We face a challenge when inferring what others know. Actions do not transparently reveal epistemic states: ignorantagents routinely ignore information too costly to obtain, and knowledgeable agents often confirm what they already knowwhen its convenient. We hypothesized that epistemic inferences are sensitive both to agents actions, and the underlyingutilities that best explain them. We tested this possibility in a simple task. Adults watched an explorer decide whetherto collect a map before searching an island for treasure. Participants (n=40) were asked to jointly infer how much theexplorer knew about the treasures location, and how much information the explorer believed the map had. Participantjudgments matched a computational model of epistemic inference structured around an expectation that agents rationallytradeoff information gain with information cost (r=0.86; 95%: 0.740.93, p¡.001). Our results suggest that adult Theory ofMind supports nuanced and graded epistemic inferences from observable action.

To infer what others know, we must consider under what epistemic states their actions were both rational and probable. We test whether preschoolers can compare the probability of different actions (and outcomes) under different epistemic states—and use this to evaluate what others know. Specifically, four- to six-year-olds (n=90) were asked to help evaluate an agent’s knowledge state by asking the agent to complete either an “undiagnostic” task (where success was assured), or a “diagnostic task” (where the probability of random success was low). By age six, children understood that the “diagnostic” task would more likely reveal the agent’s knowledge state; four- and five-year-olds had no reliable preference, although children in all age groups understood that the “diagnostic” task was harder. These results suggest that, by the end of preschool, children understand how agents’ epistemic states and environment jointly determine success—considering whether agents’ actions imply knowledge, or just luck.

When deciding whether to explore, people must consider both their need for information, and the cost of obtaining it.Thus, to judge why someone explores (or decides not to), we must consider not only their actions, but also the cost ofinformation. Do children attend to the cost of agents exploratory choices when inferring what others know or desireto know? In Experiment 1, four- and five-year-olds judged that an agent who rejected an opportunity to gain low-costinformation must have already known it. In Experiment 2, four- and five-year-olds judged that an agent who incurreda greater cost to gain information had a greater epistemic desire. In two control experiments, we show that these resultscannot be explained by a low-level heuristic linking competence with knowledge. Our results suggest that childrens Theoryof Mind includes expectations about how costs interact with epistemic desires to produce action.

Much research has investigated how children track and reason about accuracy when deciding who to trust. The majority of this work assumes a static link between accuracy and knowledge; that is, children are expected to attribute greater knowledge to accurate agents. But while accuracy often reveals knowledge, the two are not deterministically related. Ignorant agents can be accurate (for example, one could take a lucky guess), and knowledgeable agents can be inaccurate (for example, one could accidentally err). Given this, how do children reason about the relation between knowledge and accuracy? Across three experiments, we show that four- and five-year-olds are sensitive to the distinction between knowledge and accuracy. Specifically, children judge that an agent who accurately predicts an outcome is knowledgeable, but an agent who merely observes and then accurately describes the same outcome is not. Our findings show that when children gauge agents’ knowledge, they do not rely on accuracy alone; they infer knowledge only when an agent is right in the right kind of way

Fairness depends on the principles that people use to justifytheir actions, and on the outcomes that they produce. Here wepropose that, from early in childhood, we also judge fairnessbased on whether we believe the resulting outcomes werecaused by the underlying principles. In Experiment 1 we showthat four- five- and six-year-olds believe that an agent who paidattention when distributing resources is more fair than an agentwho was distracted when distributing resources, even whenthey both produce identical outcomes. In Experiment 2 weshow that children of the same ages believe that an agent whocounts when distributing resources is more fair than an agentwho does not count, even when both agents attend to how theydistribute their resources and produce identical outcomes.Together, our findings suggest that children do not judgefairness based on the outcome alone, and they add to a growingbody of work suggesting that, from early childhood, ourintuitions about fairness are tightly linked with intuitions aboutexactness.

