Bio
Marta Kryven is an Assistant Professor in Computer Science, Psychology and Neuroscience, at Dalhousie University. She directs the Computation and Cognition Lab, which studies how humans and machines construct models of the world, come up with explanations, and plan. Her work has appeared in NeurIPS, ICLR, and Nature, including a NeurIPS 2025 Spotlight paper, and research recognized in Nature Computational Science's annual highlights.
She serves as Conference Co-Chair of CogSci 2027, Proceedings Committee of CCN 2027, Scientific Expert for Creative Destruction Lab, as an Area Chair for ICML 2027, and founder and organizer of Computation and Cognition Symposium. She is also the Founder and Chief Scientist of Northlight Intelligence, where she applies research insights to practical challenges in AI adoption and human–AI decision-making.
Talk abstracts
How Minds Build Models of the World
Target General academic audience across cognitive science, AI, neuroscience.
Humans can learn from remarkably little experience, solve unfamiliar problems, and navigate novel complex environments on their first exposure. Which mental representations make this possible?
This talk explores how people construct and use models of the world to support learning and reasoning. Drawing on research in cognitive maps, compositional abstraction, and causal learning, I will discuss computational mechanisms through which intelligent agents discover structure, invent hypotheses, and reuse knowledge across problems. I will show how behavioral experiments and computational models can help identify the principles underlying these abilities, and discuss what they suggest about the design of artificial systems capable of learning and reasoning more flexibly.
Inventing Explanations: How Humans Learn from Uncertain Evidence
Target General academic audience across cognitive science, AI, neuroscience and developmental psychology.
Much of what we know about the world must be inferred from incomplete, ambiguous, or unreliable evidence. Traditional computational models of learning often assume that the learner begins with a predefined set of possible explanations. But how do people discover explanations that they have never considered before? In this talk, I will present research investigating how children, adults and LLMs generate and evaluate causal hypotheses while learning about unfamiliar systems. What are the cognitive mechanisms by which we examine how learners formulate hypotheses, interpret uncertain evidence, and decide when to explore alternative explanations. I will also discuss comparisons between human-machine cognition, highlighting what these reveal about the computational mechanisms supporting learning in humans and machines, and its implications for rigorous model evaluation.
What Should We Delegate to AI?
Target Technology companies, business leaders, AI industry conferences, innovation organizations
Advanced AI can perform tasks that recently required substantial human expertise. Yet deploying these capabilities effectively remains surprisingly difficult. AI systems excel at some complex tasks while struggling with apparently simple problems involving ambiguous goals or incomplete information. How can organizations determine which decisions to automate and where human judgment remains essential?
In this talk, I draw on research in cognitive science and AI capabilities to examine the strengths and limitations of human and machine reasoning. I will introduce a framework for analyzing work in terms of its underlying cognitive demands and discuss practical examples of how organizations can allocate tasks between humans and AI, design appropriate oversight, and evaluate whether automation is improving outcomes. As AI systems increasingly integrate in our society, the greatest value of AI comes from understanding how different forms of intelligence can work together.
Computational Modelling for Non-Specialists
Target Researchers, graduate students, and professionals interested in computational approaches to human behavior.
How can we turn informal theories about human behavior into precise, testable computational models?
This introductory workshop explains what computational models are, why they are useful, and how they can advance scientific understanding. Using examples from human decision-making, participants learn how to translate verbal explanations into mathematical models, generate behavioral predictions, and evaluate competing theories against empirical evidence. The workshop also explores the relationship between model complexity, explanatory power, and falsifiability.