Action understanding as inverse planning

Humans are adept at inferring the mental states underlying other agents’ actions, such as goals, beliefs, desires, emotions and other thoughts. We propose a computational framework based on Bayesian inverse planning for modeling human action understanding. The framework represents an intuitive theory of intentional agents’ behavior based on the principle of rationality: the expectation that agents will plan approximately rationally to achieve their goals, given their beliefs about the world. The mental states that caused an agent’s behavior are inferred by inverting this model of rational planning using Bayesian inference, integrating the likelihood of the observed actions with the prior over mental states. This approach formalizes in precise probabilistic terms the essence of previous qualitative approaches to action understanding based on an “intentional stance” [Dennett, D. C. (1987). The intentional stance. Cambridge, MA: MIT Press] or a “teleological stance” [Gergely, G., Nádasdy, Z., Csibra, G., & Biró, S. (1995). Taking the intentional stance at 12 months of age. Cognition, 56, 165-193]. In three psychophysical experiments using animated stimuli of agents moving in simple mazes, we assess how well different inverse planning models based on different goal priors can predict human goal inferences. The results provide quantitative evidence for an approximately rational inference mechanism in human goal inference within our simplified stimulus paradigm, and for the flexible nature of goal representations that human observers can adopt. We discuss the implications of our experimental results for human action understanding in real-world contexts, and suggest how our framework might be extended to capture other kinds of mental state inferences, such as inferences about beliefs, or inferring whether an entity is an intentional agent.     

Spontaneous goal versus spontaneous trait inferences: How ideology shapes attributions and explanations

Past research documented liberals’ greater tendency than conservatives to take situational determinants of others’ actions into account when forming causal attributions, and conservatives’ greater tendency to seek consistency. We hypothesize that liberals (vs. conservatives) should be more likely to make spontaneous goal inferences (SGIs). Conservatives, however, should tend to implicitly infer invariant rather than variant characteristics from others’ behaviors, drawing spontaneous trait inferences (STIs) rather than SGIs. Experiment 1 and 2 supported those hypotheses by illustrating differences in the type of implicit inferences formed by liberals and conservatives in a false recognition paradigm common to the STI literature. Experiment 3 revealed similar differences in conservatives’ and liberals’ goal and trait inferences when making open-ended causal explanations for others’ actions.     

One‐year‐old infants use teleological representations of actions productively

Two experiments investigated whether infants represent goal‐directed actions of others in a way that allows them to draw inferences to unobserved states of affairs (such as unseen goal states or occluded obstacles). We measured looking times to assess violation of infants' expectations upon perceiving either a change in the actions of computer‐animated figures or in the context of such actions. The first experiment tested whether infants would attribute a goal to an action that they had not seen completed. The second experiment tested whether infants would infer from an observed action the presence of an occluded object that functions as an obstacle. The looking time patterns of 12‐month‐olds indicated that they were able to make both types of inferences, while 9‐month‐olds failed in both tasks. These results demonstrate that, by the end of the first year of life, infants use the principle of rational action not only for the interpretation and prediction of goal‐directed actions, but also for making productive inferences about unseen aspects of their context. We discuss the underlying mechanisms that may be involved in the developmental change from 9 to 12 months of age in the ability to infer hypothetical (unseen) states of affairs in teleological action representations.     

Processing inferences drawn from the logically equivalent frames half full and half empty

Choice-based experiments indicate that readers draw sophisticated inferences from logically equivalent frames. Readers may infer that a glass was previously full if described as currently half empty, and previously empty if described as currently half full. The information leakage framework suggests these inferences are made because information about a previous state is leaked from speaker's choice of frame. We examine if similar inferences are made during reading in two eye-tracking experiments. In Experiment 1, participants read a passage where a character describes a glass as currently half full or half empty before making a statement about the previous volume. We hypothesised that participants would infer that the glass was previously empty or full, respectively. Results suggest processing a previous volume of full is simpler regardless of the frame provided. In Experiment 2, materials were constructed to ensure inferences were based on participants' beliefs as opposed to characters'. Results support the information leakage framework; previous volumes of full and empty were processed more easily after current volumes of half empty and half full, respectively. We suggest that processing discrepancies between the two experiments are driven by word-related factors (e.g., markedness) or by participants' integration of characters' expectations.     

