Genies, lawyers, and smart-asses: Extending proxy failures to intentional misunderstandings.

We propose that the logic of a genie - an agent that exploits an ambiguous request to intentionally misunderstand a stated goal - underlies a common and consequential phenomenon, well within what is currently called proxy failures. We argue that such intentional misunderstandings are not covered by the current proposed framework for proxy failures, and suggest to expand it.

Partial mental simulation explains fallacies in physical reasoning.

People can reason intuitively, efficiently, and accurately about everyday physical events. Recent accounts suggest that people use mental simulation to make such intuitive physical judgments. But mental simulation models are computationally expensive; how is physical reasoning relatively accurate, while maintaining computational tractability? We suggest that people make use of partial simulation, mentally moving forward in time only parts of the world deemed relevant. We propose a novel partial simulation model, and test it on the physical conjunction fallacy, a recently observed phenomenon [Ludwin-Peery et al. (2020). Broken physics: A conjunction-fallacy effect in intuitive physical reasoning. Psychological Science, 31(12), 1602-1611. https://doi.org/10.1177/0956797620957610] that poses a challenge for full simulation models. We find an excellent fit between our model's predictions and human performance on a set of scenarios that build on and extend those used by Ludwin-Peery et al. [(2020). Broken physics: A conjunction-fallacy effect in intuitive physical reasoning. Psychological Science, 31(12), 1602-1611. https://doi.org/10.1177/0956797620957610], quantitatively and qualitatively accounting for deviations from optimal performance. Our results suggest more generally how we allocate cognitive resources to efficiently represent and simulate physical scenes.

Learning physical parameters from dynamic scenes.

Humans acquire their most basic physical concepts early in development, and continue to enrich and expand their intuitive physics throughout life as they are exposed to more and varied dynamical environments. We introduce a hierarchical Bayesian framework to explain how people can learn physical parameters at multiple levels. In contrast to previous Bayesian models of theory acquisition (Tenenbaum, Kemp, Griffiths, & Goodman, 2011), we work with more expressive probabilistic program representations suitable for learning the forces and properties that govern how objects interact in dynamic scenes unfolding over time. We compare our model to human learners on a challenging task of estimating multiple physical parameters in novel microworlds given short movies. This task requires people to reason simultaneously about multiple interacting physical laws and properties. People are generally able to learn in this setting and are consistent in their judgments. Yet they also make systematic errors indicative of the approximations people might make in solving this computationally demanding problem with limited computational resources. We propose two approximations that complement the top-down Bayesian approach. One approximation model relies on a more bottom-up feature-based inference scheme. The second approximation combines the strengths of the bottom-up and top-down approaches, by taking the feature-based inference as its point of departure for a search in physical-parameter space.

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.

Ingredients of intelligence: From classic debates to an engineering roadmap.

We were encouraged by the broad enthusiasm for building machines that learn and think in more human-like ways. Many commentators saw our set of key ingredients as helpful, but there was disagreement regarding the origin and structure of those ingredients. Our response covers three main dimensions of this disagreement: nature versus nurture, coherent theories versus theory fragments, and symbolic versus sub-symbolic representations. These dimensions align with classic debates in artificial intelligence and cognitive science, although, rather than embracing these debates, we emphasize ways of moving beyond them. Several commentators saw our set of key ingredients as incomplete and offered a wide range of additions. We agree that these additional ingredients are important in the long run and discuss prospects for incorporating them. Finally, we consider some of the ethical questions raised regarding the research program as a whole.

Teaching Without Thinking: Negative Evaluations of Rote Pedagogy.

When people make decisions, they act in a way that is either automatic ("rote"), or more thoughtful ("reflective"). But do people notice when others are behaving in a rote way, and do they care? We examine the detection of rote behavior and its consequences in U.S. adults, focusing specifically on pedagogy and learning. We establish repetitiveness as a cue for rote behavior (Experiment 1), and find that rote people are seen as worse teachers (Experiment 2). We also find that the more a person's feedback seems similar across groups (indicating greater rote-ness), the more negatively their teaching is evaluated (Experiment 3). A word-embedding analysis of an open-response task shows people naturally cluster rote and reflective teachers into different semantic categories (Experiment 4). We also show that repetitiveness can be decoupled from perceptions of rote-ness given contextual explanation (Experiment 5). Finally, we establish two additional cues to rote behavior that can be tied to quality of teaching (Experiment 6). These results empirically show that people detect and care about scripted behaviors in pedagogy, and suggest an important extension to formal frameworks of social reasoning.

Ambivalence by design: A computational account of loopholes.

