I really enjoyed this conversation with Michael. Here's the outline: 0:00 - Introduction 2:45 - Influence from literature and journalism 7:39 - Are most people good? 13:05 - Ethical algorithm 24:28 - Algorithmic fairness of groups vs individuals 33:36 - Fairness tradeoffs 46:29 - Facebook, social networks, and algorithmic ethics 58:04 - Machine learning 58:05 - Machine learning 59:19 - Algorithm that determines what is fair 1:01:25 - Computer scientists should think about ethics 1:05:59 - Algorithmic privacy 1:11:50 - Differential privacy 1:19:10 - Privacy by misinformation 1:22:31 - Privacy of data in society 1:27:49 - Game theory 1:29:40 - Nash equilibrium 1:30:35 - Machine learning and game theory 1:34:52 - Mutual assured destruction 1:36:56 - Algorithmic trading 1:44:09 - Pivotal moment in graduate school
Although this isn't one of the more highly viewed videos, I find equal if not more value from it. Michael Kearns comes off as a truly authentic and brilliant mind within the field and I'm thankful that you shared his insights with us.
Ha! Just found this. Was a grad student with Mike at Harvard, at the time he mentions at the end. Had a few fun times. Can't wait to give this a listen!
Michael, just have the same feeling as you shared in this interview, about your Bachelor's time in Berkeley and then your Master's in Harvard. Today I finished my Master's Capstone, and I walked through the aile. I thought that, "actually I only needed a Bachelor's to equip myself with the majority of the skillsets I am using in my daily job. Master's degree is for people who really want to do research." Berkeley was also the first place I visited in the US. But later you mentioned about "had most of the interaction with math people and computer science people". That was what made me happy when I first started my Master's as I felt I was with like-minded people. So, feelings change overtime when we are in different life stages. Thanks for the great sharing and Lex's great questions.
32:08 I would just say that protection against false incarceration isn't bout "fairness for the criminals", or it wouldn't be false incarceration. It's important to remember that not every defendant has committed a crime, and not every plaintiff is a victim.
Great interview....... you need to always put a low cut filter on the vocal tracks, like 18dB per octave at around 80 hz. Because there's a lot of sub bass plosives coming from the mics. And i hear lots of other sub bass rumbles coming from the building and from Michael banging on the table!........ But yeah, i do happen to be listening in a mastering studio LOL!
FWIW, my take: If someone hasn't embarrassed themself for a good long while, they haven't truly sought mastery or tried to attain difficult knowledge...I wouldn't even feel embarrassed about it, that's just the price of admission. I'm sure I don't have to say this, but gems can still be uncovered whilst 'embarassing yourself', and sometimes at the end of the journey, you'll pull some of those gems back out, seemingly serving no purpose until you see it all come together.
Well said, @mhill88ify Eric Weinstein articulated it even better IMO when he coined the term 'Embedded Growth Obligations' to describe how institutions built on a growth model (college professors looking for tenure teaching their students to become professors, etc.) have stagnated any original research in the past few decades.
@@ishansaraf1576 haha, 'embedded growth obligations' - I love it! We all should start taking this idea to heart. Being alive entails having 'embedded growth obligations"!.
This is excellent content, worth the entire almost 2 hour time. Really good discussion but sadly the points about algorithm operation (start at 49:00) lacks the issue of driving profits for the 'googles' and advertisers out there and how the algorithm optimizes for those two sides while claiming that the user's experience is "enhanced". However, the discussion redeems itself when Kearns adds a third side (start at 51:45) wherein the user adjusts the algorithm for the user's desires NOT just the other two sides. Examples of the former is when Netflix/Amazon Prime Video give me FIRST what videos they produced vs. what I really like to watch or a google search feeds back what maximizes ads and/or clicks based on my search. Sadly, I can't give an example of the latter representing that third side where I get what I want - precisely what I want. Again, an excellent content, worth the entire time.
Also: Lex, I would take a guess that your taking notes and selecting the best ideas from these great minds to create your own one true AI? 👍 Keep up the good work
How do you address that bias is everywhere (i.e. the background of Lex's guests) and leads to goal optimization and/or evolution (specific podcast content)? Evolution & goal optimization are often driven by human bias (learning about AI from highly educated guests). Is there consideration that by ending all human bias in algorithms could slow overall forward human progress and evolution?
For anyone interested in learning more about algorithmic fairness I recommend reading Game Theory and the Pursuit of Algorithmic Fairness by Jack Frostwell. It is a great book on the topic.
1:01:00 sure the computer shouldn't automatically decide what's fair, but it could find hard-to-spot cases to flag for human evaluation and ask "Hey, I found a discrepancy--is this fair?" Kind of like Chomsky's anarchism--systematically questioning power differences...
Great talk as usual and a wonderful guest. Very nuanced and gentle in its presentation of multi generational class warfare, given the political atmosphere in the USA this is the safe choice. Looking forward to googling more about the author.
