Home Lex Fridman Notes
Lex Fridman · 2020-12-26 · 1h 57m

Charles Isbell and Michael Littman: Machine Learning and Education | Lex Fridman Podcast #148

Two longtime friends and machine learning professors riff on what ML really is, the meaning of education, and life advice.

Charles Isbell and Michael Littman: Machine Learning and Education | Lex Fridman Podcast #148
The guest

Charles Isbell and Michael Littman — Charles Isbell is dean of the College of Computing at Georgia Tech and Michael Littman is a computer science professor at Brown; the two are longtime friends who co-taught a hugely popular online machine learning course.

The gist

Lex Fridman hosts his first two-guest episode with friends Charles Isbell and Michael Littman, who met at Bell Labs/AT&T Labs before the famous research diaspora. They debate whether machine learning is 'just computational statistics' (concluding it is fundamentally a computer-science and data-centric discipline), and dig into teaching philosophy, the role of struggle versus suffering in learning, and what online education (MOOCs) and COVID reveal about why people really go to college. The conversation ranges through Westworld, simulation theory, video games, and closes with heartfelt advice about pursuing passion and the value of friendship.

Big reveals

  • The pair claim they agree on everything, then immediately stage their long-running debate over whether machine learning is just computational statistics.
  • Charles reveals his ML assignments give zero points for working code, telling students to 'steal the code' and focus entirely on the data and analysis.
  • They got laid off from AT&T Labs but their automated-calendar-assistant project helped birth the DARPA program that became CALO and ultimately Siri.
  • Charles argues the real danger of AI is not superintelligence but using data to make terrible everyday decisions more efficiently.
  • Charles contends people pay for college for the 'college experience' and rite of passage, not the education itself.
  • Charles alludes to a police encounter he 'wasn't supposed to survive,' framing his outlook that life is all gravy.

Things worth remembering

  • Charles and Michael's co-taught machine learning course has reached over 10,000 students.
  • Lex argues education should be hard, that struggle is what increases the joy of finally learning something.
  • Georgia Tech historically required a 'drown proofing' class to graduate, ending around 1983-84.
  • Bell Labs was effectively funded by a tax on phone bills, letting it do pure research and invent the transistor and laser.
  • Economists can trace fewer young men in paying jobs partly to time spent on video games.
  • Georgia Tech's online master's degree costs about $6,600-7,000 versus $46,000 on campus, with an ~60% acceptance rate and 11,000 students.
  • Lecturers teach over 50% of credit hours in Georgia Tech's College of Computing despite being far fewer than tenure-track faculty.
  • Michael notes all programming reduces to three things: reading a variable, writing a variable, and conditional branching.
  • Michael fixed a memory-overwrite bug in his master's thesis code by declaring a 400K dummy array to absorb the stray writes.

Recommended in this episode

Books, products and media the guest or host genuinely endorsed here — with the buy link.

Affiliate link — we may earn a commission at no extra cost to you.

RecommendedBook

Calculating God

Robert Sawyer

“you really need to be calculating god by sawyer it's a you'll read it in a night it's a very easy read” — guest 01:04:37
Find it on Amazon