publications

2023

1. Why think step by step? Reasoning emerges from the locality of experience

   Ben Prystawski, Michael Y Li, and Noah D Goodman

   Advances in Neural Information Processing Systems, 2023

   Abs

   Humans have a powerful and mysterious capacity to reason. Working through a set of mental steps enables us to make inferences we would not be capable of making directly even though we get no additional data from the world. Similarly, when large language models generate intermediate steps (a chain of thought) before answering a question, they often produce better answers than they would directly. We investigate why and how chain-of-thought reasoning is useful in language models, testing the hypothesis that reasoning is effective when training data consists of overlapping local clusters of variables that influence each other strongly. These training conditions enable the chaining of accurate local inferences to estimate relationships between variables that were not seen together in training. We prove that there will exist a "reasoning gap", where reasoning through intermediate variables reduces bias, for the simple case of an autoregressive density estimator trained on local samples from a chain-structured probabilistic model. We then test our hypothesis experimentally in more complex models, training an autoregressive language model on samples from Bayes nets but only including a subset of variables in each sample. We test language models’ ability to match conditional probabilities with and without intermediate reasoning steps, finding that intermediate steps are only helpful when the training data is locally structured with respect to dependencies between variables. The combination of locally structured observations and reasoning is much more data-efficient than training on all variables. Our results illustrate how the effectiveness of reasoning step by step is rooted in the local statistical structure of the training data.

2. Cultural reinforcement learning: a framework for modeling cumulative culture on a limited channel

   Ben Prystawski, Dilip Arumugam, and Noah Goodman

   In Proceedings of the 45th Annual Conference of the Cognitive Science Society, 2023

   Abs

   Humans’ capacity for cumulative culture is remarkable: we can build up vast bodies of knowledge over generations. Communication, particularly via language, is a key component of this process. Previous work has described language as enabling posterior passing, where one Bayesian agent transmits a posterior distribution to the next. In practice, we cannot exactly copy our beliefs into the minds of others–we must communicate over the limited channel language provides. In this paper, we analyze cumulative culture as Bayesian reinforcement learning with communication over a rate-limited channel. We implement an agent that solves a crafting task and communicates to the next agent by approximating the optimal rate-distortion trade-off. Our model produces documented effects, such as the benefits of abstraction and selective social learning. It also suggests a new hypothesis: selective social learning can be harmful in tasks where initial exploration is required.

3. Psychologically-informed chain-of-thought prompts for metaphor understanding in large language models

   Ben Prystawski, Paul Thibodeau, Christopher Potts, and Noah D Goodman

   In Proceedings of the 45th Annual Conference of the Cognitive Science Society, 2023

   Abs

   Probabilistic models of language understanding are valuable tools for investigating human language use. However, they need to be hand-designed for a particular domain. In contrast, large language models (LLMs) are trained on text that spans a wide array of domains, but they lack the structure and interpretability of probabilistic models. In this paper, we use chain-of-thought prompts to introduce structures from probabilistic models into LLMs. We explore this approach in the case of metaphor understanding. Our chain-of-thought prompts lead language models to infer latent variables and reason about their relationships in order to choose appropriate paraphrases for metaphors. The latent variables and relationships chosen are informed by theories of metaphor understanding from cognitive psychology. We apply these prompts to the two largest versions of GPT-3 and show that they can improve performance in a paraphrase selection task

4. Toward a normative theory of (self-) management by goal-setting

   Nishad Singhi, Florian Mohnert, Ben Prystawski, and Falk Lieder

   In Proceedings of the 45th Annual Conference of the Cognitive Science Society, 2023

   Abs

   People are often confronted with problems whose complexity exceeds their cognitive capacities. To deal with this complexity, individuals and managers can break complex problems down into a series of subgoals. Which subgoals are most effective depends on people’s cognitive constraints and the cognitive mechanisms of goal pursuit. This creates an untapped opportunity to derive practical recommendations for which subgoals managers and individuals should set from cognitive models of bounded rationality. To seize this opportunity, we apply the principle of resource-rationality to formulate a mathematically precise normative theory of (self-)management by goal-setting. We leverage this theory to computationally derive optimal subgoals from a resource-rational model of human goal pursuit. Finally, we show that the resulting subgoals improve the problem-solving performance of bounded agents and human participants. This constitutes a first step towards grounding prescriptive theories of management and practical recommendations for goal-setting in computational models of the relevant psychological processes and cognitive limitations.

2022

1. The Emergence of Gender Associations in Child Language Development

   Ben Prystawski, Erin Grant, Aida Nematzadeh, Spike W. S. Lee, Suzanne Stevenson, and Yang Xu

   Cognitive Science, 2022

   Abs

   Abstract Gender associations have been a long-standing research topic in psychological and social sciences. Although it is known that children learn aspects of gender associations at a young age, it is not well understood how they might emerge through the course of development. We investigate whether gender associations, such as the association of dresses with women and bulldozers with men, are reflected in the linguistic communication of young children from ages 1–5. Drawing on recent methods from machine learning, we use word embeddings derived from large text corpora including news articles and web pages as a proxy for gender associations in society, and we compare those with the gender associations of words uttered by caretakers and children in children’s linguistic environment. We quantify gender associations in childhood language through gender probability, which measures the extent to which word usage frequencies in speech to and by girls and boys are gender-skewed. By analyzing 4,875 natural conversations between children and their caretakers in North America, we find that frequency patterns in word usage of both caretakers and children correlate strongly with the gender associations captured in word embeddings through the course of development. We discover that these correlations diminish from the 1970s to the 1990s. Our work suggests that early linguistic communication and social changes may jointly contribute to the formation of gender associations in childhood.

