MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)

It was a real pleasure to welcome Richard Sutton to MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) for a visit that included time with faculty and students, a thoughtful talk, and a wide-ranging discussion about a deep topic in AI: experience as the core of intelligence. His work has shaped how many of us think about learning, agents, and the long-term foundations of AI. What made the conversation especially compelling was the reminder that intelligence is about experience, interaction, adaptation, and learning from the world over time. Thank you, Rich, for an inspiring visit and for a conversation that will stay with us. #AI #ReinforcementLearning #Intelligence #CSAIL #MIT #MachineLearning

Congratulation to MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) Led by the visionary Daniela Rus who produced an outstanding Entrepreneur day! Kicked off by Dean Daniel Huttenlocher Included: Keynote fireside: “My path from CSAIL to startup.” 3 CSAIL Success Stories - Ramin Hasani, Hari Balakrishnan and Wojciech Matusik; From Paper to Product: Commercializing Deep Tech": what to spin out, IP/licensing, early customers, grant vs VC vs corporate funding; to “What VCs Look for in CSAIL born Companies”: term sheet basics, deep tech vs SaaS expectations, timelines, and team composition; to Founder lightning talks: “5 minute startup stories” And the part I (John Werner)hosted dozen+ ventures (many in stealth) presenting 5 minutes pitches. Range of topics: - AI generates more vulnerable code than ever, and exploitation just got cheap with the release of Mythos. Forge combines sound program analysis (proving the absence of bugs) with Claude-driven fixes to guarantee that code is free of large classes of security bugs. - AI-driven system for designing task-specific robots from natural-language task descriptions. By combining LLM agents, simulation-based optimization, and modular hardware, the system rapidly generates robotic mechanisms tailored to specific physical tasks. The first application focuses on automated pipetting for wet-lab workflows, demonstrating a path toward faster, more affordable, and more scalable robotic automation. - Agents Aren’t the Problem. Your Infrastructure Is. Preparing Enterprises to deploy agents in scale - Core technology (7 years of MIT research) automatically specializes transistor models to foundry PDK parameters, accelerating simulations by up to 100x. It drops into existing workflows with zero changes to current simulators. - AI agents learn by watching humans perform their workflows. Take a quick look at the history bringing us here. - Reliability-as-a-service for AI traffic: we give agents an action layer for any site. Why loneliness in seniors is a massive health epidemic that needs to be solved NOW! - Governance middleware for AI agent fleets, inspired by 50 years of maritime watch law. One environment variable enforces work/rest cycles, auto-compresses agent context, and routes tasks by skill — cutting token costs ~88% while producing the audit trail regulated industries need. - A look into how we are reimagining the physical world as a programmable layer—where facilities, infrastructure, and environments can be queried, reasoned over, and automated just like data systems. - Personalized cancer screening and early detection - Custom AI intelligent routing for enterpeises. Michael Serrano Aditya Bansal Kyu Bum Kim Ajay Brahmakshatriya Nathan Cloos Michael Wang Rashad Haque Kanika Rajput Jonathan Roman Kiro Moussa Richa Gupta Samuel Blouir Nicholas Abate Nyari Nain

Congratulations to Makram Chahine on a remarkable PhD defense. Makram has been a PhD student in my research group at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). His thesis develops an elegant and ambitious dynamical-systems framework for Physical AI. This work is advancing selectivity, grounding, and compression for embodied intelligence in ways that are both theoretically principled and practically impactful. From robust navigation and human-robot co-autonomy to edge-efficient models and autonomous wildlife monitoring, this is a deeply impressive body of work. Bravo, Makram.

I was honored to participate in the joint Massachusetts Institute of Technology - ACADEMIE NATIONALE DE MEDECINE DE FRANCE symposium on AI in Medicine, organized with Bernard Nordlinger, and especially pleased to do so as a foreign member of the Académie. Together with Elazer Edelman from MIT Institute for Medical Engineering and Science (IMES), Polina Golland and me from MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), Pr Pierre Philip from the University of Bordeaux , Etienne Minvielle from École Polytechnique, and Bernard Nordlinger from the Académie Nationale de Médecine, we explored an important and consequential question: how AI can help reshape the future of medicine. One especially valuable aspect of the symposium was the opportunity to reflect how AI in medicine is evolving on the two sides of the Atlantic. In my presentation, I explained several core AI approaches and their implications for health, while highlighting MIT CSAIL projects in imaging-based diagnosis and treatment, AI in surgery, the role of sleep in disease, and the future of digital twins for better and more personalized care. The Académie Nationale de Médecine was founded in 1820. It stands in a long tradition of scientific and medical thought, shaped by figures of exceptional consequence, including Louis Pasteur, Victor Babes,  and  Marie Curie, who became its first woman member in 1922. What continues to distinguish the Académie today is its enduring culture of independence and intellectual depth. It is becoming increasingly clear is that AI has the potential to transform medicine at multiple levels: from detecting disease earlier, to tailoring therapies more precisely, to enabling a more continuous and individualized understanding of health over time. The most exciting future is one in which AI helps medicine become more powerful, more preventive, more personalized, and more deeply centered on the needs of each patient. Realizing that future will require sustained partnership across disciplines and institutions. It will depend on advances in computation, clinical science, and systems design, and also on thoughtful leadership in how technologies are developed and deployed. I am deeply grateful to Bernard Nordlinger and all of the speakers and participants for making this such a substantive and forward-looking symposium. #AIinMedicine #HealthAI #PersonalizedMedicine #DigitalTwins #MedicalImaging #MIT #CSAIL #IMES #AcadémieNationaleDeMédecine