I really enjoyed this conversation with Peter. Here's the outline: 0:00 - Introduction 0:37 - Artificial Intelligence: A Modern Approach 9:11 - Covering the entire field of AI 15:42 - Expert systems and knowledge representation 18:31 - Explainable AI 23:15 - Trust 25:47 - Education - Intro to AI - MOOC 32:43 - Learning to program in 10 years 37:12 - Changing nature of mastery 40:01 - Code review 41:17 - How have you changed as a programmer 43:05 - LISP 47:41 - Python 48:32 - Early days of Google Search 53:24 - What does it take to build human-level intelligence 55:14 - Her 57:00 - Test of intelligence 58:41 - Future threats from AI 1:00:58 - Exciting open problems in AI
I really appreciate these breakdowns!! I like to go through your videos multiple times and these really help :). Please keep doing them for as long as you can!
I think his article is fun to read, but in my opinion it's title is clickbait. The point of simple tutorials is to get the reader started with something resembling what they want to get out of their learning.
FYI Lex has talked about this "don't burn out" advice before. He thinks it is said and heard too often and that people could benefit from hearing more ambitious advice.
Peter is my favorite computer scientist, I got interested in Lisp and history of programming via his work. He has a very personal style of writing tech too, Sudoku solver for example. I had a similar motivation when created a Yazzy automatic player in Python few years ago. Glad to see him interview in this manner.
Hi Lex, your questions are always so good and deep. Thanks for your time in preparing and then getting these amazing guests on. Keep up the great work.
I loved this one. He is so simple and questions were also very good. Leaving 50 MOOCs half way oddly made me feel good as I feel guilty when I don’t finish one !
31:29 It turns out that during the Covid19, most classes go full online. It happens sooner than we thought. 35:28 I am wondering if most researchers always have these problems or the tendency to learn everything?
At 52:49 Peter Norvig acknowledges Al Gore's role in promoting the idea of a 'commercial' version of Arpanet. Gore has always been ridiculed by some for claiming that the 'invented' the Internet when in fact he made no such claim...what is fair to say...and Peter Norvig supports this notion...is that Gore was a.father of the commercial Internet and therefore a father of the vibrant technological marvel that 'the' Internet is today.
I met Peter Norvig once, and he is such a nice guy for real in person. He has a course on Udacity called "Design of Computer Programs" which I highly recommend.
Really great questions and really interesting answers. The AI to help coders was an interesting idea, especially for education. Students would find this really useful.
Artificial intelligence will very much be a part of our life in the near future but i also think that we should be connected to our natural world as well
I like the idea of a "what if" conversation, along with explanations, being a requirement of AI systems that decide legal outcomes so people can challenge the algorithm. Whether that leads to answers to why questions is going to be an interesting experiment.
36:00 totaly! You dont need to know how the universe work to boil an egg. Form other side its good, from time to time to check what and how is working, so we can optimize the structure and to make the things more resource effective
Interesting quote he says is information and knowledge is good but even better is the motivation, the attitude. Soo true about this part cuz u gotta learn new things motivated. Thats why jim kwik says emotions with information. Long term memory. Humans must have good mood when we learn.
Why can't AI systems operate without bias as talked about at 7:00. It would seem you would just remove properties such as race and sex from the algorithms??
Race and sex and others can be derived by combining other properties like name, income, city, job etc and the algorithm can find some that we can't even think about.
@@Polyian also, if there is bias in the data sets themselves - black incarceration rates (for example) in specific districts, which may be historic judicial bias, may then be replicated in the learnt behaviour of the algorithm by weighing an area as more risk prone.
I think AI also needs the symbolic approach. As an example: so that an agent with AI can understand and generate humor. In the connectionist/neural approach I see that task very very far in the future.
4:00 Hopefully this isn't going take away my ability to make mistakes and learn from them and feel accomplished and proud doing so. Especially ethics which seem to have evolved over time. Anyway, love this field and where it's heading us.
