Two very different kinds of views exist of how infants learn about physics: on the one hand, through domain-specific rules, on the other, through general purpose simulation. We attempt to reconcile these two views through a model that uses simulation to bootstrap rule learning. This model makes a variety of predictions about rapid concept acquisition in young infants which are consistent with experiments performed by developmental cognitive scientists. Consistent with the developmental literature, our model learns physical rules from just a few examples, but only when those examples are consistent with general physical principles. A model without simulation shows no such biases. Our approach provides a general mechanism for explaining how simulation and rule learning might bootstrap off of each other throughout development, and opens up a number of new questions about how children learn physical representations.

Humans have a remarkable ability to go beyond the observable. From seeing the current state of our shared kitchen, we can infer what happened and who did it. Prior work has shown how the physical state of the world licenses inferences about the causal history of events, and the agents that participated in these events. Here, we investigate a previously unstudied source of evidence about what happened: social evaluations. In our experiment, we present situations in which a group failed to optimally coordinate their actions. Participants learn how much each agent was blamed for the outcome, and their task is to make inferences about the situation, the agents' actions, as well as the agents' capabilities. We develop a computational model that accurately captures participants' inferences. The model assumes that people blame others by considering what they should have done, and what causal role their action played. By inverting this generative model of blame, people can figure out what happened.

We introduce a new problem solving paradigm: solving physical puzzles by placing tool-like objects in a scene. Thepuzzles are designed to explicitly evoke different physical concepts such as support, blocking, tipping, and launching, andare typically solved in a handful of trials. We study human participants’ problem solving strategies, including what theytry first, how they update their actions based on failed attempts, and how many attempts they eventually take to solvethe puzzles. We introduce the ‘Sample, Simulate, Remember’ model that incorporates object-based priors to generatehypotheses, mental simulation to test hypotheses, and a memory and generalization system to update across simulationsand real-world trials, and show that all three components are needed to explain human performance. Further results canbe found at https://k-r-allen.github.io/tool-games/

Humans are remarkably flexible tool users. We not only recognize a wide range of existing tools, but also producenew tools by seeing objects in new ways, or by making or repurposing objects to solve a problem confronting us. Here westudy the cognitive processes supporting flexible tool use, including deciding what makes a good tool, and how it should beused. Participants played a video game which requires selecting an object from a set of options and placing it in a virtualphysical scene in order to accomplish goals such as tipping another object over or launching it into a container. People appearto use a combination of simulation-based planning and experience-based heuristics: fast heuristics drive the initial selectionand placement of a candidate tool, and that solution can then be refined by several rounds of mental simulation interspersedwith trial-and-error experimentation to rapidly converge on goal-satisfying solutions.

Developmental psychology presents us with a puzzle: though children are remarkably apt at planning their actions, theysuffer from surprising yet consistent shortcomings. We argue that these patterns of triumph and failure can be broadlycaptured by the framework of task and motion planning, where plans are hybrid entities consisting of both a structured,symbolic skeleton and a continuous, low-level trajectory. As a proof of concept, we model two case studies from the tooluse literature and show how their results can be understood by the interaction of symbolic and continuous plans.

Our ability to make reliable physical predictions even in novel settings is a hallmark of human intelligence. Here we investigate how people infer multiple physical variables simultaneously and compose them to generalize to a novel scenario. Participants (N=203) observed a series of balls launched at different angles in a 2D virtual environment and generated predictions about their trajectories. We found that people could infer the masses of different balls based on these observations, as well as the existence of a latent "wind" force, and compose knowledge of these two variables to generalize to novel situations in a subsequent test phase. We modeled this generalization as the consequence of being able to simulate trajectories by independently combining force and mass information in accordance with Newtonian mechanics. To validate this approach, we also tested several alternative models and compared their generalization behavior to one another and to that of people. Together, our study points to the value of using generalization to probe the underlying representations supporting physical prediction.

Over the last decades, games have become one of the most popular recreational activities, not only among children but also among adults. Consequently, they have also gained popularity as an avenue for studying cognition. Games offer several advantages, such as the possibility to gather big data sets, engage participants to play for a long time, and better resemblance of real world complexities. In this workshop, we will bring together leading researchers from across the cognitive sciences to explore how games can be used to study diverse aspects of intelligent behavior, explore their differences compared to classical lab experiments, and discuss the future of game-based cognitive science research.

Humans make plans with remarkable flexibility by leveraging symbolic representations. How are these representationslearned? We present a model that starts out with a language of low-level physical constraints and, by observing expertdemonstrations, builds up a library of high-level concepts that afford planning and action understanding. We demonstrateits versatility through experiments inspired by developmental psychology literature.

Many models of intuitive physical reasoning posit some kind of mental simulation mechanism, yet everyday environments frequently contain far more objects than people could plau- sibly represent with their limited cognitive capacity. What determines which objects are actually included in our repre- sentations? We asked participants to predict how a ball will bounce through a complex field of obstacles, and probed work- ing memory for objects in the scene that were more and less likely to be relevant to the ball’s trajectory. We evaluate differ- ent accounts of relevance and find that successful object mem- ory is best predicted by how frequently a ball’s trajectory is expected to contact that object under a probabilistic simulation model. This suggests that people construct representations for mental simulation efficiently and dynamically, on the fly, by adding objects “just in time”: only when they are expected to become relevant for the next stage of simulation.

`Embodied cognition' suggests that our motor experiences shape our cognitive and perceptual capabilities broadly, but often considers tasks that directly relate to or manipulate the body. Here we study how a history of natural embodied experience affects abstract physical problem-solving in a virtual, disembodied physical reasoning task. We compare how groups with different embodied experience -- congenitally limb-different versus two-handed children and adults -- perform on this task, and find that while there is no difference in overall performance, limb-different participants solved problems using fewer actions, and spent a longer time thinking before acting. This suggests that differences in embodied experience drive the acquisition of different meta-strategies for balancing acting with thinking, even on tasks that are designed to equalize differences in embodiment.