Results 341 - 350 of 24331
Hats off to the many members of the worldwide Simons Institute community who have been making a splash in the last weeks.
In his keynote address at the workshop on Governance at the Technological Frontier: Translating Research into Policy for AI Oversight, California State Senator Jerry McNerney spoke about prospects and challenges for enacting meaningful AI regulation via the legislative process. Senator McNerney holds a PhD in Mathematics.
Not all convex functions have finite minimizers; some can only be minimized by a sequence as it heads to infinity. In this talk from our recent workshop on The Role of TCS in Modern Machine Learning, Robert Schapire (Microsoft Research) presented joint work with Miro Dudík and Matus Telgarsky aiming to develop a theory for understanding such minimizers at infinity.
Recent advances in vision and vision-language models have renewed interest in a fundamental question: do these systems acquire internal models of the world? Although such models are trained on images, videos, and other human-produced representations rather than direct experience, they often exhibit behaviors suggestive of knowledge about objects, spaces, and dynamics. Determining what these behaviors reveal about underlying world models, however, remains challenging.
In this talk, I will discuss how Kenneth Craik's conception of mental models can provide a useful lens for thinking about this question. Craik argued that internal models enable organisms to reason about alternatives, anticipate future events, and apply past experience to new situations. Building on this perspective, I will examine two recent lines of work that probe these capabilities in modern AI systems: spatial reasoning in language models and motion forecasting in visual models. Together, these examples illustrate both the promise and the limitations of current approaches to studying world models in artificial systems.
World models are now a central paradigm in artificial intelligence, providing latent dynamical structures that support prediction, planning, and counterfactual reasoning. Yet prevailing formulations remain largely individualistic: agents model environments, but rarely other model-bearing agents in a principled way. In contrast, biological intelligence is intrinsically relational. Neural systems are embodied, socially coupled, and dynamically co-regulated across multiple timescales.
In this talk, I argue for a reformulation of world modeling through the framework of Social NeuroAI. Drawing on dynamical systems theory, inter-brain neuroscience, and multi-agent computational modeling, I show how intelligence can be understood as emerging from coupled embodied dynamics rather than isolated inference. I will present recent work on embodied neural agents demonstrating how metastable coordination and collective decision making arise from structured neural coupling, and discuss how these mechanisms motivate a shift toward relational computation in AI architectures.
I propose that next-generation world models should treat other agents as dynamical systems with their own latent models, enabling collective intelligence to emerge through interaction. This perspective connects embodied dynamics, social inference, and generative world modeling within a unified computational framework. The goal is to move from solitary simulators to socially embedded systems whose intelligence is constructed through coordination, alignment, and shared representation.
