WEBVTT 99:59:59.999 --> 99:59:59.999 Okay, so, good morning everyone. 99:59:59.999 --> 99:59:59.999 I'll just get started. 99:59:59.999 --> 99:59:59.999 My name is Shailesh and I give these talks 99:59:59.999 --> 99:59:59.999 almost every year so this is a very deja-vu feeling for me. 99:59:59.999 --> 99:59:59.999 The only thing different this time 99:59:59.999 --> 99:59:59.999 is the stage is slightly thinner. 99:59:59.999 --> 99:59:59.999 But great crowd, great list of talks so far. 99:59:59.999 --> 99:59:59.999 So, Daniel called me a couple of weeks ago and said 99:59:59.999 --> 99:59:59.999 "Why don't you give a keynote again?" 99:59:59.999 --> 99:59:59.999 And I said, "You know, I'm running out of things to say now." 99:59:59.999 --> 99:59:59.999 I've given four talks at different forums 99:59:59.999 --> 99:59:59.999 with The Fifth Elephant and I wasn't so sure 99:59:59.999 --> 99:59:59.999 what I want to talk about. 99:59:59.999 --> 99:59:59.999 So, then, one of these days I was talking 99:59:59.999 --> 99:59:59.999 to one of my non-geek friends 99:59:59.999 --> 99:59:59.999 and he was very excited about what I do 99:59:59.999 --> 99:59:59.999 so he said, 'What do you do?' 99:59:59.999 --> 99:59:59.999 and I, you know, it was on the phone 99:59:59.999 --> 99:59:59.999 and I started talking to him about this, that, and the other. 99:59:59.999 --> 99:59:59.999 And for about 45 minutes I was rambling 99:59:59.999 --> 99:59:59.999 and this guy was very quiet. 99:59:59.999 --> 99:59:59.999 I didn't realise he wasn't a techie 99:59:59.999 --> 99:59:59.999 and I was going on and on and after 45 minutes I stopped 99:59:59.999 --> 99:59:59.999 and said, "Are you still there? 99:59:59.999 --> 99:59:59.999 "Are you listening?" 99:59:59.999 --> 99:59:59.999 And he said, "Yeah, I'm listening. 99:59:59.999 --> 99:59:59.999 "Can you tell me what do you do again?" (audience laughs) 99:59:59.999 --> 99:59:59.999 And then I realized, how do I summarize this in 2 words? 99:59:59.999 --> 99:59:59.999 So then I told him, "Hey, I'm building thinking machines." 99:59:59.999 --> 99:59:59.999 And that's when he said, "Why didn't you say that before? 99:59:59.999 --> 99:59:59.999 "It was so easy to say that, right?" 99:59:59.999 --> 99:59:59.999 So that's how the title came by 99:59:59.999 --> 99:59:59.999 and obviously we're not building thinking machines 99:59:59.999 --> 99:59:59.999 but what I'm going talk about is towards thinking machines, right? 99:59:59.999 --> 99:59:59.999 So, we have a long way to go. 99:59:59.999 --> 99:59:59.999 So I added the word "towards" later. 99:59:59.999 --> 99:59:59.999 So what I'm gonna talk about is all over the place. 99:59:59.999 --> 99:59:59.999 I'm going to talk about philosophy, science fiction. 99:59:59.999 --> 99:59:59.999 I'm going to talk about algorithms 99:59:59.999 --> 99:59:59.999 and I'm going to talk about, you know, deep learning 99:59:59.999 --> 99:59:59.999 and how to think about things beyond deep learning. 99:59:59.999 --> 99:59:59.999 All right? 99:59:59.999 --> 99:59:59.999 And let me give you a perspective and then we'll start. 99:59:59.999 --> 99:59:59.999 So I'll take questions at the end. 99:59:59.999 --> 99:59:59.999 Start working this. 99:59:59.999 --> 99:59:59.999 All right, so, I ended my last year's talk on this quotation 99:59:59.999 --> 99:59:59.999 So I thought I'll start on this quotation this time. 99:59:59.999 --> 99:59:59.999 So I like this quotation because it puts a lot 99:59:59.999 --> 99:59:59.999 of things into perspective of what we're doing, 99:59:59.999 --> 99:59:59.999 how our civilisation got here, and where we're headed. 99:59:59.999 --> 99:59:59.999 So it says, "Our technology, our machines, is part of our humanity. 