@Explicit Relativity I don't like the social engineering and the bias removal, sounds like communism, and is suggested to the EU by Cedric Villani in a paper a few years ago, and that scares me. Equal opportunities yes, and the use of historical data on all categories in order to pinpoint the root causes from problems we might experience nowadays. I use ML/PM to solve daily practical problems not the high level stuff. I'm from Eindhoven, Techno(logy) City, can't release it my friend, although I play the shakuhachi.
@Explicit Relativity Google is a great technology company and comes with great (social) responsibility. Their realisation is a fact though. Unfortunately nowadays everything needs to be black or white, for or against, depending on your own bias. I'd rather have good discussions and be proven wrong so I learned something. That's what I like about mr Fridman's channel, many different experts with different backgrounds. Sometimes I don't agree, many times I'm inspired. The only thing mr Fridman should do is having a 1 dollar Patreon subscription. I'd rather support 100 people instead of 81 as that is my budget to support content creators.
If your dataset of human features involves race (say, it's photos), the machine learning algorithm will likely implicitly learn the race as one of high level representations. Algorithm may learn something you did not intend it to learn. For example, US Army had a project where the goal was to classify US and Russian tanks. In lab tests system did well, but in field tests failed miserably. Turns out, all it learned to classify was the quality of the photos - of course, Russian tank photos were low quality spy images...
He meant that language shapes the way you think about the problem. People think in language terms because it's natural for them, but machine does not work that way. What is really happening is that machine is optimising a surface in a million or so dimensional space, and nobody really knows how to reason about such complex things.
There was actually an experiment of this kind. Chatbots developed their own language: www.snopes.com/fact-check/facebook-ai-developed-own-language/ (the rating is "false", but keep reading - the essence is in the nuance).
You can't satisfy both the protected class criteria and accurate prediction re: recidivism because race is real and manifests itself in behavior. There is no AI ethical quandary, there is simply the fact that black people have higher recidivism rates than other demographics.
I really enjoyed this conversation with Peter. Here's the outline:
0:00 - Introduction
0:37 - Artificial Intelligence: A Modern Approach
9:11 - Covering the entire field of AI
15:42 - Expert systems and knowledge representation
18:31 - Explainable AI
23:15 - Trust
25:47 - Education - Intro to AI - MOOC
32:43 - Learning to program in 10 years
37:12 - Changing nature of mastery
40:01 - Code review
41:17 - How have you changed as a programmer
43:05 - LISP
47:41 - Python
48:32 - Early days of Google Search
53:24 - What does it take to build human-level intelligence
55:14 - Her
57:00 - Test of intelligence
58:41 - Future threats from AI
1:00:58 - Exciting open problems in AI
I really appreciate these breakdowns!! I like to go through your videos multiple times and these really help :). Please keep doing them for as long as you can!
instablaster.
53:24 53:24 @@kishorevenkateshan397753:24 53:24
53:2 53:24 4
This podcast series is becoming some of my favourite content on the internet. Thank you so much for providing us with this!
"Learning to program in 10 years" - my favourite part of the interview. Thank you!
second. It's really enlightening to see him acknowledging that not knowing everything is another way.
I think his article is fun to read, but in my opinion it's title is clickbait. The point of simple tutorials is to get the reader started with something resembling what they want to get out of their learning.
if you do full time programming for 3 years at 50 hours per week it will get to you near that 10 year point.
Lex, you have been working too hard these days. Thanks for another great guest. Don't burn yourself out sir.
FYI Lex has talked about this "don't burn out" advice before. He thinks it is said and heard too often and that people could benefit from hearing more ambitious advice.
It's a gem of an interview and your astute questions have help bring Peter's very clear and structured thinking out well. Thanks Lex.
Crazy! I started with his book just last night! Really happy with your recent guests Lex, keep it up
Peter is my favorite computer scientist, I got interested in Lisp and history of programming via his work. He has a very personal style of writing tech too, Sudoku solver for example. I had a similar motivation when created a Yazzy automatic player in Python few years ago. Glad to see him interview in this manner.
Norvig, a living legend, great podcast!
This is legendary!
Hi Lex, your questions are always so good and deep. Thanks for your time in preparing and then getting these amazing guests on. Keep up the great work.
Dear Lex, thank you so much for sharing this interesting interview.
Thank you Lex!
