Please register for the seminar here: https://forms.cloud.microsoft/r/c6jgLrTNsx

Abstract:

 The AI community faces a fundamental crisis of understanding. While some researchers herald large language models as the dawn of artificial intelligence, others—including prominent voices like Yann LeCun—dismiss them as sophisticated autocomplete systems doomed to statistical mimicry. Both camps miss a crucial insight.

Our research resolves this debate by repositioning LLMs not as failed intelligence, but as powerful unconscious processors—high-capacity pattern repositories operating in what cognitive science calls System-1 mode (fast, automatic responses). This reframing unlocks a systematic pathway to artificial general intelligence through our UCCT-MACI framework, which predicts and controls the transition to deliberative System-2 reasoning (slow, conscious deliberation).

The breakthrough lies in three interconnected innovations. First, our United Cognitive Consciousness Theory (UCCT) provides the first quantitative “information dial” that predicts when systems shift from automatic responses to conscious deliberation. Second, our Multi-Agent Collaborative Intelligence (MACI) framework introduces a “behavior dial” that orchestrates regulated debate between AI agents, enabling them to anchor each other’s reasoning and escape the narrow constraints of their training data. Third, our SagaLLM and ALAS systems provide the missing ingredient: spatiotemporal continuity that enables persistent memory, evolving world models, and long-horizon planning.

Together, these components create something unprecedented: AI systems that can form perspectives, develop retrospectives, and engage in sustained reasoning trajectories across space and time—the hallmarks of genuine intelligence. Our comprehensive validation across planning, diagnosis, and policy domains demonstrates measurable improvements in reasoning depth, convergence speed, and system reliability.

This work charts a concrete roadmap from today’s pattern-matching systems to tomorrow’s general intelligence, offering both theoretical foundations and practical implementation strategies for the next phase of AI development. (This talk synthesizes insights from a comprehensive 17-chapter study on the path to AGI, forthcoming from ACM Books.)

Biography:

Edward Y. Chang is an adjunct professor in the Computer Science Department at Stanford University, where he has taught since 2019. He previously served as President of HTC DeepQ Healthcare (2012–2021) and as a Director of Research at Google (2006–2012), where he pioneered efforts in scalable machine learning, Web-scale image annotation (2008), data-centric AI (2010), and sponsored the ImageNet project. He was a visiting professor at UC Berkeley (2017–2020), working on surgical planning with virtual reality, and held a tenured faculty position in Electrical and Computer Engineering at UC Santa Barbara (1999–2006). Chang earned his M.S. in Computer Science and Ph.D. in Electrical Engineering from Stanford University. In addition to his technical training, he pursued over ten courses in philosophy and literature—an intellectual breadth reflected in his research and writing. He is a Fellow of both ACM and IEEE for contributions to scalable machine learning and AI for healthcare. His honors include the NSF CAREER Award, Google Innovation Award, the ACM SIGMM Test of Time Award, and the US$1 million XPRIZE for AI-driven disease diagnosis.