece handle and a head is still judged a hammer if its overall shape, strength, and size are within certain bounds. However, a dog-like mammal that is naturally hot pink, striped, and naturally has eight legs, is a much more problematic dog. Causal relevance intuitions seem more basic than intuitions about the causal density and structure. One might not have much of a sense at all of the causal complexity that underlies the basis of being a particular species while still believing that color, number of parts, and surface patterns tend to matter more for biological species than for tools.

Study 3 capitalized on this ability to make causal relevancy judgments by asking children which of two alleged experts were more plausible as a function of domains. If even kindergartners were able to make contrasting judgments about expertise for living kinds and artifacts by considering the kinds of properties an alleged expert emphasized, then we can see that the difficulties demonstrated by kindergarteners in the preliminary study and Study 1 are not simply reflecting problems thinking about experts. As reviewed earlier, even preschool children do seem to be able to quickly sense, even in novel artificial systems, what classes or types of interventions are likely to be linked to what properties. That ability may make them effective here in the ability to evaluate experts.

5.2. Results

Each choice was scored as either a 0 (picking the person who emphasized domain inconsistent properties) or a 1 (picking the person who emphasized domain consistent properties). These scores were then added together and divided by the number of items in each category (4 tool pairs and 4 animal pairs). A 2 × 3 repeated measures ANOVA was used to analyze which person was chosen, with category type as the within-subject factor and age (K, 2, 4) as the between-subject factor. Effect size estimates were computed using partial eta squared. There was a main effect of category type (F(1, 88) = 30.318, p < .001, η² = .256). The grade by category type interaction was not significant (F(2,91) = 1.01, η² = .022). There was a main effect of grade (F(2,88) = 7.27, p < .001, η² = .142).

As seen in Fig. 4, even kindergarteners showed different response patterns for tools and animals. A t-test of just the kindergarten age group showed that their responses for tools and animals were significantly different (t = 2.38, p < .025, df = 29). The other two age groups showed more dramatic differences significant at the p < .001 level.

Open in a separate window

Fig. 4

Judgments of quality of experts as a function of property type by domain interactions, with standard errors shown. A value of 0 favors people who talked about color, surface markings and precise number of parts. A value of 4 favors experts who talked about shape, size, and fragility.

6. General Discussion

If the feasibility of folk science rests on the extent to which individuals have fully detailed mechanistic models of the world around them, there is not much hope. If, however, folk science is more like the formal sciences in the extent to which individuals powerfully leverage their understandings by developing ways of relying on knowledge in other minds, then it may well be feasible. This is not to diminish the powerful cognitive processes and structures in the mind of each individual doing folk science. It is just that those processes and structures are quite different from intricate fully articulated theories. Instead, there appears to be a diverse set of ways in which we build on the crude mechanistic fragments in our heads. Moreover, these ways of going beyond such fragments appear early in development, with some being present in kindergartners and others rapidly emerging by the second grade. Given the rapid rise in the ability to detect where causal complexity lies, it could be argued that folk science is much more plausible in second graders than it is in kindergarteners, although a firm conclusion along those lines must await further studies asking if other task variations might improve performance in younger children.

To be able to augment our imperfect mechanistic understandings, we need several cognitive components. First, we must sense causal and relational patterns in the world in concise, abstract ways that allow us to know how to link various phenomena to appropriate groups of experts or domains of expertise. This process would normally work at a more fine-grained level than the very weak constraints provided by “framework theories”, presumably by looking at patterns that are particularly diagnostic for particular domains. Second, we must have ways of evaluating the causal intricacy of phenomena in a domain without being swamped by the details. Third, we must have some heuristics for evaluating alleged experts, some of which rely on hunches about what kinds of properties and relations matter in a specific domain, and some of which rely on past behaviors or inferred motives of an expert (Harris, 2007; Mills & Keil, 2005, 2008). Finally, we must have a sense of the knowledge formation process itself so as to know what kinds of knowledge are probably learnable on one's own and what kinds almost surely require testimony by others given a particular environment. This wide range of capacities may not be an elaborate product of higher education, but instead may be manifested in ways in which all of us make sense of our world from an early age.