Humans often communicate through seemingly arbitraryactions, like winks, waves, and nods. While these non-iconicgestures derive their meanings from cultural consensus,people, and especially children, must be able to identify thesemovements as gestures. Here we propose that people expectthat communicative actions will be shaped to reveal that theyhave no external goal. In Experiment 1, we show that peoplejudge inefficient actions as more likely to be communicative.In Experiment 2, we show that these judgments are trulydriven by efficiency, rather than a movement’s visualcomplexity. Finally, in Experiment 3, we show that repetition– which unambiguously reveals that the goal of the action isthe movement itself – has a strong influence on inferencesabout communicativeness, independent of the motion’sefficiency. Our findings show how expectations about non-iconic communicative actions can be folded into a generalgoal inference framework structured around an expectationfor efficiency.

Inferences about other people's knowledge and beliefs are central to social interaction. In many situations, however, it's not possible to be sure what other people know because their behavior is consistent with a range of potential epistemic states. Nonetheless, this behavior can give us coarse intuitions about how much someone might know, even if we cannot pinpoint the exact nature of this knowledge. We present a computational model of this kind of broad epistemic-state inference, centered on the expectation that agents maximize epistemic utilities. We evaluate our model in a graded inference task where people had to infer how much an agent knew based on the actions they chose. Critically, the agent's behavior was always under-determined, but nonetheless contained information about how much knowledge they possessed. Our model captures nuanced patterns in participant judgments, revealing a quantitative capacity to infer amorphous knowledge from minimal behavioral evidence.

When deciding how to act in new situations, we expect agents todraw on relevant prior experiences. This expectation underliesmany of our mental-state inferences, allowing us to infer agents’prior knowledge from their current actions. Do children sharethis expectation, and use it to infer others’ epistemic states? InExperiment 1, we find that five- and six-year-olds (but not four-year-olds) attribute additional knowledge to agents whose priorexperiences cannot explain their success. In Experiment 2, wefind that six-year-olds (but not younger children) also attributegreater knowledge to agents whose prior experience cannotexplain their failure. We show that by age five or six, childrenexpect ignorant agents’ beliefs (and therefore their actions) to beguided by their prior knowledge. This work adds to a growingbody of research suggesting that, while infants can representmental states, the ability to infer mental states continues todevelop throughout early childhood.

As we navigate our information-rich world, we frequently interpret and integrate testimony from external sources (friends, teachers, books, internet articles, etc.) – deciding which pieces of information to believe, and which to discard. One cue to a statement’s trustworthiness is whether it comes from a consensus (i.e., when a majority of people agree). But what counts as consensus? When presented with a set of agreeing sources, do we evaluate the quality of consensus – for example, asking whether each source arrived at their conclusion by independent means? In a first experiment, we demonstrate that individuals are insensitive to the quality of a consensus, and are equally confident in conclusions drawn from a ‘true’ consensus (i.e., one derived from many primary sources) and those drawn from a ‘false’ consensus (i.e., one derived from many secondary sources but only a single primary source). In a second experiment, we find that this continues to be true even when the expertise of the secondary sources is minimized. Together, our experiments provide converging evidence that people do not properly discount (or discount at all) information from a ‘false’ consensus.

Although we possess intuitions about pedagogy from early in life, adults commonly fail to teach effectively in real-world situations. Why might adults struggle in more complex teaching tasks? Here we develop a simple teaching task where adults fail to teach naïve learners, despite reporting high confidence that they taught effectively. Using a formal model of a rational teacher, we analyze the sources of our adult teachers’ failures. Our model-based analyses reveal that teachers successfully provided high-quality examples, but failed to address hypotheses that naïve learners find plausible. We validate these results in a second experiment, where we find that constraining learners’ hypothesis space increases their performance in the task. Our findings help bridge the gap between children’s teaching proficiency in constrained tasks, and adults’ teaching failures in more naturalistic tasks.