Loopholes offer an opening. Rather than comply or directly refuse, people can subvert an intended request by an intentional misunderstanding. Such behaviors exploit ambiguity and under-specification in language. Using loopholes is commonplace and intuitive in everyday social interaction, both familiar and consequential. Loopholes are also of concern in the law, and increasingly in artificial intelligence. However, the computational and cognitive underpinnings of loopholes are not well understood. Here, we propose a utility-theoretic recursive social reasoning model that formalizes and accounts for loophole behavior. The model captures the decision process of a loophole-aware listener, who trades off their own utility with that of the speaker, and considers an expected social penalty for non-cooperative behavior. The social penalty is computed through the listener's recursive reasoning about a virtual naive observer's inference of a naive listener's social intent. Our model captures qualitative patterns in previous data, and also generates new quantitative predictions consistent with novel studies (N = 265). We consider the broader implications of our model for other aspects of social reasoning, including plausible deniability and humor.

Causal implicatures from correlational statements.

Correlation does not imply causation, but this does not necessarily stop people from drawing causal inferences from correlational statements. We show that people do in fact infer causality from statements of association, under minimal conditions. In Study 1, participants interpreted statements of the form "X is associated with Y" to imply that Y causes X. In Studies 2 and 3, participants interpreted statements of the form "X is associated with an increased risk of Y" to imply that X causes Y. Thus, even the most orthodox correlational language can give rise to causal inferences.

Non-commitment in mental imagery.

We examine non-commitment in the imagination. Across 5 studies (N > 1, 800), we find that most people are non-committal about basic aspects of their mental images, including features that would be readily apparent in real images. While previous work on the imagination has discussed the possibility of non-commitment, this paper is the first, to our knowledge, to examine this systematically and empirically. We find that people do not commit to basic properties of specified mental scenes (Studies 1 and 2), and that people report non-commitment rather than uncertainty or forgetfulness (Study 3). Such non-commitment is present even for people with generally vivid imaginations, and those who report imagining the specified scene very vividly (Studies 4a, 4b). People readily confabulate properties of their mental images when non-commitment is not offered as an explicit option (Study 5). Taken together, these results establish non-commitment as a pervasive component of mental imagery.

An approximate representation of objects underlies physical reasoning.

People make fast and reasonable predictions about the physical behavior of everyday objects. To do so, people may use principled mental shortcuts, such as object simplification, similar to models developed by engineers for real-time physical simulations. We hypothesize that people use simplified object approximations for tracking and action (the body representation), as opposed to fine-grained forms for visual recognition (the shape representation). We used three classic psychophysical tasks (causality perception, time-to-collision, and change detection) in novel settings that dissociate body and shape. People's behavior across tasks indicates that they rely on coarse bodies for physical reasoning, which lies between convex hulls and fine-grained shapes. Our empirical and computational findings shed light on basic representations people use to understand everyday dynamics, and how these representations differ from those used for recognition. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Self-orienting in human and machine learning.

A current proposal for a computational notion of self is a representation of one's body in a specific time and place, which includes the recognition of that representation as the agent. This turns self-representation into a process of self-orientation, a challenging computational problem for any human-like agent. Here, to examine this process, we created several 'self-finding' tasks based on simple video games, in which players (N = 124) had to identify themselves out of a set of candidates in order to play effectively. Quantitative and qualitative testing showed that human players are nearly optimal at self-orienting. In contrast, well-known deep reinforcement learning algorithms, which excel at learning much more complex video games, are far from optimal. We suggest that self-orienting allows humans to flexibly navigate new settings.

Rethink reporting of evaluation results in AI.

Aggregate metrics and lack of access to results limit understanding.

What Could Go Wrong: Adults and Children Calibrate Predictions and Explanations of Others' Actions Based on Relative Reward and Danger.

When human adults make decisions (e.g., wearing a seat belt), we often consider the negative consequences that would ensue if our actions were to fail, even if we have never experienced such a failure. Do the same considerations guide our understanding of other people's decisions? In this paper, we investigated whether adults, who have many years of experience making such decisions, and 6- and 7-year-old children, who have less experience and are demonstrably worse at judging the consequences of their own actions, conceive others' actions as motivated both by reward (how good reaching one's intended goal would be), and by what we call "danger" (how badly one's action could end). In two pre-registered experiments, we tested whether adults and 6- and 7-year-old children tailor their predictions and explanations of an agent's action choices to the specific degree of danger and reward entailed by each action. Across four different tasks, we found that children and adults expected others to negatively appraise dangerous situations and minimize the danger of their actions. Children's and adults' judgments varied systematically in accord with both the degree of danger the agent faced and the value the agent placed on the goal state it aimed to achieve. However, children did not calibrate their inferences about how much an agent valued the goal state of a successful action in accord with the degree of danger the action entailed, and adults calibrated these inferences more weakly than inferences concerning the agent's future action choices. These results suggest that from childhood, people use a degree of danger and reward to make quantitative, fine-grained explanations and predictions about other people's behavior, consistent with computational models on theory of mind that contain continuous representations of other agents' action plans.

Dangerous Ground: One-Year-Old Infants are Sensitive to Peril in Other Agents' Action Plans.