I loved this one. He is so simple and questions were also very good. Leaving 50 MOOCs half way oddly made me feel good as I feel guilty when I don’t finish one !
Really love this conversation. Got me into thinking on lot of key things.
Mad respect for this channel, and Peter Norvig
31:29 It turns out that during the Covid19, most classes go full online. It happens sooner than we thought.
35:28 I am wondering if most researchers always have these problems or the tendency to learn everything?
Wow...he is a legend in the field.. thanks again for sharing this lex..
When is the 'Artificial Intelligence: A Modern Approach 4th edition' going to be released?
Feb 2020
I'm seeing April 12th, 2020 on amazon.com
Just bought 3rd edition before watching this...
Waiting for the Kindle version to come out
Loved your book, I read it many times during high school! Great how it presents many AI topics backed up with theory, showcased with clear examples.
Awesome interview, Lex!
At 52:49 Peter Norvig acknowledges Al Gore's role in promoting the idea of a 'commercial' version of Arpanet. Gore has always been ridiculed by some for claiming that the 'invented' the Internet when in fact he made no such claim...what is fair to say...and Peter Norvig supports this notion...is that Gore was a.father of the commercial Internet and therefore a father of the vibrant technological marvel that 'the' Internet is today.
LISP part of interview was nice!
big Norvig fan, ty
A more elaborate introduction of the Peter Norvig would be great
I met Peter Norvig once, and he is such a nice guy for real in person. He has a course on Udacity called "Design of Computer Programs" which I highly recommend.
Why did you take down the Siraj Raval interview? 🤔
Seth Nuzum apparently he’s a shady character
Great work Lex!
Love this conversation! As you said.. His research really inspired us!
AIMA has literally become the Bible of AI. Thanks for bringing the man himself .
Great interview, thanks Lex
Really great questions and really interesting answers. The AI to help coders was an interesting idea, especially for education. Students would find this really useful.
Artificial intelligence will very much be a part of our life in the near future but i also think that we should be connected to our natural world as well
Thank you Lex.
This was fantastic.
I like the idea of a "what if" conversation, along with explanations, being a requirement of AI systems that decide legal outcomes so people can challenge the algorithm. Whether that leads to answers to why questions is going to be an interesting experiment.
36:00 totaly! You dont need to know how the universe work to boil an egg.
Form other side its good, from time to time to check what and how is working, so we can optimize the structure and to make the things more resource effective
Doubling in vertical or horizontal direction 👉 left or right 😢
Interesting quote he says is information and knowledge is good but even better is the motivation, the attitude. Soo true about this part cuz u gotta learn new things motivated. Thats why jim kwik says emotions with information. Long term memory. Humans must have good mood when we learn.
34:40 He said Java Script!! The Revolution is happening
Your book is awsome, cant wait to get 4 th edition :)
living legend...
Why do I feel like Peter is 3blue1brown, but old?
Because he’s very tame and logical/mathematical-minded.
Love it. Rock on! Thx for video !
In programming, assembling indeed makes it much faster to complete projects, but I find manufacturing far more interesting.
It is an absolute must that you should have Geordie Rose as your guest in the next podcast. You will truly know how AI will come into existence.
Why can't AI systems operate without bias as talked about at 7:00. It would seem you would just remove properties such as race and sex from the algorithms??
Race and sex and others can be derived by combining other properties like name, income, city, job etc and the algorithm can find some that we can't even think about.
@@Polyian also, if there is bias in the data sets themselves - black incarceration rates (for example) in specific districts, which may be historic judicial bias, may then be replicated in the learnt behaviour of the algorithm by weighing an area as more risk prone.
It's in the data whether you ignore it or not.
@59:24 "I'm certainly not worried about the robot apocaLISP"
Good catch :)
this is gas thanks
I think AI also needs the symbolic approach. As an example: so that an agent with AI can understand and generate humor. In the connectionist/neural approach I see that task very very far in the future.
You don’t need written or spoken language for humor.
In the beginning he says predicate logic... but he really means propositional logic, right?
4:00 Hopefully this isn't going take away my ability to make mistakes and learn from them and feel accomplished and proud doing so. Especially ethics which seem to have evolved over time. Anyway, love this field and where it's heading us.