Intersubjectivity and Receptive Experience

Wilfrid Sellars's iconic exposé of the ‘myth of the given’ taught us that experience must present the world to us as normatively laden, in the sense that the contents of experience must license inferences, rule out and justify various beliefs, and rationalize actions. Somehow our beliefs must be governed by the objects as they present themselves to us. Often this requirement is cashed out using language that attributes agent‐like properties to objects: we are described as ‘accountable to’ objects, while objects ‘hold us’ to standards, and so forth. But such language is either deeply anti‐naturalistic or trades on a set of metaphors in need of a literal translation. We offer an explanation of how the material features of the world, as received in experience, can rationally constrain our beliefs and practices—one that makes no recourse to this imagery. In particular, we examine the structure of ostensive practices (that is, practices of directing one another's attention to objects and features of the world) and the distinctive role they play in making us jointly beholden to how things actually are.     

Understanding the glue of narrative structure: Children's on- and off-line inferences about characters’ goals

Because characters’ goals play a key role in the structure of narratives, the ability to make inferences about goals is essential to narrative comprehension. Despite their importance, no previous studies have examined the process by which children make these goal inferences. In the current study, we examined 6- and 8-year-old children's goal inference making processes through think-aloud protocols. We also examined the product of comprehension, the mental representation of text, through free recall and comprehension questions. The results revealed that children of both ages regularly made appropriate goal inferences while listening to narratives. In addition, the number of goal inferences predicted children's recall of the stories. Thus, children as young as 6 years old are sensitive to the vital role of characters’ goals in narrative structure, and they can engage in sophisticated cognitive processing while they listen to narratives to form coherent mental representations of them.     

Do 10-month-old infants understand others' false beliefs?

As adults, we know that others’ mental states, such as beliefs, guide their behavior and that these mental states can deviate from reality. Researchers have examined whether young children possess adult-like theory of mind by focusing on their understanding about others’ false beliefs. The present research revealed that 10-month-old infants seemed to interpret a person’s choice of toys based on her true or false beliefs about which toys were present. These results indicate that like adults, even preverbal infants act as if they can consider others’ mental states when making inferences about others’ actions.     

WHAT LIARS CAN TELL US ABOUT THE KNOWLEDGE NORM OF PRACTICAL REASONING

If knowledge is the norm of practical reasoning, then we should be able to alter people's behavior by affecting their knowledge as well as by affecting their beliefs. Thus, as Roy Sorensen (2010 ) suggests, we should expect to find people telling lies that target knowledge rather than just lies that target beliefs. In this paper, however, I argue that Sorensen's discovery of “knowledge‐lies” does not support the claim that knowledge is the norm of practical reasoning. First, I use a Bayesian framework to show that in each of Sorensen's examples, knowledge‐lies alter people's behavior by affecting their beliefs. Second, I show that while we can imagine lies that target knowledge without targeting beliefs, they cannot alter people's behavior. In other words, knowledge‐lies actually work (i.e., manipulate behavior) by targeting beliefs or they do not work at all.     

Spinoza on Counterpossible Inferences

As part of his geometrical method, Spinoza uses inferences involving impossibilities: what would follow if some impossibility were true. This paper examines a puzzle surrounding his use of such ‘counterpossible’ inferences. The puzzle consists of Spinoza's apparent acceptance of the following three claims: that counterpossible inferences can produce knowledge; that inference‐making produces knowledge only on the basis of a transition between ideas; and that counterpossible inferences are unthinkable. I argue for a solution to this puzzle, which interprets Spinoza's geometrical method as a form of syntactical manipulation.     

Faulty Self‐Assessment: Why Evaluating One's Own Competence Is an Intrinsically Difficult Task

People's perception of their competence often diverges from their true level of competence. We argue that people have such erroneous view of their competence because self‐evaluation is an intrinsically difficult task. People live in an information environment that does not contain all the data they need for accurate self‐evaluation. The information environment is insufficient in two ways. First, when making self‐judgments, people lack crucial categories of information necessary to reach accurate evaluations. Second, although people receive feedback over time that could correct faulty self‐assessments, this feedback is often biased, difficult to recognize, or otherwise flawed. Because of the difficulty in making inferences based on such limited and misleading data, it is unreasonable to expect that people will prove accurate in judgments of their skills.     

