A Fireside Chat with Turing Award Winner Geoffrey Hinton, Pioneer of Deep Learning (Google I/O'19)

In this rare interview since (jointly) winning the 2018 Turing Award for his work on neural networks, hear about the conceptual and engineering breakthroughs that have made deep neural networks a critical element of computing. Their research has allowed artificial intelligence technologies to progress at a rate that was not possible in the past and has reinvented the way technology is built.
Watch more #io19 here: Inspiration at Google I/O 2019 Playlist → goo.gle/2LkBwCF
TensorFlow at Google I/O 2019 Playlist → bit.ly/2GW7ZJM
Google I/O 2019 All Sessions Playlist → goo.gle/io19allsessions
Learn more on the I/O Website → google.com/io
Subscribe to the TensorFlow Channel → bit.ly/TensorFlow1
Get started at → www.tensorflow.org/
Speaker(s): Geoffrey Hinton, Nicholas Thompson
TDAA69 event: Google I/O 2019; re_ty: Publish; fullname: Geoffrey Hinton, Nicholas Thompson;

[@jsn355]

25:04 How excited he gets when asked about dreams :) just an amazing scientist

[@BR-hi6yt]

I love the way Geoffrey thinks, almost the way I do rather than using mathematical language to describe stuff that's computational. Universe uses computation rather than math imo.

[@graysonyin8955]

A great conversation comes from good understanding from both sides, insightful questions and insightful answers. Seems like the interviewer could have understand Deep Learning and Hinton's work better before starting the interview.

[@jossydere3930]

Geoffrey Hinton is living legend of this century period !!!

[@PistachoGirasol]

Oh, I get it, he is from WIRED...

[@sukyansingh2910]

ad hominem fallacy

[@PS1212]

I AM SOOOO GLAD I WATCHED THIS!!!

[@JeremiahTownsend]

The interviewer seemed to have walked in with (incorrect) assumptions, and when the interview didn’t go as planned, he couldn’t adapt. Really took away from the presentation.

[@blasttrash]

Well the Hinton dude was also missing out some questions early on. Like the interviewer asked at what stage in Hinton;s life did he come up with this idea or think that this is how brain might work, Hinton kept dodging the question. I really wanted the answer for that.

[@TheReferrer72]

@@blasttrash why is it a dodge, every kid knows how the brain works.... they just don't know the fined detail.

[@babeeshmohanan7686]

It's seems like he is interrogating instead of interviewing..

[@deeXaeed]

Hinton says 'bound'. Subtitles say 'band'.

[@ShaulKedem]

This was excellent, thank you both

[@luisluiscunha]

This was great. Thank you very much. Nice interview. And no need to be so humble, as at about 14:20, although I understand, given the guest being Hinton.

[@gyimahfrancis60]

Love this Geoffrey guy

[@user-xq8si8pz3h]

I would like to translate this video, is it possible to open for translation?

[@littech4637]

The discussion about machine and dreams reminds me of a scene from i-Robot.

[@hafty9975]

realized halfway through I had automatic captions on which are there because of the discoveries he's talking about

[@zhuuxi]

Was this video shoot in 120 fps?

[@diannashuller5692]

39:01 come in a little late but I am thrilled to listen to your presentation I think it's supports My photographs that are holographic views of rocks that have the history of our world magnified by computers and knowing that the removal of input in my life has possibly created the connection recognition to dreams and other realms not fully recognized by the society where I live thank you for sharing fully enjoyed it

[@abstractTYPE]

aha, mhm, yeap, right, ok...

[@TheAcolossus]

you missed 'ha'

[@royeyono6512]

I want to hear the remaining 2 theories

[@BiancaAguglia]

At 37:08 the transcript for the videos says, "AI, which meant you're logic inspired and you do manipulations of cymbal strings." 😁 So that's the reason AI systems are not as advanced as we want them to be: they're manipulating cymbal strings instead of working on symbol strings. #funny_closed_captioning_fails 😁 Great interview. Geoff is one of those people who can talk about complex concepts in a way that even non-experts can understand.

[@d_wigglesworth]

Isn't it ironic? A, presumably, deep learning (ie a non-logic-based) AI System made this particular error while transcribing informed, expert comments on the shortcomings of logic-based AI systems. Furthermore, "Cymbal" is just another name for a gong... GONG! .... But I totally respect the apparent promise of the deep learning approach to AI. ... imho, It will take merely (?) sufficient training of a suitably structured deep learning NN to automatically produce comments like this one.

