I really enjoyed this conversation with Judea. Here's the outline: 0:00 - Introduction 3:18 - Descartes and analytic geometry 6:25 - Good way to teach math 7:10 - From math to engineering 9:14 - Does God play dice? 10:47 - Free will 11:59 - Probability 22:21 - Machine learning 23:13 - Causal Networks 27:48 - Intelligent systems that reason with causation 29:29 - Do(x) operator 36:57 - Counterfactuals 44:12 - Reasoning by Metaphor 51:15 - Machine learning and causal reasoning 53:28 - Temporal aspect of causation 56:21 - Machine learning (continued) 59:15 - Human-level artificial intelligence 1:04:08 - Consciousness 1:04:31 - Concerns about AGI 1:09:53 - Religion and robotics 1:12:07 - Daniel Pearl 1:19:09 - Advice for students 1:21:00 - Legacy
a suggestion for a future guest could be William "Bill" Easterly. he is a an economics professor at NYU (focusing on political economy and international development) and had roughly 15 years experience at the world bank as their head of research. He has written many great books. His big idea is that in contrast to the status quo approach of economists and aid agencies trying to reduce poverty through technocratic approaches and top down plans... a better more time tested approach is to expand political and economic freedoms to the poor. He also taught at MIT. Highly recommend giving him a youtube search and listening to some of his stuff. Thanks!
Great interview with a man that I have incredible respect for, and am in awe of. Reading "The Book of Why" was a great experience, although I'm going to need to read it multiple times I think, to *really* "get it". As far as suggestions for future guests, I would love to see one or more of: Ben Goertzel Marcus Hutter Pei-Wang Fei-Fei Li
Siraj is just one name. Every other person who has a RU-vid channel or an account on FB or LinkedIn think they are representative of the field of Machine Learning and Data Science but in reality their whole understanding is built by some shitty medium blog and all they know is how to stack layers.
@@dream1430 Check LinkedIn someday. Every one is calming that they are doing research but all they have to offer is some Tensorflow/Pytorch 101 course.
Thank you for being very respectful while chatting with your guests, even when you are discussing a sensitive topic. It is an amazing quality that every host should have.
This has to be one of the best interviews I have ever seen. Both Judea and Lex are continuously challenging their point with enormous respect. Judea is also an amazing communicator of complex ideas. thanks Lex for providing us with this content!!
Lex's podcast have a way of doing that lol. I never liked math but I now can see the beauty and importance of it thanks to him and all the wonderful guests.
[11:30] “Faking it, is having it. …Faking intelligence, is intelligence, because it’s not easy to fake. It’s very hard to fake…and you can only fake it if you have it.”
but that's not true. watch an interview with almost any famous actor: the actor isn't actually as intelligent or cool or whatever as the rolls he convincingly plays.
@@chrissmithdoe2100 Give yourself some time and think about the original sentence and your reply - maybe write it down...there is a clear answer, that is why I had to react. Wonderful podcast and a wonderful audience. Best wishes!
@@phlipsterroxor9068 well i thought i understood your point, which was the obvious point that, being actors, an observer arguably can't know whether they're acting stupid or are stupid. my next message was just trying to play along in a fun way. but i'm not sure why you're treating me like an idiot now?
Beautiful. Life is short, but you make my life so much richer with interviews like this one. I will watch this again with affection. Point of personal history. I was at UCLA in 1968 and 1969. The internet began on October 29, 1969 when Leonard Kleinrock and his team at UCLA sent a message to Stanford (only the first two letters made it through). At that time I was interested in film, art, music, parties, political demonstrations, and working to save money to go to Europe, and I knew nothing about computers. I traveled to Europe, and lived in Paris, working as a construction worker. I returned to UCLA, switched majors, and graduated from the UCLA School of Engineering in Computer Science. Judea Pearl’s office reminds me of the offices in the school where my friends and I would hang out with professors drinking coffee, smoking cigarettes (times have changed) and talking about self-aware robots, Boolean algebra, transfer phenomena, and multi-dimensional spaces. So many wonderful professors then, and now, who were so kind to us students. Watching these videos of so many extraordinary people makes me wonder about the many paths that I did not take, but if I had taken any of those paths then I would not exist to say--thank you. William L. Ramseyer
I love how Prof Pearl seems to keep Lex a bit off balance, not in a bad way, but by continually questioning and challenging various points as they're made.
I think around 56-58 mins-ish I believe the rifle man example it’s clear that Lex hadn’t grasped the principles early in the conversation. It’s easy for the mind to wander off the path 😂
Mr. Judea Pearl is so amazing! My hart is bleeding for him and his late son. But he is so full of joy and life! I wish I could ever become at least half a person he is...
Lex, your talent for extracting knowledge from your interviewees is amazing, inspirational, and very much appreciated by this armchair aficionado of the sciences.
It feels like Lex is trying to ask very cautiously how to build an AGI and Judea politely declines to break our hopes about it happening any time soon :D Great podcast, refreshing to see people that actually know what they are discussing
Amazing interview. Beautiful exploration of what basic concepts in causality actually mean, complete with examples. I've watched several speeches by Pearl, but never anything as clear this, Lex Fridman very craftily structures the conversation.
How did you cover such a magnificent span of topics and range of intellectuality and emotion. Can't believe you asked that deep and perfect pair of questions about his son.