99:59:59.999 --> 99:59:59.999 "We created them to extend ourselves 99:59:59.999 --> 99:59:59.999 "and that is what is unique about human beings!" 99:59:59.999 --> 99:59:59.999 And if you look at chairs, and dogs, and animals, and cats 99:59:59.999 --> 99:59:59.999 they don't create machines to extend themselves. 99:59:59.999 --> 99:59:59.999 They just have instincts and they follow their instincts. 99:59:59.999 --> 99:59:59.999 Right, that's very unique about human civilization. 99:59:59.999 --> 99:59:59.999 We've created Taj Mahal, and space flights, and internet 99:59:59.999 --> 99:59:59.999 and so we've come a very long way. 99:59:59.999 --> 99:59:59.999 So if you think about the tools, right? 99:59:59.999 --> 99:59:59.999 The cavemen had tools and now we have 99:59:59.999 --> 99:59:59.999 a completely robotic assembly line with no humans 99:59:59.999 --> 99:59:59.999 and you could turn the lights off and nothing will happen 99:59:59.999 --> 99:59:59.999 the cars will get produced, right? 99:59:59.999 --> 99:59:59.999 We've gone from just on-road, bullock carts, 99:59:59.999 --> 99:59:59.999 to massive amounts of transportation that we can do now. 99:59:59.999 --> 99:59:59.999 If you look at our ability to look further 99:59:59.999 --> 99:59:59.999 into space, again, since Galileo, 99:59:59.999 --> 99:59:59.999 we have made a lot of progress. 99:59:59.999 --> 99:59:59.999 Recently we saw the news of Pluto flyby 99:59:59.999 --> 99:59:59.999 so now we're able to send satellites into space. 99:59:59.999 --> 99:59:59.999 If you look at the first computer we built 99:59:59.999 --> 99:59:59.999 and where we are today, right? 99:59:59.999 --> 99:59:59.999 We have a huge data center, and really 99:59:59.999 --> 99:59:59.999 if you look at the whole thing in perspective 99:59:59.999 --> 99:59:59.999 we have made an enormous amount of progress 99:59:59.999 --> 99:59:59.999 in the last so many centuries, right? 99:59:59.999 --> 99:59:59.999 So if you look just at the technical part 99:59:59.999 --> 99:59:59.999 the IT kind of intelligent machines 99:59:59.999 --> 99:59:59.999 we're not talking about mixies 99:59:59.999 --> 99:59:59.999 and other things, just look at what AI 99:59:59.999 --> 99:59:59.999 and deep learning, this stuff, has produced. 99:59:59.999 --> 99:59:59.999 Today's machines can play chess. 99:59:59.999 --> 99:59:59.999 And there's no human on the planet 99:59:59.999 --> 99:59:59.999 who can play chess better than the machine. 99:59:59.999 --> 99:59:59.999 I want to take a pause and think about where we are. 99:59:59.999 --> 99:59:59.999 There's no human on the planet 99:59:59.999 --> 99:59:59.999 who can play chess better than the machine. 99:59:59.999 --> 99:59:59.999 There's no human on the planet 99:59:59.999 --> 99:59:59.999 who can play Jeopardy better than the machine. 99:59:59.999 --> 99:59:59.999 And recently, Google came out with automatic cars 99:59:59.999 --> 99:59:59.999 so the machines can drive cars and record show 99:59:59.999 --> 99:59:59.999 that these cars are better than humans under ideal conditions 99:59:59.999 --> 99:59:59.999 And they have much less accident rates 99:59:59.999 --> 99:59:59.999 and all the accidents happened because of other humans drivers. 99:59:59.999 --> 99:59:59.999 They're not because of cars. 99:59:59.999 --> 99:59:59.999 And recently also saw how machines 99:59:59.999 --> 99:59:59.999 are able to create pictures, right? 99:59:59.999 --> 99:59:59.999 So this is one of the things 99:59:59.999 --> 99:59:59.999 that deep learning is internally doing. 99:59:59.999 --> 99:59:59.999 And now think about all this. 99:59:59.999 --> 99:59:59.999 Just think about where machines have gone today. 