9:59:59.000,9:59:59.000 Ok so good morning everyone.[br]I'll just get started. 9:59:59.000,9:59:59.000 My name is Shailesh and I give these talks[br]almost every year so this is a very 9:59:59.000,9:59:59.000 deja-vu feeling for me. The only thing[br]different this time is the stage is 9:59:59.000,9:59:59.000 slightly thinner. But great crowd.[br]Great list of talks so far. 9:59:59.000,9:59:59.000 So, Daniel called me a couple of weeks[br]ago and said, 'Why don't you give a 9:59:59.000,9:59:59.000 keynote again?' And I said, 'You know, I'm[br]running out of things to say now.' 9:59:59.000,9:59:59.000 I've given four talks at different forums[br]with The Fifth Elephant and I wasn't so 9:59:59.000,9:59:59.000 sure what I want to talk about. So, then[br]one of these days I was talking to one of 9:59:59.000,9:59:59.000 my non-geek friends and he was very[br]excited about what I do, so he said, 9:59:59.000,9:59:59.000 'What do you do?' and I, you know, it was[br]on the phone and I started talking to him 9:59:59.000,9:59:59.000 about this, that and the other. And for[br]about 45 minutes I was rambling 9:59:59.000,9:59:59.000 and this guy was very quiet. I didn't[br]realise he wasn't a techie and I was 9:59:59.000,9:59:59.000 going on and on, and after 45 minutes[br]I stopped and said, 9:59:59.000,9:59:59.000 'Are you still there? Are you listening?'[br]And he said, 'Yeah, I'm listening. 9:59:59.000,9:59:59.000 Can you tell me what do you do again?'[br](audience laughs) 9:59:59.000,9:59:59.000 And then I realised, how do I summarise[br]this in 2 words? 9:59:59.000,9:59:59.000 So then I told him, 'Hey, I'm building[br]thinking machines.' 9:59:59.000,9:59:59.000 And that's when he said, 'Why didn't you[br]say that before. 9:59:59.000,9:59:59.000 It was so easy to say that, right?'[br]So that's how the title came by, 9:59:59.000,9:59:59.000 and obviously we're not building[br]thinking machines but 9:59:59.000,9:59:59.000 what I'm going talk about is towards[br]thinking machines, right? 9:59:59.000,9:59:59.000 So, we have a long way to go. So I[br]added the word 'towards' later. 9:59:59.000,9:59:59.000 So what I'm gonna talk about is all over[br]the place. I'm going to talk about 9:59:59.000,9:59:59.000 philosophy, science fiction. I'm going to[br]talk about algorithms and 9:59:59.000,9:59:59.000 I'm going to talk about, you know, deep[br]learning and how to think about things 9:59:59.000,9:59:59.000 beyond deep learning. Alright?[br]And let me give you a perspective 9:59:59.000,9:59:59.000 and then we'll start. So I'll take[br]questions at the end. 9:59:59.000,9:59:59.000 Start working this. 9:59:59.000,9:59:59.000 Alright, so, I ended my last year's talk[br]on this quotation 9:59:59.000,9:59:59.000 So I thought I'll start on this quotation[br]this time. 9:59:59.000,9:59:59.000 So I like this quotation because it puts[br]a lot of things into perspective of 9:59:59.000,9:59:59.000 what we're doing, how our civilisation got[br]here, and where we're headed. 9:59:59.000,9:59:59.000 So it says, "Our technology, our machines,[br]is part of our humanity. 9:59:59.000,9:59:59.000 We created them to extend ourselves, and[br]that is what is unique about human beings" 9:59:59.000,9:59:59.000 And if you look at chairs, and dogs, and[br]animals, and cats 9:59:59.000,9:59:59.000 They don't create machines to extend[br]themselves. They just have instincts 9:59:59.000,9:59:59.000 and they follow their instincts. Right,[br]that's very unique 9:59:59.000,9:59:59.000 about human civilisation. We've created[br]Taj Mahal, and space flights, and internet 9:59:59.000,9:59:59.000 And so we've come a very long way.[br]So if you think about the tools, right? 9:59:59.000,9:59:59.000 The cavemen had tools and now we have[br]a completely robotic assembly line 9:59:59.000,9:59:59.000 with no humans and you could turn the[br]lights off and nothing will happen. 