People excel at real-life tasks that are challenging for modern AI, such as planning, learning new tools from a few examples, and social cognition. For instance, people easily find their way in new buildings, learn to use new tools from few examples, and rapidly make accurate social attributions. This data-efficiency and broad generalization remain some of the most salient gaps between human and machine intelligence, motivating the question: how does human cognition solve real-world problems so effectively?
Although people rarely adhere to normative economic theories, their behavior can be thought of optimally balancing latent costs and values according to cognitive computatinos. Our lab builds models of cognitive computations that underlie human intelligence toward a dula goal -- using AI to understand cognition, and using cognition to advance AI. To achieve this, I we build algorithms that reverse-engineer human thought, grounding them in theories of cognition, neuroscience, and behavioural experimentation.
The lab explores questions such as:
– How do people plan, navigate, and explore in the real world?
– How do children and adults discover complex causal rules?
– How can psychological methods help interpret and guide emerging cognitive abilities in AI?
- How can humans and machines work together to invent nwe concepts and tools?
Our lab conducts interdisciplinary research combining methods from AI, neuroscience, and psychology. The lab connects a highly collaborative group of researchers at various stages of their careers and actively engages in national and international collaborations. Our collaborators include experts in artificial intelligence, neuroscience, cognitive science, and psychology.
We have funded PhD positions that explore different aspects of this research, for example, but not limited to:bExisting cognitive models of planning (e.g., in games like chess) tend to pre-specify possible planning models as anchored to classic algorithms, such as MDP solvers and stochastic search. In contrast, people likely maintain and learn an evolving library of planning strategies encoded as mental programs, and grow this library through experience. In this project, we explore approaches to modeling how such libraries of programs evolve and grow through social interaction and experience. We will work in collaboration with cognitive and developmental psychologists to build AI that can grow and learn like a child.
Current clinical assessments in psychiatry rely on subjective, survey-based reports, prone to bias -- underscoring the urgent need for formal, quantitative tools for measuring depressive symptoms. Our project will focus on interpreting computational causal mechanisms of anhedonia — a core symptom of depression affecting motivation, social interaction and valuation of rewards.
As humans and artificial agents can both exhibit systematic failures of prospection and disposition inference, we will apply our computational frameworks to modelling responses of foundation models, toward a novel computational framework for principled comparison of failure modes in humans and machines. Using this framework we will examine whether misestimated reward or misaligned social dispositions in foundation models parallels cognitive failure seen in humans. Our results will inform targeted therapeutic intervention, and bring insight into improving AI safety post-training.
Contact us to express interest in graduate studies or getting involved in research. Applications for directed study and honors projects are welcome at any time.
Please include in your email:
[under construction]
[under construction]