Brendan Lake is a cognitive scientist and professor at New York University, where he investigates the computational principles that enable human learning and reasoning, with the goal of building more human-like artificial intelligence. His research sits at the intersection of Machine learning, cognitive-science, and Artificial intelligence, focusing on how machines can learn from fewer examples, generalize more flexibly, and acquire concepts in ways that mirror human cognition. Lake is particularly known for his work on probabilistic program induction, the "Human Level AI" challenge, and his advocacy for integrating insights from developmental psychology into AI research.
Lake completed his undergraduate studies at Stanford University, where he developed an early interest in both computer science and psychology. He then pursued graduate work at the Massachusetts Institute of Technology, earning a PhD in cognitive science under the supervision of Joshua Tenenbaum. His doctoral research centered on modeling human concept learning using probabilistic programs, a framework that treats concepts as structured generative models. This work culminated in several influential papers that demonstrated how humans can learn complex concepts from very few examples, a capability that standard deep learning models at the time lacked.
Early Career and Academic Positions
After completing his PhD, Lake held postdoctoral positions at MIT and later at New York University, where he collaborated with researchers in both the psychology and computer science departments. In 2017, he joined the faculty at NYU as an assistant professor in the Department of Psychology and the Center for Data Science. He was promoted to associate professor in 2021 and to full professor in 2024. His lab, the Computation and Cognition Lab, focuses on building computational models that capture human learning and reasoning abilities, with an emphasis on compositionality, causality, and intuitive physics.
Lake has also been affiliated with the Center for Brains, Minds and Machines (CBMM), a multi-institutional research center funded by the National Science Foundation, which aims to understand intelligence by combining neuroscience, cognitive science, and computer science. Through CBMM, he has worked closely with colleagues such as Joshua Tenenbaum and Melanie Mitchell on projects exploring the limits of current AI systems.
Key Research Contributions
One of Lake's most cited contributions is the 2015 paper "Human-level concept learning through probabilistic program induction," published in Science, which introduced the Omniglot dataset. This dataset consists of 1,623 handwritten characters from 50 different writing systems, designed to test one-shot learning. Lake and his colleagues showed that a Bayesian program learning model could learn a new character from a single example and perform as well as humans on classification tasks, while also generating new, human-like variations of the characters. This work highlighted the importance of structured knowledge and compositionality, which are often absent in purely connectionist approaches.
In 2016, Lake co-authored a widely discussed paper titled "Building machines that learn and think like people," with Joshua Tenenbaum and others. The paper argued that AI should draw more heavily on cognitive science, emphasizing the need for models that can learn causal models, use intuitive physics, and reason about unobserved variables. It proposed a roadmap for combining deep learning with structured probabilistic models, a theme that Lake has continued to develop in subsequent work.
Lake has also contributed to the study of intuitive physics and intuitive psychology. His research has explored how humans make predictions about physical events, such as the stability of towers of blocks, and how these predictions can be modeled using simulation-based approaches. He has published papers on the role of physical simulation in human judgment, showing that people's expectations align closely with the outputs of physics engines, suggesting that the brain may perform approximate simulations.
The Human Level AI Challenge
In 2022, Lake introduced the "Human Level AI" challenge, a benchmark designed to test whether AI systems can achieve human-level performance on a battery of cognitive tasks. The challenge includes tasks such as visual concept learning, language understanding, and social reasoning, drawing from developmental psychology paradigms. Lake argued that current large-scale models, despite their impressive performance on narrow tasks, fail to exhibit the flexibility and robustness of human cognition. The challenge has been adopted by several research groups as a target for evaluating progress toward more general AI.
Lake has been a vocal critic of the idea that scaling up Neural network models alone will lead to human-level intelligence. In talks and papers, he has pointed out that even large language models, such as those based on the Transformer (architecture) architecture, struggle with compositional generalization and causal reasoning. He advocates for a hybrid approach that combines deep learning with symbolic reasoning and probabilistic inference, an idea that has gained traction in the broader AI community.
