Tom Mitchell is a Digital Fellow at the Stanford Digital Economy Lab and Founders University Professor at Carnegie Mellon University, where he founded the world’s first machine learning department. He is also the author of Machine Learning, a foundational textbook in the field.
Machine learning: How did we get here? is his new podcast, featuring a series of conversations about the history of machine learning with Nobel Prize winners, groundbreaking researchers, and industry leaders.
How would you describe your background in machine learning?
I’ve been researching machine learning since I was a PhD student at Stanford University in the 1970s, inventing new approaches and applying machine learning to a variety of problems. Two of my favorite applications apply this to brain imaging data (to learn how to decipher which nouns you’re thinking from fMRI brain images) and to online education (an algorithm was able to learn which hints to give to students who got stuck solving a problem, and now it’s giving hints to millions of students online!).
What is it about now that finally makes it a good time to do a podcast?
Many people hear about machine learning in the news and are interested in it. We hope that the podcast can shed some light on the mysteries of machine learning and show the human faces of the people who have made significant contributions.
What do you think your podcast audience is: long-time researchers, people who know a lot about AI but want to learn more, or both?
I thought about this question a lot and tried to build an episode that anyone could understand, but there were still some surprises for professionals working in AI.
Is there a particular moment, story, or detail from the conversation that stands out to you?
There are many memorable quotes and anecdotes from the conversations in the episodes, but what stands out to me is his personality. Interviews reveal their passion for what they do, their curiosity, and their humanity.
You could tell that they really enjoyed spending hours, months, and years on the problem of how to make machines learn. And you can enjoy the camaraderie of the research community. Everyone participated in this study together and openly shared their findings.
The opening lecture (Episode 1) discusses how questioning authority has been an important theme throughout the history of machine learning. Could you explain that a little in the context of your conversation?
It is important to ask what we can learn from the history of machine learning. I think one of the most important lessons is that great progress often comes from people questioning the conventional wisdom of the field and even redefining problems and questioning authority.
For example, one of the podcast episodes features Dean Pomerleau, who was interested in self-driving cars in the 1980s. Conventional wisdom at the time was that driving computer programs should be built manually, and all researchers did so. But Dean, a PhD student, said no, so I’m going to try training a neural network to do it. He did just that, and ultimately achieved results that far exceeded the state-of-the-art technology of the time and changed the direction of research on the problem.
Another podcast episode features Kai-Fu Li, a PhD student working on speech recognition who decides to take a very novel machine learning approach by applying hidden Markov models. Once again, he changed the direction of the field. I think one of the most dangerous things a beginner in this field can do is blindly accept a common problem definition or paradigm. Doing so deprives us of the opportunity to reconsider what is possible.