Do infants appreciate that other people's actions may fail, and that these failures endow risky actions with varying degrees of negative utility (i.e., danger)? Three experiments, including a pre-registered replication, addressed this question by presenting 12- to 15-month-old infants (N = 104, 52 female, majority White) with an animated agent who jumped over trenches of varying depth towards its goals. Infants expected the agent to minimize the danger of its actions, and they learned which goal the agent preferred by observing how much danger it risked to reach each goal, even though the agent's actions were physically identical and never failed. When we tested younger, 10-month-old infants (N = 102, 52 female, majority White) in a fourth experiment, they did not succeed consistently on the same tasks. These findings provide evidence that one-year-old infants use the height that other agents could fall from in order to explain and predict those agents' actions.

Plans or Outcomes: How Do We Attribute Intelligence to Others?

Humans routinely make inferences about both the contents and the workings of other minds based on observed actions. People consider what others want or know, but also how intelligent, rational, or attentive they might be. Here, we introduce a new methodology for quantitatively studying the mechanisms people use to attribute intelligence to others based on their behavior. We focus on two key judgments previously proposed in the literature: judgments based on observed outcomes (you're smart if you won the game) and judgments based on evaluating the quality of an agent's planning that led to their outcomes (you're smart if you made the right choice, even if you didn't succeed). We present a novel task, the maze search task (MST), in which participants rate the intelligence of agents searching a maze for a hidden goal. We model outcome-based attributions based on the observed utility of the agent upon achieving a goal, with higher utilities indicating higher intelligence, and model planning-based attributions by measuring the proximity of the observed actions to an ideal planner, such that agents who produce closer approximations of optimal plans are seen as more intelligent. We examine human attributions of intelligence in three experiments that use MST and find that participants used both outcome and planning as indicators of intelligence. However, observing the outcome was not necessary, and participants still made planning-based attributions of intelligence when the outcome was not observed. We also found that the weights individuals placed on plans and on outcome correlated with an individual's ability to engage in cognitive reflection. Our results suggest that people attribute intelligence based on plans given sufficient context and cognitive resources and rely on the outcome when computational resources or context are limited.

Sticking to the Evidence? A Behavioral and Computational Case Study of Micro-Theory Change in the Domain of Magnetism.

Constructing an intuitive theory from data confronts learners with a "chicken-and-egg" problem: The laws can only be expressed in terms of the theory's core concepts, but these concepts are only meaningful in terms of the role they play in the theory's laws; how can a learner discover appropriate concepts and laws simultaneously, knowing neither to begin with? We explore how children can solve this chicken-and-egg problem in the domain of magnetism, drawing on perspectives from computational modeling and behavioral experiments. We present 4- and 5-year-olds with two different simplified magnet-learning tasks. Children appropriately constrain their beliefs to two hypotheses following ambiguous but informative evidence. Following a critical intervention, they learn the correct theory. In the second study, children infer the correct number of categories given no information about the possible causal laws. Children's hypotheses in these tasks are explained as rational inferences within a Bayesian computational framework.

Lucky or clever? From expectations to responsibility judgments.

How do people hold others responsible for the consequences of their actions? We propose a computational model that attributes responsibility as a function of what the observed action reveals about the person, and the causal role that the person's action played in bringing about the outcome. The model first infers what type of person someone is from having observed their action. It then compares a prior expectation of how a person would behave with a posterior expectation after having observed the person's action. The model predicts that a person is blamed for negative outcomes to the extent that the posterior expectation is lower than the prior, and credited for positive outcomes if the posterior is greater than the prior. We model the causal role of a person's action by using a counterfactual model that considers how close the action was to having been pivotal for the outcome. The model captures participants' responsibility judgments to a high degree of quantitative accuracy across three experiments that cover a range of different situations. It also solves an existing puzzle in the literature on the relationship between action expectations and responsibility judgments. Whether an unexpected action yields more or less credit depends on whether the action was diagnostic for good or bad future performance.

Mind Games: Game Engines as an Architecture for Intuitive Physics.

We explore the hypothesis that many intuitive physical inferences are based on a mental physics engine that is analogous in many ways to the machine physics engines used in building interactive video games. We describe the key features of game physics engines and their parallels in human mental representation, focusing especially on the intuitive physics of young infants where the hypothesis helps to unify many classic and otherwise puzzling phenomena, and may provide the basis for a computational account of how the physical knowledge of infants develops. This hypothesis also explains several 'physics illusions', and helps to inform the development of artificial intelligence (AI) systems with more human-like common sense.

Ten-month-old infants infer the value of goals from the costs of actions.

Infants understand that people pursue goals, but how do they learn which goals people prefer? We tested whether infants solve this problem by inverting a mental model of action planning, trading off the costs of acting against the rewards actions bring. After seeing an agent attain two goals equally often at varying costs, infants expected the agent to prefer the goal it attained through costlier actions. These expectations held across three experiments that conveyed cost through different physical path features (height, width, and incline angle), suggesting that an abstract variable-such as "force," "work," or "effort"-supported infants' inferences. We modeled infants' expectations as Bayesian inferences over utility-theoretic calculations, providing a bridge to recent quantitative accounts of action understanding in older children and adults.