What Pearl says about intervention is much the same as what LeCunn says about infants. Infants observe the world (mostly) and occasionally intervene. Infants (and kittens batting a ball of yarn) are building a causal model from the data -- from some initial architecture and set of conditions -- and resolving ambiguities with interventions. Intervention is a way of pruning the causal graph to make it less "bushy", as Pearl puts it. The other lesson to draw from this interview is the importance of historical and cultural grounding in math and science education. We typically interact with disembodied knowledge about abstract structures. But Pearl is firmly situated in time and place. He knows who he is, who Descartes was, and who his people (the Jews) are. He knows how to relate to Archimedes and Daniel and the king of Babylon. All these connections mean that his personal knowledge graph is very bushy. He can approach a topic from multiple vantages and evaluate the merits of many different paths through his personal graph. But if all he knew about a subject was what he learned in textbooks, then the sparseness of his understanding would preclude insights and wisdom.
Man, that is absolutely horrible what happened to his son. He's a strong man being able to talk about it without falling to pieces. Or maybe that's just time dulling the pain. I love my children and I don't want to think about how I would grapple with such a tragedy. Love of ones offspring is a powerful emotion. Sorry for your loss sir.
I admire the financial independence of people, But you can live better if you work a little more. After watching this I think there are people out there, on the extreme, who plan to die early just to be able to retire early. To each their own but to me, retirement isn't just about not having to work, it's about having the freedom to do whatever you might reasonably want, such as travel, buying things, enjoying life, etc. I don't think I could retire with less than $3m in income-generating investments, maybe $2m at the very minimum. I plan to work until I'm at least 45
Nobody knows anything, you need to create your own process, manage risk and stick to the plan, through thick or thin while also continuously learning from mistakes and improving
@@harrisonjamie794 Having an investment adviser is the best way to go about the market right now, especially for near-retirees, I've been in touch with a coach for a while now mostly cause I lack the depth knowledge and mental fortitude to deal with these recurring market conditions, I netted over $220K during this dip, that made it clear there's more to the market that we avg joes don't know
One of the best interviews I saw in a long time! Judea Pearl is a genius with a wonderful sense of humor and a big heart!!👏👏 A rare breed in today’s world!!
A beautiful episode, loved it. I had a feeling I'm drinking a cup of coffee with the 2 of you - that's a good sign that the podcast was a success. Thank you.
Who the fuck is disliking these podcasts? Honestly, it blows my mind. I understand you might not like the topic at hand or the guest but to take the time to go on and dislike such high quality of content is absurd.
I had hard time to agree with many things that he said. However, I was in awe at the end of the discussion. Ended up buying his book. Great talk! Gave me many fruitful insights!
Thanks @Lex. It was an interesting conversation that tried to align a mental model of reality, that is very human and uses terms like causation, and structure, to some of the mental models that a pure pattern recognition algorithm can build from raw observation of the world (eg a baby of another specie of being for example). Can we 'learn' how to build mental models, is the topic that you kept coming back to.
Thank you for respecting the artistic integrity of your podcast and putting ads at the beginning! Sean Carrol has degraded his podcast significantly by having interruptions for unrelated things in the middle of a deep conversation.
The important thing to understand is that when we *do not* add arrows to the causal diagram it is an explicit assumption that there is no causality in between two variables. And that is essentially the default in many observational studies. What causal inference does is allow one to explore and better understand how those implicit or explicit assumptions actually influence the analysis of the data and conclusions that can or can't be made logically. At least, that's the important takeaway for me.
@@SalarymanNoMore ??? How can you fake a higher order of intelligence. If people try to fake intelligence by using smart words and throwing in a few premeditated comments, they appear as stupid as people walking around naked. Can you pretend to be good at chess in a chess competition?
What a brilliant man! I love that he said the best way to teach math is chronologically! I always believed so. I need to know who, when, and why first! He is so funny, I don't know why Lex isn't laughing harder. As a roboticist, this is undoubtedly my favorite episode!
2 of the biggest dangers imo is the fact that generals are going to get their hands on it for war. Secondly the creation of A.I could start a war itself if leaders realize the power advantage because someone is going to create it first.
"Ask your questions. Really, your questions are never dumb. And solve them your own way. And don't take "no" for an answer." Judea Pearl - Lex Fridman Podcast #56
There is a textbook on current ideas of learning causal relationships from data. It's Elements of Causal inference by Jonas Peters. Let's look at the firing squad example. Let o,a,b,d be boolean variables representing wether by the end of the trial an order was issued, rifleman 1 shot, rifleman 2 shot, and the prisoner was killed. If you ignore the time precedence like that (that during the trial a and b always change after o, and d always changes after a or b) it's impossible to infer the causal graph without making additional assumptions. A lot of current methods work by making assumptions. If you do take into account the time precedence it becomes much easier. One way to learn the graph in the firing squad would be to collect time series information o_i, a_i, b_i, d_i for every trial where i is in milliseconds (assuming it takes at least a millisecond for the bullet to reach the prisoner). We would know that the series o_i precedes a_i and b_i, which in turn precede d_i. We still know that a change in o_i causes a change in d_i, but we can find out that it does so only indirectly through a_i and b_i by testing for conditional independence (d_i -||- o_{i-1,i-2...} | a_{i-1,i-2,...}, b_{i-1,i-2,...} ).