2. Resource-rational Models of Human Goal Pursuit

   Ben Prystawski, Florian Mohnert, Mateo Tošić, and Falk Lieder

   Topics in Cognitive Science, 2022

   Abs

   Abstract Goal-directed behavior is a deeply important part of human psychology. People constantly set goals for themselves and pursue them in many domains of life. In this paper, we develop computational models that characterize how humans pursue goals in a complex dynamic environment and test how well they describe human behavior in an experiment. Our models are motivated by the principle of resource rationality and draw upon psychological insights about people’s limited attention and planning capacities. We find that human goal pursuit is qualitatively different and substantially less efficient than optimal goal pursuit in our simulated environment. Models of goal pursuit based on the principle of resource rationality capture human behavior better than both a model of optimal goal pursuit and heuristics that are not resource-rational. We conclude that the way humans pursue goals is shaped by the need to achieve goals effectively as well as cognitive costs and constraints on planning and attention. Our findings are an important step toward understanding humans’ goal pursuit as cognitive limitations play a crucial role in shaping people’s goal-directed behavior.

2021

1. Algorithms for Adaptive Experiments that Trade-off Statistical Analysis with Reward: Combining Uniform Random Assignment and Reward Maximization

   Jacob Nogas, Tong Li, Fernando J. Yanez, Arghavan Modiri, Nina Deliu, Ben Prystawski, Sofia S Villar, Anna Rafferty, and Joseph J Williams

   In Presented at Causal Inference Challenges in Sequential Decision Making: Bridging Theory and Practice - a NeurIPS 2021 Workshop, 2021

   Abs

   Multi-armed bandit algorithms like Thompson Sampling can be used to conduct adaptive experiments, in which maximizing reward means that data is used to progressively assign more participants to more effective arms. Such assignment strategies increase the risk of statistical hypothesis tests identifying a difference between arms when there is not one, and failing to conclude there is a difference in arms when there truly is one. We present simulations for 2-arm experiments that explore two algorithms that combine the benefits of uniform randomization for statistical analysis, with the benefits of reward maximization achieved by Thompson Sampling (TS). First, Top-Two Thompson Sampling adds a fixed amount of uniform random allocation (UR) spread evenly over time. Second, a novel heuristic algorithm, called TS PostDiff (Posterior Probability of Difference). TS PostDiff takes a Bayesian approach to mixing TS and UR: the probability a participant is assigned using UR allocation is the posterior probability that the difference between two arms is ‘small’ (below a certain threshold), allowing for more UR exploration when there is little or no reward to be gained. We find that TS PostDiff method performs well across multiple effect sizes, and thus does not require tuning based on a guess for the true effect size.

2. Modelling Recognition in Human Puzzle Solving

   Ben Prystawski, Rebekah Gelpi, Christopher G Lucas, and Daphna Buchsbaum

   In Proceedings of the 43nd Annual Conference of the Cognitive Science Society, 2021

   Abs

   Our ability to play games like chess and Go relies on both planning several moves ahead and on recognition or gist - intuitively assessing the quality of possible game states without explicit planning. In this paper, we investigate the role of recognition in puzzle solving. We introduce a simple puzzle game to study planning and recognition in a non-adversarial context and a reinforcement learning agent which solves these puzzles relying purely on recognition. The agent relies on a neural network to capture intuitions about which game states are promising. We find that our model effectively predicts the relative difficulty of the puzzles for humans and shows similar qualitative patterns of success and initial moves to humans. Our task and model provide a basis for the study of planning and intuitive notions of fit in puzzle solving that is simple enough for use in developmental studies.

2020

1. Tracing the Emergence of Gendered Language in Childhood

   Ben Prystawski, Erin Grant, Aida Nematzadeh, Spike W. S. Lee, Suzanne Stevenson, and Yang Xu

   In Proceedings of the 42nd Annual Conference of the Cognitive Science Society, 2020

   Abs

   Are gender associations in general language reflected in the words spoken to and by children? Previous work has suggested that language reveals gender differences in discourse, speech style, language use and acquisition. Work in artificial intelligence has shown that word embeddings trained on large corpora reflect human gender associations. We connect this work to developmental psychology by exploring whether gender associations in word embeddings are present in the linguistic input and output of children, and if so, how early gendered language emerges. We present a computational method that quantifies the gender associations of words and use a corpus of child-caretaker speech to show that these gender associations correlate significantly with those in word embeddings. We discover that gendered word use emerges in English-speaking children around age 2, and the gender associations cannot be explained solely by variables including word length, frequency, concreteness, and valence.

2. Incremental Hypothesis Revision in Causal Reasoning Across Development

   Rebekah Gelpi, Ben Prystawski, Christopher G Lucas, and Daphna Buchsbaum

   In Proceedings of the 42nd Annual Conference of the Cognitive Science Society, 2020

   Abs

   We explore whether children’s strategies on a causal learning task show a bias observed in adults towards “exploitative” hypothesis revision. Adults and children (ages 4–6) were presented with evidence which initially seemed to conform to a simple, salient rule (e.g. blue blocks activate a machine), but then encountered evidence that violated this rule. The true rule in the "near" condition was more complex, but could be reached through iterative revision of the salient rule, while in the "distant" condition, the true rule was comparatively simple, but incremental revision could not yield the true rule. Participants then predicted the behaviour of a set of new blocks. Adults performed better in the near condition, while in the distant condition adults did not appear to revise their initial hypothesis significantly. Unlike adults, children’s overall performance in both conditions was similar, while condition differences may reflect a broader search for alternative solutions.