TEACHING SCIENCE AND RELIGION IN THE TWENTY‐FIRST CENTURY: THE MANY PEDAGOGICAL ROLES OF CHRISTOPHER SOUTHGATE

With the goal of understanding how Christopher Southgate communicates his in‐depth knowledge of both science and theology, we investigated the many roles he assumes as a teacher. We settled upon wide‐ranging topics that all intertwine: (1) his roles as author and coordinating editor of a premier textbook on science and theology, now in its third edition; (2) his oral presentations worldwide, including plenaries, workshops, and short courses; and (3) the team teaching approach itself, which is often needed by others because the knowledge of science and theology do not always reside in the same person. Southgate provides, whenever possible, teaching contexts that involve students in experiential learning, where they actively participate with other students. We conclude that Southgate's ultimate goal is to teach students how to reconcile science and theology in their values and beliefs, so that they can take advantage of both forms of rational thinking in their own personal and professional lives. The co‐authors consider several examples of models that have been successfully used by people in various fields to integrate science and religion.     

When is the gift given? Organ donation,social representations,and an opportunity to register

The gift of life doctrine underpins Australia's approach to organ donation: in legislation, clinical practice, community awareness campaigns, and educational activities. In this paper, we present an approach that situates an understanding of organ donation within a social representation framework as a system of values, ideas, and practices. In cadaveric donation, the final giving‐of‐the‐gift can never be by the donor, leading us to ask where the potential donor's decision to give the gift really lies. We present research from three studies that explored the relationship between what was socially understood about organ donation and the registration of donation intent. Drawing from three socially and culturally diverse populations, we asked people working in a corporate city institution and those attending two football matches in the outer city area to complete a word‐association task and Likert‐scale belief questions about organ donation—followed by an opportunity to register immediately on the Australian Organ Donor Register. Driven by the interdependent themata of life/death and self/other, the gift of life doctrine is inextricably linked with the loss of life emerging as both positive and negative beliefs allowing their relationship to actual registration behaviour to be observed. Our findings suggest that in many instances, the potential donor's genuine desire to give the gift lies in the tension between positive and negative beliefs, manifesting as a consent registration when the positive beliefs about donation prevail and an immediate opportunity to register is available.     

Assessing Mathematics Misunderstandings via Bayesian Inverse Planning

Online educational technologies offer opportunities for providing individualized feedback and detailed profiles of students' skills. Yet many technologies for mathematics education assess students based only on the correctness of either their final answers or responses to individual steps. In contrast, examining the choices students make for how to solve the equation and the ways in which they might answer incorrectly offers the opportunity to obtain a more nuanced perspective of their algebra skills. To automatically make sense of step-by-step solutions, we propose a Bayesian inverse planning model for equation solving that computes an assessment of a learner's skills based on her pattern of errors in individual steps and her choices about what sequence of problem-solving steps to take. Bayesian inverse planning builds on existing machine learning tools to create a generative model relating (mis)-understandings to equation solving choices. Two behavioral experiments demonstrate that the model can interpret people's equation solving and that its assessments are consistent with those of experienced teachers. A third experiment uses this model to tailor guidance for learners based on individual differences in misunderstandings, closing the loop between assessing understanding, and using that assessment within an educational technology. Finally, because the bottleneck in applying inverse planning to a new domain is in creating the model of possible student misunderstandings, we show how to combine inverse planning with an existing production rule model to make inferences about student misunderstandings of fraction arithmetic.     

The dynamics of avoidance goal regulation

An avoidance goal is an undesired state from which a person seeks to distance themselves. Though important for understanding behavior, avoidance goals have received less attention than approach goals. In this paper, we present a dynamic, formal model that provides a framework for describing and predicting the dynamics of avoidance goal regulation. We conduct a series of simulations to examine the dynamic pattern of behavior that emerges from the model when an avoidance goal is pursued in isolation and when an approach goal is also present. Two versions of the model were examined. In the first, the avoidance goal is regulated by a positive feedback loop. In the second, the avoidance goal is regulated by a negative feedback loop. We find that the positive feedback model produces a pattern of runaway behavior, even in a scenario where an approach goal is also present. By contrast, the negative feedback loop model produces a stable pattern of behavior that is more consistent with existing theory. The findings provide an important step toward theoretical parsimony by demonstrating that avoidance goal regulation, like approach goal regulation, can be understood using a negative feedback control system framework. We discuss new insights provided by this model and its potential to spark empirical research.     

