AI summary by TwinMind
Charles Isbell and Michael Littman: Machine Learning and Education | Lex Fridman Podcast #148
Lex Fridman
1h 57m 47s
Machine Learning vs Computational Statistics Debate
- Charles: ML is not just computational statistics - missing rules, software engineering aspects
- Michael: ML is computational statistics - statistics keeps you from lying to yourself
- Distinction between ICML (computer science approach) vs NIPS (impressing statisticians)
- Neural networks changing the landscape - hyper parameters, data collection/cleaning as programming
- Focus on data matters more than algorithms in ML - data reveals algorithm differences
- Statistics vs computationalist perspectives bring different valuable viewpoints
Educational Philosophy and Assignments
- Charles's ML assignments: steal code, focus on data analysis not implementation
- Students must find 2 interesting datasets that reveal algorithm differences
- Zero points for getting code to work - all points for data analysis and justification
- Students initially hate approach but love it later
- "Give them what they need without bending to their will"
Role of Struggle in Education
- Lex: hardship/struggle essential for deep learning and joy
- Charles: distinction between hopeless vs hopeful struggle
- Need to maintain hope while challenging students
- Georgia Tech vs Brown - different institutional approaches to student experience
- Georgia Tech: "building tomorrow the night before" ethos
- Drown proofing requirement at Georgia Tech (historical)
Meeting and Collaboration History
- Met at AT&T Labs during Michael's interview
- Charles had intimidating low chair setup
- Both worked on automated assistant project before layoffs
- Project eventually influenced CALO/Siri development under Ron Brockman at DARPA
- Maintained friendship through texting, Charles forced Michael to buy iPhone
- Collaborated on Georgia Tech online master's program
Bell Labs Experience
- Described as magical research environment - university without students
- Pure research funded by phone bill tax from AT&T monopoly
- Whiteboards everywhere, constant intellectual arguments
- Rich Sutton, Michael Kearns, Peter Stone, Satinder Singh all there
- Diaspora when labs closed - everyone went to great academic positions
- Represented high water mark of basic research in US
Online Education and MOOCs
- Georgia Tech online master's: 11,000 students, $6,600 total cost vs $46,000 on-campus
- 60% acceptance rate, high completion rates
- Students created own social groups, showed strong loyalty
- 80+ graduates return as TAs for minimum wage
- Teaching together provided energy/connection missing from solo online teaching
- "Smoove and Curly" personas developed organically
COVID and Future of Education
- Disaggregation of education from place vs experience
- Students pay for college experience, not just classes
- Physical presence matters for peer interaction, rite of passage
- Online education serves different population (working adults vs traditional undergrads)
- Technology will enable more dispersed experience but won't replace universities
- Need to expand access while maintaining mission focus
Faculty Role and University Structure
- Strange system: world-class researchers also teaching and doing service
- Lecturers teach 50% of credit hours with 10% of faculty
- Alumni remember teachers who made personal impact most
- Mission-driven approach unifies research and teaching
- Tying faculty incentives to institutional mission important
Programming and Computer Science Advice
- Start small, learn building blocks patiently
- Life is long enough to build big projects eventually
- Python recommended for beginners, though "only real language is Lisp"
- Basic concepts (variables, conditionals, loops) are genuinely difficult
- Assignment vs equality major conceptual hurdle
- Being comfortable with confusion/debugging essential skill
Personal Friendship and Collaboration
- Inner product of 0.7 - similar enough to learn from each other
- Michael: Charles shows him different world, alternative life experiences
- Charles: Michael believes in him when he doesn't believe in himself
- Mutual confidence and support through long friendship
- "Luck favors the Charles" - things work out well for Charles
