Michael I. Jordan: A Berkeley professor and one of the most influential figures in machine learning, statistics, and AI, cited over 170,000 times and mentor to many leading researchers in the field.
What this episode covers
Michael I. Jordan reframes today's AI not as a quest for human-like intelligence but as an emerging engineering discipline akin to how chemical and electrical engineering grew from chemistry and electromagnetism. He argues we understand almost nothing about the human brain and that the real frontier is building large-scale systems that make consequential decisions under uncertainty, especially markets that directly connect producers and consumers. He critiques the advertising-driven business models of Google and Facebook, advocating instead for transparent producer-consumer markets that create jobs and human happiness. The conversation also dives deep into optimization, gradients, stochasticity, and the Bayesian-versus-frequentist divide in statistics, before closing on advice for young researchers and his love of languages.
Also referenced (named, not recommended)
Alexa
Amazon
“alexa done right you know could do things like i leave the water running in my garden and i say hey alexa”— Michael I. Jordan
TensorFlow
“young people getting in this field who kind of think it's all done because we have you know tensorflow well no these are all open problems”— Michael I. Jordan
Big reveals from this episode
- Jordan flatly disagrees with Elon Musk on brain-computer interfaces, saying real brain-to-computer understanding is 'not even for the century.'
- He insists we are 'even worse than the Greeks' at understanding how the brain computes, calling it a task of the next few centuries.
- He wants a new word entirely instead of 'AI,' calling the term a 'terrible choice' that led to false promises.
- He proposes a 'music market' thought experiment where AI creates jobs by directly connecting creators to fans rather than propping up a few superstars.
- He argues the core problem with Facebook and Google is the advertising model itself, and that fake news 'rides on top of that.'
- He finds it 'creepy' that Facebook could know he's going to India from a conversation, contrasting it with trusting Microsoft.
- He claims a market that feeds a city is itself a form of intelligence, robust and adaptive like a brain, and that we must move beyond human intelligence.
Worth remembering
- Yann LeCun calls Jordan 'the Miles Davis of machine learning' for periodically reinventing himself.
- Jordan compares current AI to chemical engineering emerging from chemistry in the 1930s and 40s.
- He refuses to trust a neural net to predict his heart attack, insisting on 'what if' questions and dialogue with a doctor.
- He defends advertising as a legitimate market signal that a company believes enough in its product to spend real money on it.
- He notes United Masters, where he sits on the board, has signed 100,000 artists and put their music on the NBA website.
- He explains why naive 'biggest bang for your buck' intuition about gradients is mathematically wrong.
- He calls Nesterov's acceleration, which achieves a one-over-k-squared rate, one of the deepest and most surprising ideas in optimization.
- Statistics was originally called 'inverse probability,' and Laplace named it 'statistics' for the study of data for the state.
- He stresses the field is deeply international and cooperative, at odds with current-era nationalism.
- Jordan speaks French and Italian, learned French from boredom reading books in Kansas, and considers Italian the most beautiful language.