Collaborations and Interdisciplinary Work
Throughout his career, Lake has emphasized the importance of interdisciplinary collaboration. He has worked with developmental psychologists to study how children learn concepts, with neuroscientists to understand the neural basis of cognition, and with computer scientists to develop new algorithms. His collaborations have extended to researchers at Google DeepMind, OpenAI, and other leading AI labs, though he has also maintained a critical perspective on the limitations of these systems.
Lake has served on the editorial boards of several journals, including Cognitive Science and the Journal of Machine Learning Research. He has also been a program committee member for major conferences such as NeurIPS, ICML, and CogSci. His work has been funded by grants from the National Science Foundation, the Defense Advanced Research Projects Agency (DARPA), and the Simons Foundation.
Teaching and Mentorship
At NYU, Lake teaches courses on computational cognitive science and machine learning. His courses are known for their rigor and for encouraging students to think critically about the assumptions underlying AI models. He has mentored numerous PhD students and postdoctoral fellows, many of whom have gone on to positions in academia and industry. His mentorship style emphasizes deep conceptual understanding and a willingness to question dominant paradigms.
Lake has also been active in public outreach, giving talks at venues such as the Aspen Ideas Festival and appearing in documentaries about AI. He has written opinion pieces for popular outlets, arguing for a more scientifically grounded approach to AI development. His public engagement aims to demystify AI and to highlight the gaps between current systems and human intelligence.
Awards and Recognition
Lake has received several awards for his research, including the Cognitive Science Society's Computational Modeling Prize and the Sloan Research Fellowship in Neuroscience. In 2020, he was named a CIFAR Azrieli Global Scholar in the Artificial Intelligence program, which recognizes early-career researchers with exceptional promise. His papers have been cited tens of thousands of times, and he is regularly invited to give keynote lectures at international conferences.
Current Directions
As of 2025, Lake's research continues to explore how to build AI systems that learn and think like people. His recent work has focused on developing benchmarks for measuring compositional generalization, studying how humans learn abstract rules, and investigating the role of language in shaping thought. He has also begun to explore how insights from cognitive science can inform the design of more interpretable and trustworthy AI systems, a topic of growing importance as AI is deployed in high-stakes domains.
Lake remains optimistic about the potential for AI to benefit society, but he cautions against overhyping current capabilities. He argues that achieving human-level AI will require a deeper understanding of human cognition, not just more data and compute. His vision is one where AI systems are built on principles derived from cognitive science, leading to machines that can learn, reason, and interact with the world in ways that are genuinely intelligent.
Selected Publications
Among Lake's most influential publications are:
- Lake, B. M., Salakhutdinov, R., & Tenenbaum, J. B. (2015). Human-level concept learning through probabilistic program induction. Science, 350(6266), 1332-1338.
- Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2017). Building machines that learn and think like people. Behavioral and Brain Sciences, 40, e253.
- Lake, B. M., & Baroni, M. (2018). Generalization without systematicity: On the compositional skills of sequence-to-sequence recurrent networks. ICML.
- Lake, B. M. (2019). Compositional generalization through meta-sequence-to-sequence learning. NeurIPS.
These works have shaped debates about the nature of generalization and the future of AI research.
Impact on the Field
Lake's insistence on bringing cognitive science back into AI has influenced a new generation of researchers. His critiques of deep learning have prompted many to reconsider the assumptions behind large-scale models. While some have dismissed his views as overly pessimistic, others see them as a necessary corrective to the field's tendency to equate scale with intelligence. His work on Omniglot has become a standard benchmark for few-shot learning, and his ideas about probabilistic programs have inspired numerous follow-up studies.
In summary, Brendan Lake stands out as a leading voice advocating for a more human-centered approach to artificial intelligence. By bridging cognitive science and machine learning, he has made lasting contributions to both fields and continues to challenge the AI community to aim higher than mere pattern matching.