Michael I. Jordan: Machine Learning, Recommender Systems, and Future of AI | Lex Fridman Podcast #74
Vložit
- čas přidán 1. 06. 2024
- Michael I Jordan is a professor at Berkeley, and one of the most influential people in the history of machine learning, statistics, and artificial intelligence. He has been cited over 170,000 times and has mentored many of the world-class researchers defining the field of AI today, including Andrew Ng, Zoubin Ghahramani, Ben Taskar, and Yoshua Bengio.
This episode is presented by Cash App. Download it & use code "LexPodcast":
Cash App (App Store): apple.co/2sPrUHe
Cash App (Google Play): bit.ly/2MlvP5w
PODCAST INFO:
Podcast website:
lexfridman.com/podcast
Apple Podcasts:
apple.co/2lwqZIr
Spotify:
spoti.fi/2nEwCF8
RSS:
lexfridman.com/feed/podcast/
Full episodes playlist:
• Lex Fridman Podcast
Clips playlist:
• Lex Fridman Podcast Clips
EPISODE LINKS:
(Blog post) Artificial Intelligence -- The Revolution Hasn’t Happened Yet:
hdsr.mitpress.mit.edu/pub/wot...
OUTLINE:
0:00 - Introduction
3:02 - How far are we in development of AI?
8:25 - Neuralink and brain-computer interfaces
14:49 - The term "artificial intelligence"
19:00 - Does science progress by ideas or personalities?
19:55 - Disagreement with Yann LeCun
23:53 - Recommender systems and distributed decision-making at scale
43:34 - Facebook, privacy, and trust
1:01:11 - Are human beings fundamentally good?
1:02:32 - Can a human life and society be modeled as an optimization problem?
1:04:27 - Is the world deterministic?
1:04:59 - Role of optimization in multi-agent systems
1:09:52 - Optimization of neural networks
1:16:08 - Beautiful idea in optimization: Nesterov acceleration
1:19:02 - What is statistics?
1:29:21 - What is intelligence?
1:37:01 - Advice for students
1:39:57 - Which language is more beautiful: English or French?
CONNECT:
- Subscribe to this CZcams channel
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridmanpage
- Instagram: / lexfridman
- Medium: / lexfridman
- Support on Patreon: / lexfridman - Věda a technologie
I really enjoyed this conversation with Michael. Here's the outline:
0:00 - Introduction
3:02 - How far are we in development of AI?
8:25 - Neuralink and brain-computer interfaces
14:49 - The term "artificial intelligence"
19:00 - Does science progress by ideas or personalities?
19:55 - Disagreement with Yann LeCun
23:53 - Recommender systems and distributed decision-making at scale
43:34 - Facebook, privacy, and trust
1:01:11 - Are human beings fundamentally good?
1:02:32 - Can a human life and society be modeled as an optimization problem?
1:04:27 - Is the world deterministic?
1:04:59 - Role of optimization in multi-agent systems
1:09:52 - Optimizaiton of neural networks
1:16:08 - Beautiful idea in optimization: Nesterov acceleration
1:19:02 - What is statistics?
1:29:21 - What is intelligence?
1:37:01 - Advice for students
1:39:57 - Which language is more beautiful: English or French?
Any chance of seeing Eliezer Yudkowsky on a future podcast Lex ?
Fantastic. Thank you for finally having someone who speaks directly, addresses the large, scathing gaps between what is being marketed as "Elon Musk" level AI and what we actually do with these algorithms (tanh, sigmoid, relu, maxout, softmax and so on). It's refreshing to hear someone say outright what a lot of us think internally -- that we're very far off from ever resembling a human brain and what we're doing now is at best generating approximations off the corpus of data we collect -- as if the neurons in our brain are processing some version of these basic functions. While simplicity can be beautiful, it's not always the correct answer.
It's incredibly annoying to hear people talk about AGI as a "50 year problem" implying that by 2070 this will be figured out. It really is an insult to the nuance and beautiful complexity of what makes us human. Again, thank you for having someone who understands this and speaks about it directly even in the face of obvious criticism he receives in not being an evangelist for the science. For me at least, he's inspired me to delve deeper into this field.
