Cognitive Science Colloquium

Fall 2024

All meetings take place on Thursdays, 3.30-5.30 pm in HJ Patterson Building (HJP) room 2124, unless otherwise noted.

Note: some links may require you to open a new tab.

Sept 5 — Patrick Shafto (Math & Computer Science, Rutgers-Newark).

Title: Mathematical foundations for learning agents
Abstract: Learning agents, which include humans and (ideally) AI agents, take actions in the world and learn from the outcomes. I will present our recent efforts toward an integrated theory of learning agents, focusing on cooperative communication as an extended case study. I will close with interesting applications and implications. 

Sept 19 — Melissa Koenig (Institute for Child Development, University of Minnesota).

            NOTE: this meeting will be held HJP 2118 (next door to the usual room).

Title: Children’s Testimonial Learning: Early Judgments of Epistemic and Moral Agency

Abstract: Children’s testimonial learning involves a sophisticated conception of human agency. Developmental research on children’s testimonial learning opened a window onto a form of reasoning aimed at identifying the intentions of epistemic and moral agents. Children’s testimonial reasoning takes both a critical interest in the reliability of sources, as well as a cooperative view of another person and her acts of communication. In this talk, I will focus on the role that social groups play in children’s epistemic judgements and discuss ongoing work in our lab that capitalizes on the trust children place in parents, friends and teachers.

 

Oct 10 — David Yaden (Psychiatry & Behavioral Science, Hopkins Medical School).

 

               Title: What are psychedelic experiences like?

               Abstract: The question in the title can have at least two different meanings: 1) what do psychedelic experiences feel like and 2) what other experiences are similar to psychedelic experiences (e.g., psychosis, meditation, etc.)? In this talk, I will discuss current major characterizations and measures of psychedelic experiences and compare them to mental states that seem to share some similarities but are triggered through different means. I will also discuss the (mixed) evidence for metaphysical belief changes after psychedelic experiences, as well as emerging evidence on how pre-existing beliefs might help to shape psychedelic experiences. I will conclude by discussing studies involving the administration of psychedelic substances to individuals with different belief systems (e.g., religious, spiritual-but-not-religious, and atheistic), which may shed light on how certain worldviews might change the content and the risk/benefit profiles of these experiences.

 

Oct 24 — Bertram Malle (Cognitive, Linguistic, & Psych Sciences, Brown).

 

Title: A Path to Trustworthy Artificial Agents: The Core Requirement of Norms

Abstract: Much talk about trustworthy AI and robots centers on accuracy, transparency, and fairness. I introduce a broader framework in which human trust is grounded in both performance dimensions (e.g., capability, reliability) and moral dimensions. The moral dimensions include considerations of transparency, benevolence, and ethical integrity, of which fairness is just one aspect. I show that people conceptualize trust—toward both humans and machines—in this multi-dimensional way, and I propose that the moral dimensions are a requirement for norm competence. I clarify what such competence involves theoretically, provide empirical evidence of its properties, and sketch one path of implementing it computationally.

Nov 7 — Alexandra Rosati (Psychology, Michigan).

Title: The primate origins of complex cognition
Abstract: Human cognition is marked by especially high levels of cognitive control and flexible decision-making. What are the evolutionary roots of these cognitive traits? I will present research examining different aspects of value-based decision-making, executive function, and social decision-making capacities across primate species that vary in features of their natural history such as diet, social structure, and life history. I will use this data to test hypotheses about why different facets of ‘intelligent’ behavior emerge across species, as well as how these differences emerge in comparative development.

Nov 21 — Isabelle Dautriche (Psychology & Neuroscience, Aix-Marseille University).

Title: Language Foundations: Insights from Acquisition, Communication, Cognition and More.

Abstract: Human languages exhibit an incredible diversity, from the sounds they use to form words to the grammatical rules they follow. Despite this variation, languages share deep structural properties like compositionality and exhibit striking regularities, such as a preference for placing subjects before objects. How do children manage to quickly learn such diverse languages? Why do languages share certain properties while differing in others? Are some of these properties deeply rooted in our cognition? In my talk, I will provide an overview on how I address these questions through experimental data from infants and animals, computational modeling, and corpus data. I will offer some evidence that infants are remarkably adept learners, that communication shapes language, and that several properties of language can be explained by non-linguistic cognition shared with other species. 

Dec 5 — Sangeet Khemlani (US Naval Research Lab, Washington DC).

Title: Toward a unified process theory of human reasoning
Abstract: Cognitive scientists have proposed more than a dozen theories of syllogisms, deductions about simplified quantified descriptions of properties. Such theory proliferation reveals the science's failure to deliver a unified account of how humans think. A unified theory should explain why people are capable of reasoning accurately, and also why they often make predictable errors. It should also explain common mental processes that underlie human thinking and reasoning across many tasks and contexts. I describe efforts towards developing such a theory by introducing mReasoner, a computational cognitive process model that assumes that people interpret linguistic and perceptual information by constructing iconic mental simulations. For example, the system builds mental simulations to represent quantified assertions, so it can explain reasoning about syllogisms and related inferences. Its high-level algorithms implement abstract processes, and so it can also reason about causal, sentential, kinematic, and spatiotemporal information. And because mental simulations can be constructed from perceptual information, it's possible to use the system, not just as a cognitive model that explains laboratory data, but also as a representation engine that facilitates human-centric communicative tasks. I conclude by demonstrating how mReasoner can operate in the context of real-time, real-world human-machine teaming.