[@peter_castle]

awesome!

[@jorjiang1]

anyone got a link to dream theory 3 and 4?

[@GautamSingh-yn9cb]

Is it viva going on in school ?

[@dhoroniboruah]

19:14 Biochemistry and Molecular Biology.

[@OHOHOHCOME]

13:50 There is a handful of researchers developing neural networks that perform reasoning, and last year Stanford published one of the first reasoning models that work for visual reasoning. In the next few years I believe we will see their technology applied to other areas and whoever can capitalize on this will be the next millionaire. I also think we should invest more into making use of graph structures as all knowledge can be modelled with graphs.

[@franciscolima278]

How about "could you PLEASE explain"?

[@muralisethuraman5754]

Is there a specific paper that Geoffrey Hinton is referring to at 6:22?

[@cheukkinpoon4428]

I didnt read any paper but I GUESS by band he may mean something like bandwidth of signals sending through the network. Data could be considered as signals passing into the network and the number sending through the links are signals. So you can plot the distribution of those incoming signals for each neuron and you can tell how the distributions are. If there are many extreme values in the plot, it could probably mean that there are many extreme cases that "surprise" the neuron too much. For a sigmoid activation, there is no big difference between a "surprise" and a very big "surprise", in another words the neuron would not be able to pass on the information about the difference of those data points.

[@conorheins8880]

@@cheukkinpoon4428 the word is 'bound', as in an upper bound on surprise. This is the same as negative (log) evidence. by approximating the log evidence with a bound, you implement approximate bayesian inference. so it's not a property of the distributions of incoming signals, the bound on surprise is a functional of the generative model embodied by the neural network itself (the whole configuration of its weights). this bound he's referring to is called the variational free energy. the line of work that works on optimizing this 'bound on surprise' was developed by people like Geoff Hinton, Richard Zemel, and Radford Neal in the 1990s. there are several papers that he was referring to with this comment. the most famous one is called 'the helmholtz machine'

[@cheukkinpoon4428]

@@conorheins8880 thanks for pointing to the right dirrction!

[@mattizzle81]

I love the way he explains things. Explanations straight from the mind of one of the pioneers. If I ever get into a 'debate' with someone over AI who doesn't know who this man is, yet says he is wrong, I just laugh, and yes there are a lot of know-it-alls out there who are saying neural nets are bunk and 'not related at all' to how the human brain works. He says right out it is *inspired* by the human brain. Not an exact true to life replica (yet), but inspired by it. Yet some will say it is bunk and AI is a fraud, etc etc blah blah blah. Well, the tech he is explaining right now is put into practice and works every day.

[@babeeshmohanan7686]

Really helpful for understanding Real AI . He is right we are making AI not a Brain. We are just taking brain as a model for creating a auto self learning and understanding machine

[@hakankanplay]

Aye das my prof

[@waqasakram117]

Today my deep nets predict 2 dreams that i will dream tonight.

[@user-ic1vr8dk6z]

Review settings

[@alfcnz]

14:14 Nicholas Thompson: «The human brain is not necessarily the most efficient computational machine ever created» WHATTTT??? You must be kidding me, or, you have no whatsoever clue what you're actually saying. God. I'm speechless.

[@willrocks41]

It's clearly not in some cases. E.g think about how much energy it takes for a human brain to do long division vs a computer. Regardless, I think he fumbled his words a bit there, the sentiment is more that the human brain is not the most efficient computational machine that could be created. Which, I think, is hard to argue against.

[@alfcnz]

@@willrocks41, a human brain is not optimised for divisions (or algebraic computations). The things that it performs astonishingly well is perception, which we managed to tackle only recently with our machines, and we still struggle to be as efficient as our brains. So... about divisions, I'd argue the most efficient machine is my portable CASIO calculator. But then, that would be a non general purpose processor. So, I think I'm still of the opinion that the brain is indeed the most computationally efficient general purpose machine ever created / existed. Or is it not?

[@deeXaeed]

The interviewer could have learnt a bit more about AI.

[@AnthonyBecker9]

31:18 Google I/O 2020 lmao

[@limitless.discomfort]

Lol here looking for I/O 2020/2021 Lmao

[@caglardemir5339]

Awful interviewer

[@AnthonyBecker9]

He's not that bad. He keeps him talking and doesn't interrupt too much, and lets the audience clap.

[@VinBhaskara_]

terrible interviewer