9:59:59.000,9:59:59.000 The car would get ???, right? We've gone[br]from just on-road, bullock carts, 9:59:59.000,9:59:59.000 to massive amounts of transportation we[br]can do now. 9:59:59.000,9:59:59.000 If you look at our ability to look further in[br]the space, again, since Galileo, 9:59:59.000,9:59:59.000 we've made a lot of progress, ???[br]he's certainly a thousand years off our 9:59:59.000,9:59:59.000 Pluto fly by. So now we're able to send[br]satellites into space. 9:59:59.000,9:59:59.000 If you look at the first computer we built[br]and where we are today, right? 9:59:59.000,9:59:59.000 We have a huge data centre, and really, if[br]you look at the whole thing in perspective 9:59:59.000,9:59:59.000 we have made an enormous amount of[br]progress in the last so many centuries. 9:59:59.000,9:59:59.000 So if youlook just at the technical part,[br]the IT kind of intelligent machines, 9:59:59.000,9:59:59.000 we're not talking about mixies? and other[br]things, just look at what AI 9:59:59.000,9:59:59.000 and deep learning, this stuff, has[br]produced. Today's machines can play chess. 9:59:59.000,9:59:59.000 And there's no human on the planet who can[br]play chess better than the machine. 9:59:59.000,9:59:59.000 I want to take a pause and think about[br]where we are. 9:59:59.000,9:59:59.000 There's no human on the planet who can[br]play chess better than the machine. 9:59:59.000,9:59:59.000 There's no human on the planet who can[br]play Jeopardy better than the machine. 9:59:59.000,9:59:59.000 And recently, Google came up with[br]automatic cars, so the machine can 9:59:59.000,9:59:59.000 drive cars and record show, that this cars[br]are better than humans under rider? 9:59:59.000,9:59:59.000 conditions. And they have much less[br]accident rates, and all the accidents 9:59:59.000,9:59:59.000 happened because of other humans drivers.[br]They're not because of cars. 9:59:59.000,9:59:59.000 And recently also saw how machines are[br]able to create pictures, right, so this is 9:59:59.000,9:59:59.000 one of the things that deep learning is[br]internally doing. 9:59:59.000,9:59:59.000 And now think about all this. Just think[br]about where machines have gone today. 9:59:59.000,9:59:59.000 How many things they can do which are[br]way beyond our imagination 9:59:59.000,9:59:59.000 that machines could have done.[br]So obviously there's a lot they've done. 9:59:59.000,9:59:59.000 But can they do the following?[br]We would want to stress their limits 9:59:59.000,9:59:59.000 So one of the holy grails of AI is to have[br]a machine have a conversation with 9:59:59.000,9:59:59.000 a human being. We all know the Turing test[br]and the repercussions of this will be huge 9:59:59.000,9:59:59.000 We could think about how we talk to the[br]internet today. We carefully craft word 9:59:59.000,9:59:59.000 for word queries, right, and you know, we[br]allow the internet to make mistakes 9:59:59.000,9:59:59.000 We craft queries again, and we take the[br]suggestions or not 9:59:59.000,9:59:59.000 We talk to the internet like we're talking[br]to a 3-year-old 9:59:59.000,9:59:59.000 Now in the daily needs of massive data[br]computers, NLP? and all its deploying 9:59:59.000,9:59:59.000 staff, imagine how shameful it is to talk[br]to a computer like a 3-year-old. 9:59:59.000,9:59:59.000 So it's got the capacity of thousands of[br]people but it can't understand language. 9:59:59.000,9:59:59.000 So we need to change that. Now imagine[br]beyond keywords what can happen 9:59:59.000,9:59:59.000 We can do question answering, but how do[br]we do question answering today? 9:59:59.000,9:59:59.000 We've created Yahoo Answers. We've created[br]Quora, where people can type questions 9:59:59.000,9:59:59.000 We do a match between the questions[br]and the answers, and then 9:59:59.000,9:59:59.000 we again do retrieval. So not answering[br]questions. 9:59:59.000,9:59:59.000 Now think about conversations.[br]Conversation is an even more complex thing 9:59:59.000,9:59:59.000 If it works out, what are the[br]repercussions? I don't want to study 9:59:59.000,9:59:59.000 physics from my physics teacher. I want to[br]study from Einstein or Feynman. 