99:59:59.999 --> 99:59:59.999 How many things they can do 99:59:59.999 --> 99:59:59.999 which are way beyond our imagination 99:59:59.999 --> 99:59:59.999 that machines could have done. 99:59:59.999 --> 99:59:59.999 So obviously there's a lot they've done. 99:59:59.999 --> 99:59:59.999 But can they do the following? 99:59:59.999 --> 99:59:59.999 So obviously there's a lot they've done. 99:59:59.999 --> 99:59:59.999 So one of the holy grails of AI is to have a machine 99:59:59.999 --> 99:59:59.999 have a conversation with a human being. 99:59:59.999 --> 99:59:59.999 We all know the Turing test and the 99:59:59.999 --> 99:59:59.999 repercussions of this will be huge. 99:59:59.999 --> 99:59:59.999 We could think about how we talk to the internet today. 99:59:59.999 --> 99:59:59.999 We carefully craft word-for-word queries, right? 99:59:59.999 --> 99:59:59.999 And you know, we allow the internet to make mistakes. 99:59:59.999 --> 99:59:59.999 We craft queries again, and we take the suggestions or not. 99:59:59.999 --> 99:59:59.999 We talk to the internet like we're talking to a three-year-old. 99:59:59.999 --> 99:59:59.999 Now in the daily needs of massive data computers, NLP 99:59:59.999 --> 99:59:59.999 and all its deploying staff, imagine how shameful it is 99:59:59.999 --> 99:59:59.999 to talk to a computer like a 3-year-old. 99:59:59.999 --> 99:59:59.999 So it's got the capacity of thousands of people 99:59:59.999 --> 99:59:59.999 but it can't understand language. 99:59:59.999 --> 99:59:59.999 So we need to change that. 99:59:59.999 --> 99:59:59.999 Now imagine beyond keywords what can happen. 99:59:59.999 --> 99:59:59.999 We can do question answering 99:59:59.999 --> 99:59:59.999 but how do we do question answering today? 99:59:59.999 --> 99:59:59.999 We've created Yahoo Answers, we've created Quora 99:59:59.999 --> 99:59:59.999 where people can type questions 99:59:59.999 --> 99:59:59.999 We do a match between the questions and the answers 99:59:59.999 --> 99:59:59.999 and then we again do retrieval. 99:59:59.999 --> 99:59:59.999 So not answering questions. 99:59:59.999 --> 99:59:59.999 Now think about conversations. Conversation is an even more complex thing 99:59:59.999 --> 99:59:59.999 If it works out, what are the repercussions? I don't want to study 99:59:59.999 --> 99:59:59.999 physics from my physics teacher. I want to study from Einstein or Feynman. 99:59:59.999 --> 99:59:59.999 We already know all the language and knowledge of these people. 99:59:59.999 --> 99:59:59.999 Can we not have a persona or a person Feynman or Einstein, 99:59:59.999 --> 99:59:59.999 and have a conversation with that person, right? So just imagine the future 99:59:59.999 --> 99:59:59.999 of what will happen if we're just able to have conversations with the machines. 99:59:59.999 --> 99:59:59.999 So there's a long way to go between keyword search and conversations. 99:59:59.999 --> 99:59:59.999 Can we discover a cure for cancer? There are a lot of diseases out there. 99:59:59.999 --> 99:59:59.999 Now obviously there is a lot of research pharma companies are doing. 99:59:59.999 --> 99:59:59.999 There's a lot of new initiatives on how to use the high end machine learning 99:59:59.999 --> 99:59:59.999 in pharma research. But my contention is I believe that the cure for a lot of 99:59:59.999 --> 99:59:59.999 diseases is already out there. In all the medical literature, if somebody could 99:59:59.999 --> 99:59:59.999 actually read them, hold that knowledge in the brain, in RAM, and do interconnections 99:59:59.999 --> 99:59:59.999 we should be able to find a lot of things. But what is the problem? 