Lex, so far my experience with cash app has been awful. They have horrible customer service. Ive so far lost 150$ and ant get them to help me. Unimpressed.
Thank u
He's calling for sobriety and hell i thought I'll never hear that from a .. let's say computational categorizer :)) he's a real scientist down to earth and I love that! An other highlight of Lex's podcast - aren't we listeners spoiled guys?
Just had Professor Jordan this past Fall 2019 semester at UC Berkeley and thoroughly enjoyed a new course he and other faculty piloted called “Data, Inference, and Decisions.” As a fan of Lex’s as well, it’s so awesome to see worlds collide.
Is his coursework accessible online?
@@SM-vg6xk thanks for the link
You can view the current semester’s material here: www.data102.org! Professor Jordan is currently not teaching it this semester but instructors for each course often cycle between Fall and Spring semester so that those from Fall 2019 may return for Fall 2020.
"Technically MJ is the Michael I. Jordan of Basketball"
I agree.
"Young people get into this field and think its all done because we have tensorflow" :)
If the alternative is that you will not invent AI in "500 years from now" I can see why though. ;) Why even bother to believe a pessimist.
Wait, did you seriously just say that one of the foremost machine learning experts isn't doing anything to advance the field?
@@ben2258 I agree that it sound a bit harsh but when you I think about it, the professor said so himself when he claimed we won't see any significant progress towards AGI in the next 500 years.
Michael Jordan naturally speaks at 1.25x speed
XD XD XD
And Lex Fridman talks at .75
Michael I Jordan is literally for me is a brutal Realist and knows the stuff inside out as he had consumed it,chewed it,and spits it out whenever he desires with no icing..
THIS IS JUST SOOOOOO GOOOOD!!!! I hope we get more of these... Thanks to Both Lex and Michael...
Michael Jordan is such a goddamn national treasure. I wonder how he grew up. I wish Lex had asked him
This occurred to me while listening to Lex and Michael I Jordan go back and forth about recommendation systems: using all past behavior to generate new recommendations is not the best way b/c most interesting things about us may not be in the data of our historical usage/puchases/etc...wanting recommendation systems to get at principals and our values and serving up recommendations that could tap into those (but are not intuitive based on past history alone) we have an analogy in recruiting to this: culture fit vs culture add culture fit generally refers to the degree to which candidates' values/beliefs/principals/etc.. mesh with those of the existing organization whereas culture add is about how do candidates' values/beliefs/mindsets/points of view/life experiences/etc... enhance existing organization in new ways
Well said Sean!
I love you mate, without a doubt my favorite podcast right now
I appreciate your choices of guests Lex. I feel that you interview and ask all the questions I wanted to ask to the people I always wanted to talk to. You are amazing and I am happy you are successful in this realm.
It is all so fascinating and it fills my void for intelligent fulfillment.
Thank you,
Mark
The clarity of his point of view is mind boggling !! Really enjoyed the way he thinks, not only in the context of academia but taking the market into perspective.
Michael I Jordan finally helped me understand math and I actually understand what he's saying. He doesn't beat around the bush, only when he uses those words of jargon.
how did he help you understand math? tell me more... thx
You ask the most brilliant questions and allow your guests the time and space to answer them without rushing into other subjects. Also I love when you ask them to elaborate on the interesting points
Excellent podcast, keep going!
Hello Lex, I am 3 rd year phd student, Thank you so much dear Lex for your awesome videos, alongside my research I have been following your videos online for about a year now, You and your amazing guests have educated me to an unbelievable level of knowledge regarding AI and my research which is focused on vision-language tasks using attentive deep neural networks. Again thank you and please keep up the good work and do not stop any soon.
Favourite episode so far, great stuff, thx! 👍
1st🙀🙌🙌😃. Lex Friedman I greatly appreciate the amount of hard work you dedicate to conducting these interviews, considering you’re only doing this as a side project.
Your podcasts open my mind in so many ways, cheers from Argentina
OMG....Mr. Jordan tells it like it is...instead of telling us what we want to hear. Thank You!
Thanks for this! Michael Jordan is one of my favorite scientists.
Still think this is one of lex's best interviews. Mike is such a smart no nonsense guy.
