Back to main page

Notes from CogSci 2026

48th Annual Conference of the Cognitive Science Society · Rio de Janeiro, Brazil · July 2026

Two caveats before you dive in. First, these notes cover only a small slice of the conference: what you see here is just the parts that were most interesting for me. Second, I took these notes live (or wrote them up shortly after from memory), so there's a decent chance I misheard a name or got a number slightly off somewhere.

Workshop

Closing panel: does anything actually solve the frame problem?

Framing the Problem workshop — normative vs. process models, and whether LLMs count

Talks

Interactive, multi-turn evaluation of language models

Robert D. Hawkins — reference games, politeness, and a negotiation task that exposes grounding failures

Why cognitive-science-inspired benchmarks keep failing us

Tomer Ullman — the Ali and Kogi story, the chocolate-bag task, illusions and cartoons

Do multilingual models and brains share a format?

Andrea Gregor de Varda — shared units across languages, and a 21-language brain-alignment study

Poverty, income instability, and child cognitive development

Julia Hermida — deficit vs. adaptation, the maternal-migration reversal, rural vs. urban Argentina

A resource-rational model of memory-guided deliberation

Marcelo Mattar — eye tracking, "looking at nothing," and a meta-RL model of memory sampling

Poster follow-up: the memory-sampling model, in more depth

Jonathan Nicholas — the same project, a more technical Q&A

What working memory limits mean for planning

Fred Calloway — decision trees, forgetting, and whether representation is just attention

Prize Lectures

Glushko Prize lecture: rethinking scientific discovery

Marina Dubova — simulated scientist communities, and "excess capacity learning"

Rumelhart Prize lecture

Douglas Medin — a tribute to Rumelhart, then diversity-based reasoning and psychological distance from nature

Lila Gleitman Prize lecture: cognitive tools

Judy Fan — number lines, sketches, data visualization, and building things together

Symposium

Large language models and personal memory

Paul Smart — encoding, elaboration, retrieval, joint reminiscing, and hallucination as a possible virtue