I understand that algorithms can help in the performance of justice, the administration and delivery, in terms of accuracy and precision. It's this very increased performance that concerns me. Some law that is ten percent unjust is not so bad, but what if it's enforced with one hundred percent efficiency? Hard to decide on these matters. Do we really want to have, say, nearly perfect lie detectors on all government and medical devices?
At 11:12 Michael Kearns had an opportunity to acknowledge that academics are a group of people in a faulty culture normalized by its being a bubble, and then he used it to say outsiders don’t get it. Ironically this proves his earlier point that people acclimate to normalize their cultural bubbles.
All your podcasts with table-mics are of lower quality. Get them permanently off the table (even when offloc) - or add some sort of filter to prevent the constant thumps some of your guests are causing. I just finished watching your interview with Christof Koch and in ~57mins he must have banged the table 500+ times.
Made by Jimbob Nor is inequality. “Equality”, like “liberty” and “fraternity” (or human solidarity in modern, non-sexist terms) are the fundamental values of western enlightenment civilization. You need to give the sense in which inequality is not inherently ethical.... and why would you only pick on equality. Why not say the same about liberty or solidarity? Is the liberty of a slave owner ethical? Of course no
being motivated by power or money or control will produce different behaviors than being motivated by truth, beauty and the good. This man is naive to the motivations of the most powerful, the history of their legacies, and how what is Good can be ignored by those who merely want to win.
I really enjoyed this conversation with Michael. Here's the outline:
0:00 - Introduction
2:45 - Influence from literature and journalism
7:39 - Are most people good?
13:05 - Ethical algorithm
24:28 - Algorithmic fairness of groups vs individuals
33:36 - Fairness tradeoffs
46:29 - Facebook, social networks, and algorithmic ethics
58:04 - Machine learning
58:05 - Machine learning
59:19 - Algorithm that determines what is fair
1:01:25 - Computer scientists should think about ethics
1:05:59 - Algorithmic privacy
1:11:50 - Differential privacy
1:19:10 - Privacy by misinformation
1:22:31 - Privacy of data in society
1:27:49 - Game theory
1:29:40 - Nash equilibrium
1:30:35 - Machine learning and game theory
1:34:52 - Mutual assured destruction
1:36:56 - Algorithmic trading
1:44:09 - Pivotal moment in graduate school
Lex Fridman thank you so much bro. I’m so grateful for all that you do.
Thank you for adding the outline to the comments. This is the only way timestamps work for mobile users.
Great pod. You should suspend the guests mic, a lot of thumping while listning to headphones. Thought it was my neighbor moving stuff at first
thank you
Although this isn't one of the more highly viewed videos, I find equal if not more value from it. Michael Kearns comes off as a truly authentic and brilliant mind within the field and I'm thankful that you shared his insights with us.
Indeed!
Ha! Just found this. Was a grad student with Mike at Harvard, at the time he mentions at the end. Had a few fun times. Can't wait to give this a listen!
Thank you Mr.Lex Fridman sir and thank you Mr.Michael Kearns sir for good talking.
On behalf of myself and all of youtube I would like to congratulate you sir and thank you for your comment
Michael, just have the same feeling as you shared in this interview, about your Bachelor's time in Berkeley and then your Master's in Harvard. Today I finished my Master's Capstone, and I walked through the aile. I thought that, "actually I only needed a Bachelor's to equip myself with the majority of the skillsets I am using in my daily job. Master's degree is for people who really want to do research." Berkeley was also the first place I visited in the US. But later you mentioned about "had most of the interaction with math people and computer science people". That was what made me happy when I first started my Master's as I felt I was with like-minded people. So, feelings change overtime when we are in different life stages. Thanks for the great sharing and Lex's great questions.
highly appreciate your style of promoting sponsors🙌
I always learn so much from your podcast. Thanks Lex!
You are an ENORMOUS inspiration, my man. Not to mention BJJ Black belt and awesome guitarist. \m/ Keep crushing it!!!
32:08 I would just say that protection against false incarceration isn't bout "fairness for the criminals", or it wouldn't be false incarceration.
It's important to remember that not every defendant has committed a crime, and not every plaintiff is a victim.
Great interview....... you need to always put a low cut filter on the vocal tracks, like 18dB per octave at around 80 hz. Because there's a lot of sub bass plosives coming from the mics. And i hear lots of other sub bass rumbles coming from the building and from Michael banging on the table!........ But yeah, i do happen to be listening in a mastering studio LOL!
In the elon musk interview, I heard people walking around above or to the side of him. I kept thinking there were people in my house.
Even though there was some slight disagreements it was a super informative and interesting interview! 👍 Thanks so much!
Fascinating podcast.
FWIW, my take: If someone hasn't embarrassed themself for a good long while, they haven't truly sought mastery or tried to attain difficult knowledge...I wouldn't even feel embarrassed about it, that's just the price of admission. I'm sure I don't have to say this, but gems can still be uncovered whilst 'embarassing yourself', and sometimes at the end of the journey, you'll pull some of those gems back out, seemingly serving no purpose until you see it all come together.