99:59:59.999 --> 99:59:59.999 A single human expert, even in one field cannot keep up with that quest of 99:59:59.999 --> 99:59:59.999 knowledge, right. We forget some things, we want to read certain papers. 99:59:59.999 --> 99:59:59.999 And therefore, it's the other problem. We have too much knowledge 99:59:59.999 --> 99:59:59.999 and our individual brains are not capable of forming those connections 99:59:59.999 --> 99:59:59.999 in the - because we can't even read that many docs, right? 99:59:59.999 --> 99:59:59.999 But machines could do it, the way, and then there's progress. 99:59:59.999 --> 99:59:59.999 Can we not find cures or new medicine too. 99:59:59.999 --> 99:59:59.999 Can I crack the next IIT Entrance Exam? You laughing today, but you never know. 99:59:59.999 --> 99:59:59.999 Five years from now, what will happen? We should hope that if Watson is 99:59:59.999 --> 99:59:59.999 a test of intelligence, if Igloo is a test of intelligence, could this not be 99:59:59.999 --> 99:59:59.999 a test of intelligence. The ability of AI system to be able to 99:59:59.999 --> 99:59:59.999 actually solve an IIT paper and get a rank 1. 99:59:59.999 --> 99:59:59.999 What about, can I search all the video scenes, which only have a goal shot 99:59:59.999 --> 99:59:59.999 in the football videos and nothing else. I don't want to watch the rest of it. 99:59:59.999 --> 99:59:59.999 A lot of balls going here and there. I just want to see the goal shots. 99:59:59.999 --> 99:59:59.999 Today I cannot do that. Can my machines be intelligent enough 99:59:59.999 --> 99:59:59.999 to vision part, to actually find, this is a goal, this is a goal, this is a goal - 99:59:59.999 --> 99:59:59.999 the rest of it is something else. So we can imagine the applications now. 99:59:59.999 --> 99:59:59.999 We were talking about sarcasm a lot and we all understand sarcasm is a very hard 99:59:59.999 --> 99:59:59.999 thing to do. And imagine if you could detect sarcasm, what else can you do? 99:59:59.999 --> 99:59:59.999 You writing an email to your boss You're angry, you've written 99:59:59.999 --> 99:59:59.999 a sarcastic comment, and ? says, 'Hey are you sure about this?' 99:59:59.999 --> 99:59:59.999 In the heat of the moment, can I put it this way? 99:59:59.999 --> 99:59:59.999 So, like, today we do attachments. Can you detect sarcasm and things like that. 99:59:59.999 --> 99:59:59.999 And to me the holy grail of AI is not really all these big things, 99:59:59.999 --> 99:59:59.999 but a really simple thing. Can a machine find a joke funny? 99:59:59.999 --> 99:59:59.999 Now there are a lot of - don't know if you guys watch Star Trek - but data entry 99:59:59.999 --> 99:59:59.999 300, 400 years from now, is an android who is capable of all these other things. 99:59:59.999 --> 99:59:59.999 He's a great supercomputer in human form but he's still struggling with humans. 99:59:59.999 --> 99:59:59.999 That's how hard the problem is. So obviously we have a long way to go. 99:59:59.999 --> 99:59:59.999 We've come a long way and we have a long way to go. 99:59:59.999 --> 99:59:59.999 So this talk is really about the way forward. 99:59:59.999 --> 99:59:59.999 So, what do we imagine the future to be? We want something like this. 99:59:59.999 --> 99:59:59.999 Golden ? We all want a Jarvis, right? 99:59:59.999 --> 99:59:59.999 Who takes care of the chores and gets rid of the whatever and then we all want 99:59:59.999 --> 99:59:59.999 a Jarvis right? So if you watch these movies again, after watching this talk, 99:59:59.999 --> 99:59:59.999 you'll have a very different perspective on what we need to do to get here. 99:59:59.999 --> 99:59:59.999 It's not going to happen just because we're going to make more and more 99:59:59.999 --> 99:59:59.999 Hollywood movies like this.