The session opened with a tribute to Rumelhart himself. Born in Springs, South Dakota, mother a town librarian, father a newspaper printer. He trained in math and psychology at the University of South Dakota and then at Stanford, was UCSD faculty from 1967 to 1987, and later moved back to Stanford until declining health led to his retirement in 1998 and death in 2000. A video tribute included recollections from Jay McClelland, who was initially rejected as a UCSD postdoc before being brought on, describing Rumelhart as a self-taught polymath and avid game player who used mathematics and symbolic computation before turning to neural networks in the late 1970s.
McClelland walked through the arc of Rumelhart's contributions. Early work treating concepts as points in a vector space, a direct ancestor of how modern embedding models place words in a vector space. Work with Don Norman on the symbolic paradigm and the nature of sentence meaning. A 1977 paper laying out the idea that people converge on an interpretation through multiple simultaneous constraints, an early statement of parallel distributed processing. And, with Hinton and Williams, the paper showing that backpropagation could train neural networks to learn useful representations beyond what's explicit in the data, described as arguably the single most important idea behind today's large language models. There was a quote from Hinton relayed in the video too: had Rumelhart lived to see where this ended up, he'd likely be the one most credited for it.
Rob Goldstone then introduced this year's recipient. On paper, a fairly typical path for a cognitive psychologist of his generation. Undergraduate degree in 1965, PhD from the University of South Dakota in 1968, postdoc and faculty positions eventually leading to Northwestern, editor of Cognitive Psychology, co-author of a widely used textbook. But also, notably, someone who collected soil samples in Guatemalan rainforests as part of his research, who has long doubted that findings from undergraduates generalize to human cognition broadly, and who has repeatedly revised his own theoretical commitments rather than defending them. He argued in Categories and Concepts, with Ed Smith, that concepts are fuzzy and resemblance-based rather than rule-based. Then he developed the exemplar-based context model against the prototype-theory consensus that followed. Then he developed "theory theory" with Greg Murphy once he decided similarity alone couldn't explain why a set of features counts as a meaningful concept in the first place.
Medin opened with two puzzles about categorization. The first: anthropologists and psychologists (Brent Berlin and Eleanor Rosch, working independently) both converged on the idea that natural categories have a privileged "basic level" that just stands out perceptually. But in his own fieldwork, participants' basic level for trees didn't line up cleanly with that prediction, and it varied by community and expertise.
The second, more striking puzzle concerned the diversity principle, the standard assumption that a taxonomically diverse premise (a disease affecting both river birch and paper birch) should generalize more strongly across a category than a less diverse one. He tested this with University of Michigan undergraduates and with Itzaj Maya participants in lowland Guatemala using local trees and animals. Michigan undergraduates followed the diversity principle about 96% of the time. Itzaj Maya participants were reliably below chance, the opposite pattern, and it replicated across follow-up studies. The same "failure" showed up in a very different population too: U.S. park and forestry workers, who reasoned instead about actual ecological proximity, since species growing near each other share disease-transmission opportunities regardless of taxonomic distance. It wasn't that these groups reasoned worse. They were reasoning about a different, and arguably more realistic, kind of relationship that a pure taxonomy-based model completely misses.
The second half of the lecture covered a long-running collaboration between Northwestern, the American Indian Center of Chicago, and the Menominee Nation in Wisconsin, a community historically known for sustainable forestry, visible as a distinctly greener patch on satellite imagery at the county level, and for whom sturgeon are a sacred species.
Interviews with Menominee parents and grandparents about what they wanted children to learn revealed a consistently embedded, "part of nature" orientation. That contrasted with a more distanced, stewardship-oriented framing among European-American respondents in the same region ("I want them to take responsibility for nature," implying separation rather than membership). This mapped onto construal-level theory from social psychology, where close orientations track attention to context and relationships, and distant orientations track abstraction.
Several measures converged on the same pattern. Children's books recommended by Native literacy organizations were far more likely to use close-up, first-person, or non-human points of view than mainstream bestsellers (84% versus 41%). Picture-description tasks with seven-year-olds showed no group differences in taxonomic language but large differences in mentions of ecological relationships, and a striking gap in spontaneous animal-sound imitation (41% versus 0%). A forest-diorama task, designed at the suggestion of a Menominee research assistant who insisted decontextualized plastic animals wouldn't work, found Native children roughly twice as likely to spontaneously narrate an animal's own perspective. And fish-expert interviews found European-American anglers sorting fish taxonomically and treating fishing partly as competitive sport, while Menominee experts organized knowledge around reciprocal, life-cycle-based ecological relationships between species. That said, the same ecological richness could be drawn out of European-American experts too under slower, more open-ended questioning.
He was careful not to over-claim. "Psychological distance" alone, he said, doesn't capture the full content of these cultural models any more than psychological-distance manipulations alone would explain what makes someone good at basketball. The deeper story is in the specific content of ecological relationships and social models, not simply how close or far people feel from nature. He closed by connecting this to survey data showing widespread endorsement of statements implying humans are separate from and harmful to ecosystems, a model that, in his view, tends to cast the ideal human role as simply getting out of the way, rather than one of active, positive participation in the systems we're part of.
This was a long lecture covering decades of work, plus the tribute beforehand, so some of the percentages and study details are from memory and could be slightly off.