The studies presented here offer a glimpse into one part of the overall system that supports folk science: the ability to sense causally rich and fertile domains for a science vs. causally sterile ones. It is ironic, in light of older theories of children being hopelessly concrete, that the concrete details are often where children most need to defer to others. Most adults freely use words to refer with confidence to things that they cannot identify directly (such as the distinctions between beeches vs. elms, silver vs. platinum, and ferrets vs. weasels) because they have a sense of who does know about these things and of the chain that could connect them to that person. Similarly, young children seem to have a sense of how to find supporting details. This sense is often built on knowledge of more abstract patterns that are unique to specific domains and that can serve as ways of linking a phenomenon to an appropriate group of experts. Importantly, children who sense that “dogs that hunt” is a much more plausible category than “dogs with red collars” might nonetheless be unable to give any specific details about hunting dogs. Much of this competency also seems implicit in nature. Children often had compelling intuitions about plausible areas of expertise or plausible kinds of expert explanations, but then could provide almost no information about why. They sense relatively abstract concise patterns for domains in ways that allow them to make such judgments while knowing hardly any details.

What role do metacognitive skills play in the development of the ability to defer appropriately? It is well established that children's metacognitive understanding develops considerably in the school years (Kuhn, 2000; Schneider, 2008). Children become better at understanding the limits of their own cognitive processes and knowledge bases and at using strategies to cope with their limitations (Schneider, 2008). Their epistemelogical understanding also develops as they show marked increases in the extent to which they can appreciate that different people can hold equally valid but different ways of understanding a set of phenomena and that errors can be an intrinsic part of a science (Kuhn, 2008; Masnick & Klahr, 2003). They slowly come to appreciate that people can be biased in their interpretations without being aware of their bias (Mills & Keil, 2005, 2008; Pillow & Henrichen, 1996). Although many of these skills do show strong developmental changes during the same period when children come to see which areas of expertise are more implausible, the influence of metacognitive factors on these tasks is unclear. Metacognitive factors may be more involved in influencing the ability to evaluate one's own degree of understanding and when one needs to defer. In contrast, the main developmental change found in Studies 1–3 may be a gradually increasing appreciation of the causal complexity underlying some categories, especially natural kinds. Thus, decisions on when to defer will be more closely linked to metacognitive abilities than will decisions on whom to defer to.

Several developmental puzzles remain to be explained. One question concerns the role of the detection of local anomalies, inconsistencies, and causal ambiguities. We have seen that even adults tolerate considerable amounts of contradictions in larger belief systems that seem to be a somewhat motley collection of conceptual fragments with lots of theoretical holes and gaps. Yet inconsistencies and contradictions are also often said to be “the engines the drive conceptual change” in young children (Carey, 2009). These may primarily occur as local juxtapositions, perhaps often only transiently during a serendipitous coming together of two clashing ideas in explicit thought. Conceptual change may then be triggered in the more abstract schematic skeletal framework as well, perhaps through a heuristic that if a part has a contradiction or if some threshold of several contradictions is exceeded, there may be a problem with the framework as a whole.

What are the broader implications for folk science? In actual practice folk science often involves a “gluing” together of fragments as a local situation suggests, buttressed by notions of relevant supporting experts. For example, if adults are asked to explain how the body turns food into movement, they might have never before tried to come up with such an explanation and may have very little in terms of a preconceived systematic idea beyond the very broad constraints laid down by a “framework theory.” But as the question is posed, they start to assemble various relevant fragments (e.g. chewing, waste, things they have heard about nutrition, some ideas about muscles, etc.) and try to piece them together. That process, and the beliefs that drive it, are powerfully helped by hunches about where they are on thin ice vs. where there is a strong foundation of supporting knowledge, even if they do not know it themselves. This creation of explanations “on the fly” is now receiving more attention in the modeling approaches as well (e.g., Klenk, Friedman, & Forbus, 2008).

