Lesson 2: Why Physical AI Is Harder to Deploy Than Virtual AI | Daphne Koller

Before the turn of the 21st century, science fiction speculated that we’d all be getting around with the aid of flying cars or jetpacks by now — but, as you’ve probably noticed, we aren’t quite there yet. The slow pace of some innovations (e.g., robot cars) can frustrate expectations shaped by rapid advances in the virtual world. But insitro CEO and AI expert Daphne Koller stresses that the pace of AI’s impact in the physical world is a difference, not a hindrance. In this video lesson, Koller discusses the reasons the rates at which AI can make meaningful change vary, and why it’s smart to adjust our expectations accordingly. The physical world has less training data. People have already put massive amounts of text and images on the web. But to collect data in the physical world, you have to interact with it. The physical world has a slower feedback loop. If someone clicks a virtual ad, you can collect that data instantaneously. But in the physical world, you have to wait for it to respond. You can do more damage in the physical world. Injecting the wrong drug can kill people. You have to be more careful. Humans mostly live in the physical world. If you can use AI to meaningfully improve it, the potential for impact is much larger. © Freethink Media Inc., All Rights Reserved. ---------------------------------------------------------------------------------- Go Deeper with Big Think: ►Become a Big Think Youtube Member Get exclusive classes and early, ad-free access to new releases without leaving Youtube. https://www.youtube.com/@bigthink/membership/ ►Become a Big Think Web Member Get the entire Big Think Class library, premium print issues, live events, and more. https://bigthink.com/membership/ ►Subscribe to Big Think on Substack Get all of your favorite Big Think content delivered to your inbox. https://bigthinkmedia.substack.com/subscribe/ ---------------------------------------------------------------------------------- About Daphne Koller: Daphne Koller is the founder and CEO of insitro, where machine learning is applied to drug discovery and development. She also co-founded the digital learning platforms Coursera and Engageli. Koller spent 18 years as a professor of computer science at Stanford University, earning the Stanford Medal for Excellence in Fostering Undergraduate Research, and now serves as an adjunct faculty member. She was previously chief computing officer at Calico. She was named a MacArthur Fellow in 2004. She's also been inducted into the National Academy of Engineering, the American Academy of Arts and Sciences, the International Society of Computational Biology, and the National Academy of Sciences. She has authored more than 300 peer-reviewed publications in journals including Science and Cell and was named one of TIME’s 100 most influential people in AI in 2024.