About

Todd Gureckis is a Professor of Psychology at New York University. His research centers on the memory, learning, and decision processes that enable intelligent and adaptive behaviors. He is particularly interested in how individuals uncover important and useful regularities about the environment through experience, such as how people come to agree on concepts like 'mammal,' how learning experiences shape perception, and how new skills and behaviors are acquired. His goal is to deepen the understanding of the mechanisms supporting these behaviors, their development and change over the lifespan, and their impact by disease or brain damage. Central to his work is the use of computational models as tools for integrating and guiding research. These models are detailed psychological theories implemented as computer programs, providing a framework for evaluating theories of cognitive function. They help organize diverse findings, generate predictions to guide empirical research, and inform the development of artificial systems capable of autonomous learning. Gureckis received his B.S. in Computer/Electrical Engineering from the University of Texas at Austin in 2001, followed by a M.A. and Ph.D. in psychology from UT Austin in 2004 and 2005 respectively. After a postdoctoral position at Indiana University, he joined NYU as a faculty member in the psychology program in January 2008. His research continues to explore the mechanisms of learning, perception, and decision-making through computational modeling.

Research topics

• Computer Science
• Artificial Intelligence
• Cognitive psychology
• Psychology
• Physics
• Epistemology
• Neuroscience
• Social psychology
• Cognitive science
• Business
• Philosophy
• Process management
• Risk analysis (engineering)

Selected publications

• A Comprehensive Behavioral Dataset for the Abstraction and Reasoning Corpus

  Scientific Data · 2025-08-07

  articleOpen accessSenior author

  The Abstraction and Reasoning Corpus (ARC) is a visual program synthesis benchmark designed to test out-of-distribution generalization in machines. Comparing AI algorithms to human performance is essential to measure progress on these problems. In this paper, we present H-ARC (Human-ARC): a novel large-scale dataset containing solution attempts from over 1700 humans on ARC problems. The dataset spans the full set of 400 training and 400 evaluation tasks from the original ARC benchmark, and it is the largest human evaluation to date. By publishing the dataset, we contribute human responses to each problem, step-by-step behavioral action traces from the ARC user-interface, and natural-language solution descriptions of the inferred program/rule. We believe this dataset will be of value to researchers, both in cognitive science and AI, since it offers the potential to facilitate the discovery of underlying mechanisms supporting abstraction and reasoning in people. The insights to be gained from these data not only have value for cognitive science, but could in turn inform the design of more efficient, human-like AI algorithms.

  PublisherOA PDFDOI

• Distinct paths to false memory revealed in hundreds of narrative recalls

  2025-10-15

  preprintOpen accessSenior author

  Memory distortions emerge from a complex interplay between prior knowledge and ongoing experience—dynamics which are not readily provoked in controlled laboratory experiments. Here we investigate naturally occurring memory distortions using the largest known dataset of narrative recall, comprising hundreds of spoken recollections. Using large language models (LLMs), we developed an automated pipeline to detect and classify spontaneous false memories. Across two validation experiments, we demonstrate that human-AI agreement matches inter-human reliability in detecting and cataloging memory distortions. We show that false memories reflect two distinct phenomena which are driven by separable semantic factors: similarity to prototypical narrative patterns drives factual errors (distortions of actual content), whereas contextual surprise drives confabulations (entirely fabricated details). Through this combination of large-scale naturalistic data and AI-powered automation tools, we reveal memory processes that controlled laboratory paradigms cannot easily capture and illuminate the complex dynamics of human (mis)remembering in real-world contexts.

  PublisherDOI

• Distinct paths to false memory revealed in hundreds of narrative recalls

  2025-10-19

  preprintOpen accessSenior author

  Memory distortions emerge from a complex interplay between prior knowledge and ongoing experience—dynamics which are not readily provoked in controlled laboratory experiments. Here we investigate naturally occurring memory distortions using the largest known dataset of narrative recall, comprising hundreds of spoken recollections. Using large language models (LLMs), we developed an automated pipeline to detect and classify spontaneous false memories. Across two validation experiments, we demonstrate that human-AI agreement matches inter-human reliability in detecting and cataloging memory distortions. We show that false memories reflect two distinct phenomena which are driven by separable semantic factors: similarity to prototypical narrative patterns drives factual errors (distortions of actual content), whereas contextual surprise drives confabulations (entirely fabricated details). Through this combination of large-scale naturalistic data and AI-powered automation tools, we reveal memory processes that controlled laboratory paradigms cannot easily capture and illuminate the complex dynamics of human (mis)remembering in real-world contexts.

  PublisherDOI

• Seeking new information with old questions: Children and adults reuse and recombine concepts from prior questions

  2025-06-03

  preprintOpen accessSenior author

  Question asking is a key tool for learning about the world, especially in childhood. However, formulating good questions is challenging. In any given situation, many questions are possible but only few are informative. In the present work, we investigate two ways 5- to 10-year-olds and adults simplify the challenge of formulating questions: by reusing previous questions, and by recombining components of previous questions to form new questions. In Study 1, we develop a new question asking task, verify its suitability for studying question asking in children and adults, and conduct a preliminary investigation of how children and adults reuse and recombine their own prior questions. In Study 2, we experimentally manipulate exposure to another person's questions, investigating under what conditions children and adults reuse and recombine others' questions. Our experimental results suggest that both children and adults reuse and recombine questions, and they adaptively modulate reuse depending on how informative a question will be in a particular situation. Moreover, children reuse and recombine prior questions more frequently than adults in some cases. This work shows that prior questions provide fodder for future questions, simplifying the challenge of inquiry and enabling effective learning.

