A constructivist connectionist model of transitions on false-belief tasks

A Brain-Inspired Model of Theory of Mind

Theory of mind (ToM) is the ability to attribute mental states to oneself and others, and to understand that others have beliefs that are different from one's own. Although functional neuroimaging techniques have been widely used to establish the neural correlates implicated in ToM, the specific mechanisms are still not clear. We make our efforts to integrate and adopt existing biological findings of ToM, bridging the gap through computational modeling, to build a brain-inspired computational model for ToM. We propose a Brain-inspired Model of Theory of Mind (Brain-ToM model), and the model is applied to a humanoid robot to challenge the false belief tasks, two classical tasks designed to understand the mechanisms of ToM from Cognitive Psychology. With this model, the robot can learn to understand object permanence and visual access from self-experience, then uses these learned experience to reason about other's belief. We computationally validated that the self-experience, maturation of correlate brain areas (e.g., calculation capability) and their connections (e.g., inhibitory control) are essential for ToM, and they have shown their influences on the performance of the participant robot in false-belief task. The theoretic modeling and experimental validations indicate that the model is biologically plausible, and computationally feasible as a foundation for robot theory of mind.

A Bayesian framework for the development of belief-desire reasoning: Estimating inhibitory power

A robust empirical finding in theory-of-mind (ToM) reasoning, as measured by standard false-belief tasks, is that children four years old or older succeed whereas three-year-olds typically fail in predicting a person's behavior based on an attributed false belief. Nevertheless, when the child's own belief is undermined by increasing their subjective uncertainty about the truth, as introduced in low-demand false-belief tasks, three-year-olds can better appreciate another person's false belief. Inhibition is believed to play a critical role in such developmental patterns. Within a Bayesian framework, using meta-data, we present the first computational implementation of inhibition, as specified by the Theory of Mind Mechanism (ToMM) model, to account for both the developmental shift from three to four years of age and the change in children's performances between high-demand and low-demand false-belief tasks. A Bayesian framework enables us to evaluate the predictive power of the model and infer the underlying psychological parameters. Together with behavioral evidence, we discuss the critical role of inhibitory control, as specified by ToMM, in children's theory-of-mind development.

False-belief reasoning from 3 to 92 years of age

False-belief reasoning, defined as the ability to reason about another person’s beliefs and appreciate that beliefs can differ from reality, is an important aspect of perspective taking. We tested 266 individuals, at various ages ranging from 3 to 92 years, on a continuous measure of false-belief reasoning (the Sandbox task). All age groups had difficulty suppressing their own knowledge when estimating what a naïve person knew. After controlling for task-specific memory, our results showed similar false-belief reasoning abilities across the preschool years and from older childhood to younger adulthood, followed by a small reduction in this ability from younger to older adulthood. These results highlight the relative similarity in false-belief reasoning abilities at different developmental periods across the lifespan.

A Bayesian Framework for False Belief Reasoning in Children: A Rational Integration of Theory-Theory and Simulation Theory

Two apparently contrasting theories have been proposed to account for the development of children's theory of mind (ToM): theory-theory and simulation theory. We present a Bayesian framework that rationally integrates both theories for false belief reasoning. This framework exploits two internal models for predicting the belief states of others: one of self and one of others. These internal models are responsible for simulation-based and theory-based reasoning, respectively. The framework further takes into account empirical studies of a developmental ToM scale (e.g., Wellman and Liu, 2004): developmental progressions of various mental state understandings leading up to false belief understanding. By representing the internal models and their interactions as a causal Bayesian network, we formalize the model of children's false belief reasoning as probabilistic computations on the Bayesian network. This model probabilistically weighs and combines the two internal models and predicts children's false belief ability as a multiplicative effect of their early-developed abilities to understand the mental concepts of diverse beliefs and knowledge access. Specifically, the model predicts that children's proportion of correct responses on a false belief task can be closely approximated as the product of their proportions correct on the diverse belief and knowledge access tasks. To validate this prediction, we illustrate that our model provides good fits to a variety of ToM scale data for preschool children. We discuss the implications and extensions of our model for a deeper understanding of developmental progressions of children's ToM abilities.

Building machines that learn and think like people

Recent progress in artificial intelligence has renewed interest in building systems that learn and think like people. Many advances have come from using deep neural networks trained end-to-end in tasks such as object recognition, video games, and board games, achieving performance that equals or even beats that of humans in some respects. Despite their biological inspiration and performance achievements, these systems differ from human intelligence in crucial ways. We review progress in cognitive science suggesting that truly human-like learning and thinking machines will have to reach beyond current engineering trends in both what they learn and how they learn it. Specifically, we argue that these machines should (1) build causal models of the world that support explanation and understanding, rather than merely solving pattern recognition problems; (2) ground learning in intuitive theories of physics and psychology to support and enrich the knowledge that is learned; and (3) harness compositionality and learning-to-learn to rapidly acquire and generalize knowledge to new tasks and situations. We suggest concrete challenges and promising routes toward these goals that can combine the strengths of recent neural network advances with more structured cognitive models.