A Biased Inferential Naivety learning model for a network of agents

We propose a Biased Inferential Naivety social learning model. In this model, a group of agents tries to determine the true state of the world and make the best possible decisions. The agents have limited computational abilities. They receive noisy private signals about the true state and observe the history of their neighbors' decisions. The proposed model is rooted in the Bayesian method but avoids the complexity of fully Bayesian inference. In our model, the role of knowledge obtained from social observations is separated from the knowledge obtained from private observations. Therefore, the Bayesian inferences on social observations are approximated using inferential naivety assumption, while purely Bayesian inferences are made on private observations. The reduction of herd behavior is another innovation of the proposed model. This advantage is achieved by reducing the effect of social observations on agents' beliefs over time. Therefore, all the agents learn the truth, and the correct consensus is achieved effectively. In this model, using two cognitive biases, there is heterogeneity in agents' behaviors. Therefore, the growth of beliefs and the learning speed can be improved in different situations. Several Monte Carlo simulations confirm the features of the proposed model. The conditions under which the proposed model leads to asymptotic learning are proved.     

Children's understanding of habitual behaviour

Research into the development of Theory of Mind (ToM) has shown how children from a very early age infer other people's goals. However, human behaviour is sometimes driven not by plans to achieve goals, but by habits, which are formed over long periods of reinforcement. Habitual and goal‐directed behaviours are often aligned with one another but can diverge when the optimal behavioural policy changes without being directly reinforced (thus specifically hobbling the habitual learning strategy). Unlike the flexibility of goal‐directed behaviour, rigid habits can cause agents to persist in behaviour that is no longer adaptive. In the current study, all children predict agents will tend to behave consistently with their goals, but between the ages of 5 and 10, children showed an increasing understanding of how habits can cause agents to persistently take suboptimal actions. These findings stand out from the typical way the development of social reasoning is examined, which instead focuses on children's increasing appreciation of how others' beliefs or expectations affect how they will act in service of their goals. The current findings show that children also learn that under certain circumstances, people's actions are suboptimal despite potentially ‘knowing better.’     

A computational-level explanation of the speed of goal inference

The ability to understand the goals that drive another person’s actions is an important social and cognitive skill. This is no trivial task, because any given action may in principle be explained by different possible goals (e.g., one may wave ones arm to hail a cab or to swat a mosquito). To select which goal best explains an observed action is a form of abduction. To explain how people perform such abductive inferences, Baker, Tenenbaum, and Saxe (2007) proposed a computational-level theory that formalizes goal inference as Bayesian inverse planning (BIP). It is known that general Bayesian inference–be it exact or approximate–is computationally intractable (NP-hard). As the time required for computationally intractable computations grows excessively fast when scaled from toy domains to the real world, it seems that such models cannot explain how humans can perform Bayesian inferences quickly in real world situations. In this paper we investigate how the BIP model can nevertheless explain how people are able to make goal inferences quickly. The approach that we propose builds on taking situational constraints explicitly into account in the computational-level model. We present a methodology for identifying situational constraints that render the model tractable. We discuss the implications of our findings and reflect on how the methodology can be applied to alternative models of goal inference and Bayesian models in general.     

EPISTEMIC ACTION,EXTENDED KNOWLEDGE,AND METACOGNITION

How should one attribute epistemic credit to an agent, and hence, knowledge, when cognitive processes include an extensive use of human or mechanical enhancers, informational tools, and devices which allow one to complement or modify one's own cognitive system? The concept of integration of a cognitive system has been used to address this question. For true belief to be creditable to a person's ability, it is claimed, the relevant informational processes must be or become part of the cognitive character of the agent, as a result of a process of enculturation. We argue that this view does not capture the role of sensitivity to epistemic norms in forming true beliefs. An analysis of epistemic actions, basic and extended, is proposed as offering an appropriate framework for crediting an agent with knowledge.     

Instance‐based learning in dynamic decision making

This paper presents a learning theory pertinent to dynamic decision making (DDM) called instancebased learning theory (IBLT). IBLT proposes five learning mechanisms in the context of a decision‐making process: instance‐based knowledge, recognition‐based retrieval, adaptive strategies, necessity‐based choice, and feedback updates. IBLT suggests in DDM people learn with the accumulation and refinement of instances, containing the decision‐making situation, action, and utility of decisions. As decision makers interact with a dynamic task, they recognize a situation according to its similarity to past instances, adapt their judgment strategies from heuristic‐based to instance‐based, and refine the accumulated knowledge according to feedback on the result of their actions. The IBLT's learning mechanisms have been implemented in an ACT‐R cognitive model. Through a series of experiments, this paper shows how the IBLT's learning mechanisms closely approximate the relative trend magnitude and performance of human data. Although the cognitive model is bounded within the context of a dynamic task, the IBLT is a general theory of decision making applicable to other dynamic environments.