@Explicit Relativity I don't like the social engineering and the bias removal, sounds like communism, and is suggested to the EU by Cedric Villani in a paper a few years ago, and that scares me. Equal opportunities yes, and the use of historical data on all categories in order to pinpoint the root causes from problems we might experience nowadays. I use ML/PM to solve daily practical problems not the high level stuff. I'm from Eindhoven, Techno(logy) City, can't release it my friend, although I play the shakuhachi.
@Explicit Relativity Google is a great technology company and comes with great (social) responsibility. Their realisation is a fact though. Unfortunately nowadays everything needs to be black or white, for or against, depending on your own bias. I'd rather have good discussions and be proven wrong so I learned something.
That's what I like about mr Fridman's channel, many different experts with different backgrounds. Sometimes I don't agree, many times I'm inspired.
The only thing mr Fridman should do is having a 1 dollar Patreon subscription. I'd rather support 100 people instead of 81 as that is my budget to support content creators.
after a couple of months of this video, lex saw online universities 😂😂😂
Long attentive course is ineffective and messy. Breaking down to structured parts make sense.
32:00
He may have foreseen the global pandemic. He was just 10% wrong. It happened in a year not in decade.
6:00 - why incorporate race, etc into the model?
If your dataset of human features involves race (say, it's photos), the machine learning algorithm will likely implicitly learn the race as one of high level representations.
Algorithm may learn something you did not intend it to learn. For example, US Army had a project where the goal was to classify US and Russian tanks. In lab tests system did well, but in field tests failed miserably. Turns out, all it learned to classify was the quality of the photos - of course, Russian tank photos were low quality spy images...
Management? Sir
Satyajit Roy 🎉
Pure AI versus Sicial AI?
Good idea turn on the google home and assistant into the inverse reinforcement learning collection machines. :) Ex machina idea hahaha :)
watching the last few minutes of this video after the emergence of copilot.
Cool grandpa i wish i had.
31:56 - You have no idea how possible...
Multimedia Madness 🎉😂❤
To talk to and train managers in creative field without killing people
Reading manuals is not optimized for my cognition. Too many targets....too little time.
22:58
"Part of the problem is we're seduced by our low-dimensional metaphors." I agree. Just like how squirrels are seduced by acorns.
not really
He meant that language shapes the way you think about the problem. People think in language terms because it's natural for them, but machine does not work that way. What is really happening is that machine is optimising a surface in a million or so dimensional space, and nobody really knows how to reason about such complex things.
44:00 lisp
Java + Lisp = JScheme small pure code
Human management and talking respectfully to a child like me
Here in 2024
watched
What if moocs were taught second life style? Why no one has ventured into this possibility?
Let me pay a monthly fee for apps that are on my side please. IDGAF about free cost I want a free mind.
Ooooo Utpal dutta
Advocate of developer
Insomnia 😢
And the suffering
All approaches to artificial intelligence are modern approaches.
what will happen if two AI systems have a conversation ? can they get angry :)
There was actually an experiment of this kind. Chatbots developed their own language:
www.snopes.com/fact-check/facebook-ai-developed-own-language/
(the rating is "false", but keep reading - the essence is in the nuance).
What aatyalingam podepuram 😂
give me haskell, or give me buffer-overflows. :|
Discus white holes.
Do you think it’s old black holes? I don’t think that’s the easy idea.
Earlier black holes seems an easy out.
Like nuke? At Japan 😂❤
And I am a critical mental health care 😂 teacher u know 😢 of your AI
🥹
The kindhearted drop phytogeographically command because throne especially reign during a didactic clover. cumbersome, abusive harp
A sili way of killing people with soft heart 😅 ummmmm
3 idiot umm mmmm makechudesekihasi an ki in Japan
Tumi ki ki jono go? Dance Singing painting and like a layman
You can't satisfy both the protected class criteria and accurate prediction re: recidivism because race is real and manifests itself in behavior. There is no AI ethical quandary, there is simply the fact that black people have higher recidivism rates than other demographics.
🥹🥲❤️