100% spot on
Michael was Great, I think we need to have another round with him
couldnt agree more
I think I understood parts of that. It's impressive to see someone that intelligent, talk so casually, yet feel like I have only a glimpse into what is being said. Think I will need to listen to that a few times to get more concrete details from the conversation. Great work Lex.
I love how down to eath, conscious men he is. We need more people like him.
This is a truly interesting conversion. Thank you for sharing this.
6:02 - Do you think there is something deeper in AI than [...]
- No.
Michael I. Jordan is just great.
Most unappreciated interview of all time for Lex
1:44:00 That was a cheeky "merci beacoup" from Lex
These podcasts are invaluable. This is spectacular. Anyone not seeing that this is pure gold, doesn’t know shit of where human society is heading. Amazing. Keep up the astounding work.
Michael Jordan;s first 5 mins into the interview , just really sums up the game. clear , precise, and even actionable--- what we as engineers , scientist, philosophers, and even doomsayers should inspire to build ..... a System for Intelligence, where engineer is becoming less adhoc
Amazing personality! He is an institution by himself! 👏🏻👏🏻👍🏻👍🏻
So agree with his view on current recommendation systems.
I like his reinforced point about connecting the real producers and consumers to create real value. This is what the right AI algorithm should do. Not companies who make lots of money by feeding from advertisement without caring about the right information flow.
Great discussion, I enjoyed it! Thanks both, Marco
I was waiting for this episode. Thanks!
Jordan dropping some huge bombs, finally someone setting things straight about AI
Fantastic episode
OMG completely agree. Try and promote this sit down every time I can. Matrix is not in our or our kids lifetime folks. Dont buy the lie...
Can you please interview David Blei, another great professor and a student of Michael Jordan. I am a fan of his work.
Another amazing conversation between two brilliant scientists. Love his different perception and perspective about AI. I don’t like that term 😂
I am a Brazilian astrophysicist and I am enjoying your podcasts. I would like to suggest Lee Smolin, a humanistic physicist and philosopher, known by his work in quantum gravity and the nature of time.
Michael is awesome to listen to because you can barely fault on him on the purely technical aspects of a conversation - but more importantly he has a grasp on economic and marketing concepts that totally defuses Lex's typical hype about the Elons, Facebooks, and Googles of the world
“Don’t try to use your AI to kind of figure me out.. then put me in a world.. where you figured me out!”
😅 love it.
Michael Jordan The GOAT
IMO its tragically insane how undervalued this interview is. One of Lex's GOAT. Theres no pill or whatever for the common cold. But in our lifetime a human to computer interface like that in the matrix is remotely plausible. Not happening folks.... You've been sold a dreamers lie.
1:01:35 "We have blinders on...We don't see the other person's pain, that easily" A very true sentence to be pondered by everyone to learn appropriate level of empathy.
Great podcast. I learnt a lot.
This is a gem.
fascinating how we impemented game theory in our tech to solve optiisation instead of thinking everything is tensor flow related and to hear Michael confirm the same makes me happy
Thank you so much. Keep up the good work .
Talking about Artificial Intelligence today is like the ancient Greeks talking about how to get to the moon. Wow! That's someone really calling it like it is. Michael Jordan. The most all round intelligent man you will ever hear.
Agreed. We cant cure the common cold. But believe be we will have Matrix like interface in our lifetime..... So I have this bridge to sell someone....
just brilliant, thank u both.
What you refer to as Artificial Intelligence is really Artificial Skill. If a Skill is learned through a program or a network, static or dynamic, is only a matter of methodology and not a question of what you are trying to achieve. Try to replace AI with AS every time you say it and you'll see that Skill is what it is all about. The field of engineering that is emerging, as your guest was talking about, is computer skill engineering.
Forgot to say: Great interview!
Awesome Point
@@skierpage I have never heard of that idea of intelligence. The skill is a tool for the intelligence as I see it but interesting none the less that others have thought intelligence is being above a certain threshold of skill.
I don't think AGI is a real thing. It sound like magic to me. Intelligence I would say, is not about developing a new skill but rather deciding if a skill is useful to learn at all. When we encounter new things we use our intuition which is very different from intelligence.
But, as you say, if we really are going to use the word intelligence to describe skill then we might as well use the word intelligence to describe intuition and call it AGI.