  PublisherOA PDFDOI

• Seeking New Information With Old Questions: Children and Adults Reuse and Recombine Concepts From Prior Questions

  Open Mind · 2025-01-01

  articleOpen accessSenior author

  Question asking is a key tool for learning about the world, especially in childhood. However, formulating good questions is challenging. In any given situation, many questions are possible but only few are informative. In the present work, we investigate two ways 5- to 10-year-olds and adults simplify the challenge of formulating questions: by reusing previous questions, and by recombining components of previous questions to form new questions. In Study 1, we develop a new question asking task, verify its suitability for studying question asking in children and adults, and conduct a preliminary investigation of how children and adults reuse and recombine their own prior questions. In Study 2, we experimentally manipulate exposure to another person's questions, investigating under what conditions children and adults reuse and recombine others' questions. Our experimental results suggest that both children and adults reuse and recombine questions, and they adaptively modulate reuse depending on how informative a question will be in a particular situation. Moreover, children reuse and recombine prior questions more frequently than adults in some cases. This work shows that prior questions provide fodder for future questions, simplifying the challenge of inquiry and enabling effective learning.

  PublisherDOI

• Correction: A Comprehensive Behavioral Dataset for the Abstraction and Reasoning Corpus

  Scientific Data · 2025-08-22

  erratumOpen accessSenior author
  PublisherOA PDFDOI

• Decision rule inference limits social escape from learning traps

  2025-09-15

  preprintOpen accessSenior author

  Individual learners often show a tendency to engage in self-reinforcing avoidance, a pattern referred to as a learning trap. Across five experiments, we investigated the extent to which previously trapped learners can escape via social observational learning. While social observational learning did help a significant number of trapped learners escape, the majority of trapped learners remained trapped even after observing a partner demonstrate an optimal decision rule. Across several follow-up experiments, we unpack possible factors which limited the effectiveness of social observational learning. Overall, the results suggest that social decision rule inference (inferring a partner's decision rule from observed choices) was a key bottleneck for observational learning. Simulations show that these results were unanticipated by a leading model of social reward learning, and highlight a central role for inference in social learning.

  PublisherOA PDFDOI

• Goals as reward-producing programs

  Nature Machine Intelligence · 2025-02-21 · 5 citations

  article
  PublisherDOI

• Can cognitive discovery be incentivized with money?

  Journal of Experimental Psychology General · 2025-01-21 · 1 citations

  articleSenior author

  The ability to discover patterns or rules from our experiences is critical to science, engineering, and art. In this article, we examine how much people's discovery of patterns can be incentivized by financial rewards. In particular, we investigate a classic category learning task for which the effect of financial incentives is unknown (Shepard et al., 1961). Across five experiments, we find no effect of incentive on rule discovery performance. However, in a sixth experiment requiring category recognition but not learning, we find a large effect of incentives on response time and a small effect on task performance. Participants appear to apply more effort in valuable contexts, but the effort is disproportionate with the performance improvement. Taken together, the results suggest that performance in tasks that require novel inductive insights is relatively immune to financial incentives, while tasks that require rote perseverance of a fixed strategy are more malleable. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

  PublisherDOI

• Decision rule inference limits social escape from learning traps

  2025-09-16

  preprintOpen accessSenior author

  Individual learners often show a tendency to engage in self-reinforcing avoidance, a pattern referred to as a learning trap. Across five experiments, we investigated the extent to which previously trapped learners can escape via social observational learning. While social observational learning did help a significant number of trapped learners escape, the majority of trapped learners remained trapped even after observing a partner demonstrate an optimal decision rule. Across several follow-up experiments, we unpack possible factors which limited the effectiveness of social observational learning. Overall, the results suggest that social decision rule inference (inferring a partner's decision rule from observed choices) was a key bottleneck for observational learning. Simulations show that these results were unanticipated by a leading model of social reward learning, and highlight a central role for inference in social learning.

  PublisherOA PDFDOI

Recent grants

• CompCog: Towards a computational cognitive science of helping

  NSF · $646k · 2020–2024

• CAREER: The Role of Self-Directed Learning in Facilitating Concept Acquisition: Advancing Research and Training in the Cognitive Science of Learning

  NSF · $766k · 2013–2019

• NCS-FO: Using computational cognitive neuroscience to predict and optimize memory

  NSF · $955k · 2016–2020

Frequent coauthors

• Bradley C. Love

  University College London

  24 shared

• Douglas Markant

  University of North Carolina at Charlotte

  24 shared

• Neil R Bramley

  University of Edinburgh

  23 shared

• Brenden M. Lake

  18 shared

• Anna Coenen

  New York University

  18 shared

• Robert L. Goldstone

  Indiana University

  14 shared

• Alexander Rich

  Edaptive Computing (United States)

  13 shared

• Azzurra Ruggeri

  Wellcome Centre for Integrative Neuroimaging

  13 shared