Well said, @mhill88ify Eric Weinstein articulated it even better IMO when he coined the term 'Embedded Growth Obligations' to describe how institutions built on a growth model (college professors looking for tenure teaching their students to become professors, etc.) have stagnated any original research in the past few decades.
@@ishansaraf1576 haha, 'embedded growth obligations' - I love it! We all should start taking this idea to heart. Being alive entails having 'embedded growth obligations"!.
Whole heartedly agree
These podcasts are brilliant. Big thanks to MK for this one.
amazing podcast, Lex
Equality of outcome and of opportunity is huge.
This is excellent content, worth the entire almost 2 hour time. Really good discussion but sadly the points about algorithm operation (start at 49:00) lacks the issue of driving profits for the 'googles' and advertisers out there and how the algorithm optimizes for those two sides while claiming that the user's experience is "enhanced". However, the discussion redeems itself when Kearns adds a third side (start at 51:45) wherein the user adjusts the algorithm for the user's desires NOT just the other two sides. Examples of the former is when Netflix/Amazon Prime Video give me FIRST what videos they produced vs. what I really like to watch or a google search feeds back what maximizes ads and/or clicks based on my search. Sadly, I can't give an example of the latter representing that third side where I get what I want - precisely what I want. Again, an excellent content, worth the entire time.
Good interview
Hello Lex. I am a fan of your clear and forward interviewing style. Would you ever consider interviewing Andrew Yang?
+
Andrew Ng would also work 😉
Also: Lex, I would take a guess that your taking notes and selecting the best ideas from these great minds to create your own one true AI? 👍 Keep up the good work
How do you address that bias is everywhere (i.e. the background of Lex's guests) and leads to goal optimization and/or evolution (specific podcast content)? Evolution & goal optimization are often driven by human bias (learning about AI from highly educated guests). Is there consideration that by ending all human bias in algorithms could slow overall forward human progress and evolution?
Does anybody know good bibliography on the interaction between Machine Learning and Game Theory?
For anyone interested in learning more about algorithmic fairness I recommend reading Game Theory and the Pursuit of Algorithmic Fairness by Jack Frostwell. It is a great book on the topic.
1:01:00 sure the computer shouldn't automatically decide what's fair, but it could find hard-to-spot cases to flag for human evaluation and ask "Hey, I found a discrepancy--is this fair?" Kind of like Chomsky's anarchism--systematically questioning power differences...
After hearing him mention Infinite Jest I knew I was going to like this podcast!
Great talk as usual and a wonderful guest. Very nuanced and gentle in its presentation of multi generational class warfare, given the political atmosphere in the USA this is the safe choice. Looking forward to googling more about the author.
Book recommendations:
1. Infinite Jest
2. Ethical Algorithms
If you edit out all the times he says 'You know' the interview will be half as long :-)
I understand that algorithms can help in the performance of justice, the administration and delivery, in terms of accuracy and precision. It's this very increased performance that concerns me. Some law that is ten percent unjust is not so bad, but what if it's enforced with one hundred percent efficiency? Hard to decide on these matters. Do we really want to have, say, nearly perfect lie detectors on all government and medical devices?
At 11:12 Michael Kearns had an opportunity to acknowledge that academics are a group of people in a faulty culture normalized by its being a bubble, and then he used it to say outsiders don’t get it. Ironically this proves his earlier point that people acclimate to normalize their cultural bubbles.
Was it just me or were there a lot of edits?
Are we in the gray area yet?
Lex you get the best minds on the planet... that's why I fks with you... No homo. #FIVESTARS
You know, you know, you know, you know.
Ok I'll take that recommendation. pessimist archive
If this guy is so smart, why does he not realize that tap dancing on the table would be hell to headphone users!
All your podcasts with table-mics are of lower quality. Get them permanently off the table (even when offloc) - or add some sort of filter to prevent the constant thumps some of your guests are causing. I just finished watching your interview with Christof Koch and in ~57mins he must have banged the table 500+ times.
What is useful isn’t the basis for what is Good. There is a ton of behavior that is useful to certain people and isn’t good at all. Utilitarianism
"there's alt-right folks in 4chan" lmao Lex is a channer. :-)
There's also a Russian chan 2ch, which also delivered since 2008. But then %%can't pronounce%% bought it and now it's just a force of trap fap threads
Based on the fact that we can't defined what fairness is, we need ML to determine if an ML algorithm is fair. Oh wait...
Equality isn’t inherently ethical.
Made by Jimbob Nor is inequality. “Equality”, like “liberty” and “fraternity” (or human solidarity in modern, non-sexist terms) are the fundamental values of western enlightenment civilization. You need to give the sense in which inequality is not inherently ethical.... and why would you only pick on equality. Why not say the same about liberty or solidarity? Is the liberty of a slave owner ethical? Of course no
being motivated by power or money or control will produce different behaviors than being motivated by truth, beauty and the good. This man is naive to the motivations of the most powerful, the history of their legacies, and how what is Good can be ignored by those who merely want to win.
Only 10 mins on algo trading? Cmon i could talk 250 days in a row about algo trading
Yang gang :)
God!!! Did anyone count how many times this man said "you know"? Horrible.