9:59:59.000,9:59:59.000 We already know all the language and[br]knowledge of these people. 9:59:59.000,9:59:59.000 Can we not have a persona or a person[br]Feynman or Einstein, 9:59:59.000,9:59:59.000 and have a conversation with that person,[br]right? So just imagine the future 9:59:59.000,9:59:59.000 of what will happen if we're just able to[br]have conversations with the machines. 9:59:59.000,9:59:59.000 So there's a long way to go between[br]keyword search and conversations. 9:59:59.000,9:59:59.000 Can we discover a cure for cancer?[br]There are a lot of diseases out there. 9:59:59.000,9:59:59.000 Now obviously there is a lot of research[br]pharma companies are doing. 9:59:59.000,9:59:59.000 There's a lot of new initiatives on how[br]to use the high end machine learning 9:59:59.000,9:59:59.000 in pharma research. But my contention is[br]I believe that the cure for a lot of 9:59:59.000,9:59:59.000 diseases is already out there. In all the[br]medical literature, if somebody could 9:59:59.000,9:59:59.000 actually read them, hold that knowledge in[br]the brain, in RAM, and do interconnections 9:59:59.000,9:59:59.000 we should be able to find a lot of things.[br]But what is the problem? 9:59:59.000,9:59:59.000 A single human expert, even in one field[br]cannot keep up with that quest of 9:59:59.000,9:59:59.000 knowledge, right. We forget some things,[br]we want to read certain papers. 9:59:59.000,9:59:59.000 And therefore, it's the other problem. [br]We have too much knowledge 9:59:59.000,9:59:59.000 and our individual brains are not[br]capable of forming those connections 9:59:59.000,9:59:59.000 in the - because we can't even read that[br]many docs, right? 9:59:59.000,9:59:59.000 But machines could do it, the way, and[br]then there's progress. 9:59:59.000,9:59:59.000 Can we not find cures or new medicine[br]too. 9:59:59.000,9:59:59.000 Can I crack the next IIT Entrance Exam?[br]You laughing today, but you never know. 9:59:59.000,9:59:59.000 Five years from now, what will happen?[br]We should hope that if Watson is 9:59:59.000,9:59:59.000 a test of intelligence, if Igloo is a test[br]of intelligence, could this not be 9:59:59.000,9:59:59.000 a test of intelligence.[br]The ability of AI system to be able to 9:59:59.000,9:59:59.000 actually solve an IIT paper and get a[br]rank 1. 9:59:59.000,9:59:59.000 What about, can I search all the video[br]scenes, which only have a goal shot 9:59:59.000,9:59:59.000 in the football videos and nothing else.[br]I don't want to watch the rest of it. 9:59:59.000,9:59:59.000 A lot of balls going here and there.[br]I just want to see the goal shots. 9:59:59.000,9:59:59.000 Today I cannot do that.[br]Can my machines be intelligent enough 9:59:59.000,9:59:59.000 to vision part, to actually find, this is[br]a goal, this is a goal, this is a goal - 9:59:59.000,9:59:59.000 the rest of it is something else.[br]So we can imagine the applications now. 9:59:59.000,9:59:59.000 We were talking about sarcasm a lot and we[br]all understand sarcasm is a very hard 9:59:59.000,9:59:59.000 thing to do. And imagine if you could[br]detect sarcasm, what else can you do? 9:59:59.000,9:59:59.000 You writing an email to your boss[br]You're angry, you've written 9:59:59.000,9:59:59.000 a sarcastic comment, and ? says,[br]'Hey are you sure about this?' 9:59:59.000,9:59:59.000 In the heat of the moment, can[br]I put it this way? 9:59:59.000,9:59:59.000 So, like, today we do attachments. Can you[br]detect sarcasm and things like that. 9:59:59.000,9:59:59.000 And to me the holy grail of AI is not[br]really all these big things, 9:59:59.000,9:59:59.000 but a really simple thing. Can a machine[br]find a joke funny? 9:59:59.000,9:59:59.000 Now there are a lot of - don't know if you[br]guys watch Star Trek - but data entry 9:59:59.000,9:59:59.000 300, 400 years from now, is an android who[br]is capable of all these other things. 9:59:59.000,9:59:59.000 He's a great supercomputer in human form[br]but he's still struggling with humans. 9:59:59.000,9:59:59.000 That's how hard the problem is.