How do all the elements that make up a folk science fit together into a system that has some efficacy for its users? The suggestion here is that the array of elements may be far larger than it might appear at first and that such an array is needed to support the highly impoverished nature of local specific intuitive “theories.” Children and adults alike seem to have very broad and extremely schematic orientations towards domains and a sense of causal patternings that both suggest where expertise is needed and point a person towards the likely expert cluster. They also have abilities to evaluate sources of information as well as to reflect on their own ways of acquiring knowledge. Given the presence of all these abilities in young children and given how similar they are to many of the key features that make formal science succeed, it seems likely that such abilities are a foundational part of what makes folk science feasible.

Folk science in technological cultures may also carry within it notions of formal science that are absent in more traditional cultures. Thus, the average layperson in North America or Europe may have well-entrenched ideas about the discipline known as `science' itself. This may not make a difference in terms of how chains of deference are set up, but it is certainly a topic of interest for future studies. Perhaps in traditional cultures there are notions about legitimate areas of expertise and about experts as well, but those areas of expertise are organized in ways that reflect that culture's specific epistemological orientations.

We also know that individual scientists constantly make simplifying assumptions so as to prune back a causal briar patch into something more elegant and cognitively manageable. They seem to do so in several ways. First, they tend to stick to one level of a reductionist hierarchy, not diving down deeper to lower levels (Owens, 1989; Wilson & Keil, 1998). For example, some psychologists may try to envision how cognitive structures and processes are constrained by the biology of the nervous system, but they rarely if ever try to fully specify a mechanistic model of what is happening at the cellular level. Similarly, while attending to chemistry, the biologist also stops short of the sorts of details that occupy chemists. Thus, few biologists ask how quantum bond angles are relevant to an aspect of cellular metabolism. There is a skill here that is not well understood, partly because we don't have a reliable way of specifying levels of reduction and whether there really are clearly objective levels that apply in the same way across all the sub-specialities of such major areas as psychology, biology, chemistry, and physics. Somehow, scientists in practice draw these lines all the time so as to make their task more manageable and they seem to do so in a way that allows their science to advance. Folk science must certainly also engage in a somewhat similar process, perhaps acknowledging the possibility of cross level influences, but trying to focus on one level at a time. Indeed, when two distinct levels are present, as in some views of disease causation in both biological and supernatural terms, children and adults alike often have the two levels work at different time scales so as not mix them together (Legare & Gelman, 2008).

Scientists also make a science “local” by ruling out those tendrils of causal influence that, while technically present, can be ignored for the purposes of the scientific task at hand. Without such simplifications, one runs the risk of having to consider the ways in which almost any event can potentially have a causal influence on any other that follows later: a scientific version of the “butterfly effect” problem (Hilborn, 2004). Elga (2007) considers this problem in detail and suggests that scientists, and lay people, see the world as localized nets of causal relations. Those nets form clumps and clusters that can be considered as “stand alone” systems in which more remote influences are ruled out. When trying to understand the behavior of billiard balls on a table we tend to consider just the forces at play between objects on the table, even though there are very small but real influences exerted by the moon, people in the room, and countless other objects and events such as sound waves. We may technically rule out effects that would have to travel faster than the speed of light to have an effect (things outside what physicists call the “light cone;” Elga, 2007); but there is still a vast array of other real influences that are simply below some threshold in almost all matters of scientific inquiry. Folk scientists engage in such local analyses with a vengeance, but the heuristics they use to prune away causal links to other domains have not been studied. Even in the heights of analytical science, the processes through which scientists zero in on certain causal processes and relations as most relevant is described as more of an art than an analytically driven procedure (Strevens, 2009).

A second facet of deferring involves heuristics for establishing trust in others. Two commonly used heuristics rely on prior accuracy and familiarity with an informant (Corriveau & Harris, 2009). Quite young children seem to be able to employ both of these heuristics, suggesting that both are also central parts of folk science. Thus, children are capable of judging that a person who is mistaken is likely to be mistaken again in the future, and they also use various cues of affiliation to judge testimony, such as whether they are familiar with a person or whether a potential informant tends not to dissent in general (Corriveau, Fusaro, & Harris, 2009). A tendency to discount dissenters may not always be the best strategy, but it is an easy heuristic to use.

