UChicago Knowledge Lab

🧠✨ Thrilled to share our latest AI work led by the brilliant Haizi Yu and Lav Varshney! We've developed a white-box AI model that learns like humans do - from just ONE example, no pretraining needed. (https://rdcu.be/ev44p) Our "distortable canvas" mimics how babies might natively perceive similarity (such as recognizing that a "1" and a crutch share the same shape). Using this cognitive insight + simple nearest-neighbor classification, we achieved: 80% MNIST accuracy with 1 example per class Near-human performance on complex visual tasks Human-interpretable learning process This represents a major step toward data-efficient, explainable AI that thinks more like we do. The model is completely transparent - you can see exactly how it transforms one image into another, following human intuition. #AI #MachineLearning #CognitiveScience #ExplainableAI #FewShotLearning #PriorDrivenModeling

I enjoyed an awesome discussion/podcast with the BigBrains team at UChicago about the future of AI (over a pair of sessions) -- Check it out (on the commute)! https://lnkd.in/gYEXHh5r

Why do scientists disagree? With the amazing Justin Sulik, NakWon Rim, Elizabeth Pontikes, and Gary Lupyan, we show in Nature Human Behaviour how cognitive traits and dispositions align with persistent scientific divides, using article text, references, surveys, and the same cognitive trait tests psychologists have long employed for subjects on psychologists themselves (https://lnkd.in/guFxsM2V). For example, we show that tolerance for ambiguity predicts a preference for context-rich explanations and dissatisfaction with the cognition-as-computer analogy, while need for cognitive structure correlates with a preference for universal rules and realism. These cognitive traits and dispositions are not ephemeral—they shape which research questions scientists pursue and which answers they find compelling, even when facing identical evidence. Analysis of 7,973 psychologists demonstrates these divides within psychological subfields and across psychological science. Scientific consensus isn't just about data—our cognitive architectures filter how we process evidence, bolstering persistent divides that resist resolution. #ScienceofScience #CognitiveScience #Conflict #Confusion #CollectiveIntelligence #CollectiveStupidity

I'm thrilled about this new work that demonstrates the existence of cultural world models within LLMs and how local cultural and linguistic knowledge is conserved and controllable within them. To be clear, this doesn't yet occur "for free", but this makes clear how intelligent soft-control of linguistic and cultural knowledge represents the next major target for native LLM customization for global users (https://lnkd.in/ehJQsJpP). With the amazing: Veniamin Veselovsky, Berke Argın, Benedikt Ströbl, Chris Wendler, Robert West, Tom Griffiths, and Arvind Narayanan

Thrilled to share our new article in Science: "Large AI models are cultural and social technologies" Working with the brilliant Henry Farrell, Alison Gopnik, and Cosma Shalizi, we challenge the dominance of the prevailing narrative about AI models as autonomous, intelligent agents. Our key insight: Today's large AI models function more like cultural and social technologies—akin to writing, print, markets, and bureaucracies. They enable humans to access and reorganize vast stores of accumulated human knowledge in novel ways. This reframing matters. By viewing AI through this historical lens, we can better anticipate how these technologies will transform society, economy, and knowledge production—just as earlier information technologies did. This is not a diminutive framing of AI but an expansive one. Like the cultural technology of the library, these tools can construct cultural world models. Unlike them, they enable users to generatively "spin-up" multiple, competing perspectives--performing synthesis and sociality--in novel ways, with implications for the future of collaboration, productivity, and institutions. Read the full article to explore how this perspective shifts conversations about AI risks, benefits, and governance: https://lnkd.in/geMF9xUP #ArtificialIntelligence #AIResearch #ScienceMag #ArtificialSociality