MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)

Happy Friday! We have you covered with all things CSAIL below. This week: a chat with Polina Golland, advances in medicine, and news in graphics & AI. 🩻

M.C. Escher's artwork famously features depth-defying optical illusions. New MIT tool creates multi-dimensional representations of these "impossible objects" that users can relight, smooth, & solve geometry problems on w/o bending or cutting shapes: https://bit.ly/4oilDuZ

What if doctors could measure what's hidden in a heart scan? 🩻 MIT CSAIL Professor Polina Golland’s research group is using AI to convert medical images into quantifiable data, helping physicians assess heart failure severity, compare treatments, and design better clinical trials. https://bit.ly/45ur55h

I had the honor of giving the keynote at the Mens, Manus and Machina (M3) Symposium on Embodied and Scalable AI, co-hosted by our M3S program and Singapore’s Infocomm Media Development Authority (IMDA). This symposium was part of SMART (Singapore-MIT Alliance for Research & Technology), an enduring research collaboration between Massachusetts Institute of Technology and National Research Foundation Singapore. The event brought together leaders in robotics, AI, materials science, and systems engineering to explore one of the most compelling challenges in the field today: Embodied Physical AI, that is, AI systems that understand the physical world, operate within it, and learn from it. In my talk on Embodied Physical AI, I emphasized that embodiment is an opportunity for advancing intelligence itself, but it requires a fundamentally different approach to AI. Specifically, embodiment requires: 1) AI that can reason about physics, task constraints, and common sense.  2) AI that runs on-device and cannot make critical errors.  3) AI that is energy-efficient. I highlighted Liquid Neural Networks developed by our team at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and Liquid Foundation Models developed by our team at Liquid AI as one promising direction toward Physical AI that meets these requirements for embodiment. I’m grateful to Prof. Jinhua Zhao and Prof. Archan Misra, with whom I have the pleasure of co-leading the M3 program, for their vision and leadership in shaping this event and the broader research agenda. My thanks also go to our M3 team and to our collaborators across IMDA, SMART, and the National Research Foundation of Singapore for advancing the science of embodied, adaptive, and collaborative AI. As we move forward, the future of AI will be embodied, it will live in the world, learn from it, adapt to it, and partner with us through shared physical understanding.

In 2016, Turing award winner Geoffrey Hinton famously said: "We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists." About a decade on, we still have lots of radiologists. So why didn't the job disappear? I asked MIT's Polina Golland, a computer scientist who works on AI and radiology. And she says: we may be getting this job displacement thesis all wrong (and not just in radiology). I found our conversation really thought-provoking: https://lnkd.in/gDfvrXC4 Massachusetts Institute of Technology MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)

What are open-weights AI models & why are they important? An excerpt from a recent CNET article: "Arguably, the most valuable thing in large [language] models is actually the weights. You can do a lot if you have the weights, which is somewhat different from traditional software," says Omar Khattab, an MIT assistant professor & CSAIL principal investigator. "In open-weights [models], you get the weights, which are these numbers, & you get how to map those weights into the neural network structure, so the layers of the neural network, in order to actually be able to run it." The architecture of the model shows how a company structures its models. Open-weights models are primarily aimed at developers, who can integrate the model into existing projects, like helping to build AI agents. More from Katelyn Chedraoui's story: https://lnkd.in/eHTFGn_s

Lung cancer is the leading cause of cancer-related deaths worldwide. A deep learning model developed by MIT & MGH researchers called "Sybil" can accurately predict a patient’s risk of lung cancer up to 6 years in advance by analyzing a low-dose CT scan: https://bit.ly/45bcHyA

Tweet tweet! We're looking for you go-getters! Our EARLYBIRD sale is now live, and for a limited time you can save 17% on select courses and programs. This is our biggest discount of the year and you won't want to miss it! Upskill this fall with MIT xPRO -- what are you waiting for? Browse the catalog today. https://hubs.ly/Q03yJSVL0

How can we make AI education and opportunities accessible for everyone? That was the pressing question at MIT Universal AI Summit earlier this summer. Co-hosted by MIT Open Learning and StartSmart South Eastern Europe in Athens, Greece, the summit explored AI’s transformative potential for education, business, and government. We were excited to preview Universal AI, Open Learning’s online learning experience currently in development. AI can be a powerful tool for good rather than an existential threat, said MIT Sloan School of Management Prof. Dimitris Bertsimas, vice provost for open learning, and MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) Prof. Manolis Kellis. Explore the key takeaways: https://bit.ly/4oiOxeg

NEW PODCAST! 🚨 How AI is Reshaping Medical Imagery with MIT CSAIL Professor Polina Golland Hear Professor Golland’s thoughts on healthcare, AI, and the future of diagnostics in this exciting glimpse into how AI impacts medicine, both now and going forward. Listen here: https://bit.ly/4mqV19c