Any account that discusses the social dimensions of scientific understanding must also at least address the vast literature on the social construction of science, the idea that cultures and groups within cultures can create lenses that distort how phenomena are viewed and understood. In the extreme, such approaches argue that there is no scientific “truth” and that all is socially constructed, often to serve political and economic agendas. The approach taken here instead assumes (following Boghassian, 2006 and Kitcher, 1998, 2001) that, while complete fabrications and massive distortions are possible, most of science does not work that way. Scientific knowledge, of either the formal or folk variety, is not simply a relative matter. Science refers to real patterns in the world, usually of a causal nature, and is thereby constrained by them.

Kitcher (2001) suggests that the process is much like the making of maps for an area. Depending on one's goals and perspective, one might come up with a very different map of the same environment than another person would. Kitcher describes how the subway map of London distorts real distances and spatial layout in dramatic ways, but is still a “true” map with respect to representing the ordered relations between stops and intersections in an efficient and clear way. A different map for walking routes might have a radically different appearance and set of spatial relations but would also be true. Sciences can each construct their own maps that leave out some informational dimensions and distort others; but they are rarely complete social constructions. In the same manner, we might consider folk science as constructing maps that are driven by real-world causal and relational patterns even as they simplify and distort information in various ways. Moreover, different cultural groups can construct very different maps that, while all true, result in sharply contrasting reasoning patterns (Atran & Medin, 2008).

A full account of folk science requires at least the following. It must specify the kinds of relevant knowledge in the minds of individuals. In particular, this means a specification of typical knowledge fragments and the levels of abstraction at which they are usually represented. It also must specify the ways in which people use that partial knowledge in situations when they are trying to explain a phenomenon, including the ways they stitch together fragments on the fly to create a more coherent description, guided by the constraints imposed a framework theory. A comprehensive account also requires a specification of the ways in which people evaluate their own knowledge. However, beyond these descriptions of what each person knows about how the world works and their understanding of their degrees of ignorance, another critical dimension is a specification of how individuals support their own impoverished knowledge by learning to outsource their understandings to others.

Just as users of a natural language confidently talk about natural kinds such as platinum and silver, even though they may have no individual means of telling them apart (Putnam, 1975; Malt, 1990), so also do our folk science explanations rely extensively on our senses of legitimate areas of expertise and how to defer to those areas. If even young children are well on their way to navigating such areas of expertise, that ability is one major form of support for the feasibility of folk science. Thus, if one has a sense of when one's own shaky understanding is likely to be resting on firm ground in other minds, one can then use that understanding more confidently to make other inferences. For example, even if I cannot tell silver from platinum, if I am confident that through various chains of deference experts would affirm that something was silver, I can feel more entitled to make other inferences about that thing by virtue of its being silver. The same sort of process can work in folk science where for example, even if we do not know much about the molecular details that are responsible for the surface properties of a particular plant, we can make other inferences about that plant based on the secure knowledge there are experts who could explain such a causal pathway.

In summary, the finding that laypeople of all ages have extremely limited understandings at the level of mechanistic clockworks diagrams of how the world works does not mean that they are not deeply interested in causal patterns or in seeking out explanations, activities that are very much a part of folk science. Lay people are in fact immensely sophisticated at tracking certain kinds of abstract causal patterns and using that information to figure out how to better ground their incomplete understandings; and this is an ability so basic that its roots can be found before children start schooling and as surging in influence by the time they reach the second grade. Folk science may be an intrinsically social process, but one that starts off that way even before children have assimilated most of the practices of their culture.

We may attribute much of the social infrastructure of formal science to explicitly taught social institutions that are acquired quite late, when in fact basic features of deference and expertise seeking are natural ways that children approach the process of understanding. Just as some anthropologists avoid the study of children because of a misleading assumption that all the interesting aspects of cultures are only gradually learned from long-term immersion in those cultures (Hirschfeld, 2002), it may be equally misleading to assume that all the interesting facets of the social embedding of both the formal and folk sciences are learned through long-term assimilation of cultural practices. Some of that embedding may be part of how children approach their very first explorations of how the world works.

Acknowledgments

Appendix A. Training and Stimuli for Study 3

References