Hi Lex, this has been awesome. Can we have Frank Kane to talk about Recommender Systems. His Sundog course on Recommender systems is amazing
Awesome Interview!
Another awesome conversation! 👏
oh my god i wanna buy his book but no book available at amazon... please publish your book MJ. Everybody's gonna buy it
This is a great podcast. I listen to this one and JRE a lot.
14:36 he explains how "Artificial Intelligence" is an incorrect name for the field and what we do have is a "Machine Learning" field. BIG difference
Michael Jordan is a hero of reasoning … bravo
This is one of Lex's best of GOAT.
The way prediction vs. decision making is differentiated in ML reminds me of the observer effect in quantum mechanics /jumping around Hilbert spaces. Lex - keep em comin'!!!!
also i think you should think about if the advertising industry in emergent or fundmental
Thanks for an informative scientific interview.
Love your work. Any chance you could interview Garret Moddel?
Absolutely brilliant. I’ve had similar thoughts as Jordan that what deep learning has done is analogous to like the microchip industry.
I respect your honesty sir
This man is a living, talking library.
Good interview and I was waiting a lot for it. However, I want to give you some tips. Jordan is not just any interviewee and you need to prepare well. Some questions were asked well in the beginning. The issues need to be dealt with rigorously and with a lot of complexity. It would be interesting to have more public participation by asking questions so that they can answer themselves. You could select and do more in the semi-final part yourself. The ending was good and I managed to learn a lot from the interview.
Seems like you have good ideas. Maybe a podcast in your future?
Apparently every Michael Jordan holds up to their craft. Michael Jordan, Michael B Jordan and Michael I Jordan.
You know a person is wise when he/she begins answering saying they’re not an expert on the topic as opposed to blabbering about your assumptions and bias on the topic.
How have I never heard of this guy!? I love how he digs into how we need to stop putting so much stake into being so literal in the metaphors we use for our algorithms.
When allowing things to speak for themselves and for the layman to start grasping how these technologies work by building their own mental models from what they have provided, these metaphors can be a huge disservice rather than an effective explanation tool. For-sure turning the core inspiration into grandiose pop-sci metaphors is easier, but you're polluting people's first impression (when their brain is the most impressionable and attempting to build a foundation and schema on the subject) on how to understand it; and when marketing and multiple media outlets do it in tandem, of course the public is going to misunderstand these things.
The only people you mostly hear from are from bloviating communicators and front men. There are hundreds of thousands of people (in a world of billions) like this that actually make the world go.
Some of the people I'd love to see on your podcast: Russ Salakhutdinov, Frank Wood, Noah D. Goodman, Vikash Mansinghka, Kevin Ellis. Please make it happen :)
I was listening to this while I was sleeping and wow what an amazing dream! 🙏
Lex thank you for your patience with this guy
For some reason he reminds me like Mark Cuban, his straight forward tone and some facial expressions.
Andrew Ng and then his doctoral advisor? Cool!
That was a life changer. Miles Davis indeed.
I really love the latter part of this interview. His view on brand community economics are really interesting! The earlier parts of this interview I have hard to grasp even. We don't really know what we mean when we talk about intelligence nor how the brain generates the projection of such phenomenon - just as prof Jordan also suggests - yet he seem to state without any thread of doubt that he knows what it isn't and that progress isn't going to happen within his lifetime. To me this has more the smell of burned bridges and abandoned dreams to it than the leadership he clearly shows in the latter part of this interview. Young people are no just swayed by glamour of Google. They listen to prof Jordan too, as long as there's a story worth listening too. Tell one on how they''ll discover AI in their time.
Lex...accept the sober voice.....don't fight it...embrace it!
what he dismisses as "science fiction" is actually metaphysics and his entire existence is predicated on it
I totally agree with The Prof about the definition about AI and ML and the the difference between them, which we have been using in many discourses about the field.
First, As Chomsky, quotes: be humble, we dont know much about Consciousness which seem raise the human mind, human intelligence. What is the real intelligence? The question is still there.
