DeepSummary
In this episode, Sean Carroll interviews Leslie Valiant, a computer scientist known for his work on computational learning theory and the "Probably Approximately Correct" (PAC) model of learning from examples. They discuss Valiant's perspective on how machines and humans learn, and his new book "The Importance of Being Educable" which proposes "educability" as a key trait distinguishing humans from other species.
Valiant explains his definition of educability as the integration of three things: learning from experience (PAC learning), chaining together learned knowledge (reasoning), and absorbing explicit theories from others. He argues that while intelligence is hard to define, educability captures a crucial aspect of human capabilities that enabled the development of civilization.
The conversation covers topics like the efficiency of learning algorithms, extrapolating theories from data, the relationship between learning and evolution, and whether machines can be designed with educability similar to humans. Valiant also discusses potential risks and societal impacts as AI systems become more capable of learning and reasoning.
Key Episodes Takeaways
- Leslie Valiant proposes 'educability' as a key trait distinguishing humans, integrating learning from experience, reasoning by chaining knowledge, and absorbing explicit theories.
- The 'Probably Approximately Correct' (PAC) model formalizes efficient learning from examples, which underpins machine learning success but also has limitations.
- Valiant argues 'intelligence' is ill-defined, while 'educability' captures crucial aspects of human capabilities that enabled civilization.
- Both humans and machines face risks if highly 'educable' but taught harmful knowledge; ethical education is crucial.
- Valiant explores connections between computational learning theory and natural processes like evolution and human cognition.
- Developing AI systems with human-like educability may require integrating reasoning capabilities beyond pure learning from data.
- Valiant's work has influenced the theoretical foundations and experimental growth of machine learning as a field.
- Assessing and potentially enhancing human educability could lead to advancements in education and cognitive development.
Top Episodes Quotes
- “So the main downside of intelligence is that no one can define it, that it's, you know, of course people have. People have, of course, complaint that we give importance to intelligence. We test people for do intelligence tests and things. This has consequence, and we don't even know what we're testing for. What did the question, where the questions come from?“ by Leslie Valiant
- “So being educable has its great dangers as well. You can educate machines very easily, and also humans very easily, to do things you don't want done.“ by Leslie Valiant
- “And I think the take I have is that much of what we do is theory less, but it doesn't mean that it's not effective or predictable or predictable on the average, or useful, because we just so this learning process is in some sense robust and useful.“ by Leslie Valiant
Entities
Company
Person
Product
Book
Concept
Episode Information
Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas
Sean Carroll | Wondery
4/15/24
Science is enabled by the fact that the natural world exhibits predictability and regularity, at least to some extent. Scientists collect data about what happens in the world, then try to suggest "laws" that capture many phenomena in simple rules. A small irony is that, while we are looking for nice compact rules, there aren't really nice compact rules about how to go about doing that. Today's guest, Leslie Valiant, has been a pioneer in understanding how computers can and do learn things about the world. And in his new book, The Importance of Being Educable, he pinpoints this ability to learn new things as the crucial feature that distinguishes us as human beings. We talk about where that capability came from and what its role is as artificial intelligence becomes ever more prevalent.
Blog post with transcript: https://www.preposterousuniverse.com/podcast/2024/04/15/272-leslie-valiant-on-learning-and-educability-in-computers-and-people/
Support Mindscape on Patreon.
Leslie Valiant received his Ph.D. in computer science from Warwick University. He is currently the T. Jefferson Coolidge Professor of Computer Science and Applied Mathematics at Harvard University. He has been awarded a Guggenheim Fellowship, the Knuth Prize, and the Turing Award, and he is a member of the National Academy of Sciences as well as a Fellow of the Royal Society and the American Association for the Advancement of Science. He is the pioneer of "Probably Approximately Correct" learning, which he wrote about in a book of the same name.
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.