Second, the way we used to metaphor the things, in some sense and some context, its so dangerous, cos it not the way the things happened as they really are, just the way we use some tools in math, physics, and the capability of our own cognitive system, which in my understanding in practical, not just human brain, but deep in our gens, our story of the age we live, our material that construct our whole body.
How about Sapolsky and the wonderful book, Behave, on your Podcast?
Btw thanks both of you so much. The scientific talk with the awareness in balancing way always help.
Lex I love you, you’re my Братан... that’s the limit of my russian
bro
Кент просвещающий)
Alexander я собираюсь вести себя так, как будто я понял, что без гугл-перевода ... я также буду вести себя так, как будто я не набираю это с помощью гугл-перевода ... русский это сложно 😂😭
@@dylanhayslip9206 Советую поехать дауншифтить на пару лет в деревню или русскоговорящий пригород, без полного погружения в среду изучить язык по настоящему не получится к сожалению (
이철우 엄청 큰 도움이 됩니다 어떤 일이 없도록 하는 것은 아닙니다 여러분 안녕하세요 지휘관 능력 및.
Anyone: *mentions anything*
MJ: There's no good word for that.
Yeah very tiresome, gave up on this one quickly.
Finally.. someone made sense of this AI overall
Amazing 👏 We have lost DECADES of progress due to CYBERNETICS being misunderstood. Any random banker, railroad admin, paralegal or librarian looking at how their overall systems work and improving them, removing steps, cascading data, centralising access, creating feedback loops etc. etc. THAT is true cybernetics. It's not only about electronics, robots and cyborgs.
This is the most awesome interview EVER!!!!!!!!!!!!!!!!!! He's a fucking Genius!
So sad LF doesn't champion this interview for how great it is. MJ should be a reoccurring guest imo...
What he says about the music market is so so true. Hopefully Music NFTs will be able to funnel more of the money directly from consumers to artists.
Glad to see Jay Leno pursuing new interest in his retirement
We need Ben Goertzel
Does anyone know any textbooks or papers on decision making as said by mj?
This man is smart.
Is there a good textbook/paper that goes more in depth on "decision making"? It sounds rather interesting.
Yes, but it becomes very mathematical and you need a good understanding of probability theory and calculus and linear algebra. If you have those, then you can look into books by Dimitri Bertsekas and Richard Sutton. For something a little lighter, you can read the textbook Networks Crowds and Markets by Kleinberg and coauthors, which is closer to the spirit of what Michael Jordan discusses here, but doesn't require a lot of math
Yoooo boss man 4:19-6:00 🙌
This person has the kind of intelligence that inspire me, a very wide one.
"a good recommender system is better than a bad recommender system" - noted ^^
Would love a threeway Episode with Peter Thiel and Rush Limbaugh. Oh...
Thank you Lex and Professor Jordan! You guys definitely rock. However, Professor if you go abroad you will certainly meet people... Ask them for resto recommendations 🤓 Natural intelligence works pretty well for that use case!
Please think about inviting Robert sapolsky from Stanford University neurobiology a very interesting person and a good subject for you on stress the brain and complex system
yes!
Well done 👍✅✅
He is a giant in statistics
Great podcast!
Statisticians and machine learning experts, hear me out. I am currently thinking about the machine learning vs. 'traditional' statistics concept. In my mind it boils down to prediction vs. inference. Could someone give me thoughts regarding the following question: should an inference model (domain based believable model that fulfills basic model assumptions to an adequate degree) always predict new observations as well as a prediction model (machine learning model)? If it doesn't (and a prediction model does), then how can it be a believable domain-based model?
@@lepidoptera9337 I'm well into the thinking process about this issue. This is a very interesting and nuanced question.
One thought about the issue. Even if I have the absolute correct model, its usefulness is limited by variance. And variance has different sources also. So one person might be happy to have found the 'correct' model which can give horrible predictions (on average good, but any single prediction possibly horrible). Another person might treat this 'correct' model as absolutely unusable because of the bad prediction performance (average correct predictions not good enough; variance should be small also).
@@lepidoptera9337 Actually, if I had to choose between no/yes, I'd choose yes :) But it would be interesting to hear different, well explained and justified, opinions from real-world cases.
Can you invite Alan De Botton for an interview Lex?
Came here for recommender systems.
Stayed for the dressing down.