WEBVTT 00:00:01.100 --> 00:00:03.240 - Okay, so, good morning everyone. 00:00:03.240 --> 00:00:04.913 I'll just get started. 00:00:04.913 --> 00:00:08.840 My name is Shailesh and I give these talks 00:00:08.927 --> 00:00:11.946 almost every year so this is a very deja-vu feeling for me. 00:00:11.994 --> 00:00:13.330 The only thing different this time 00:00:13.330 --> 00:00:15.732 is the stage is slightly thinner. 00:00:15.732 --> 00:00:21.880 But great crowd, great list of talks so far. 00:00:21.880 --> 00:00:25.870 So, Daniel called me a couple of weeks ago and said 00:00:25.870 --> 00:00:27.703 "Why don't you give a keynote again?" 00:00:27.703 --> 00:00:30.690 And I said, "You know, I'm running out of things to say now." 00:00:30.690 --> 00:00:33.200 I've given four talks at different forums 00:00:33.200 --> 00:00:36.880 with The Fifth Elephant and I wasn't sure 00:00:36.880 --> 00:00:38.030 what I'm gonna talk about 00:00:38.030 --> 00:00:41.320 So, then, one of these days I was talking 00:00:41.320 --> 00:00:43.280 to one of my non-geek friends 00:00:43.280 --> 00:00:46.225 and he was very excited about what I do 00:00:46.225 --> 00:00:47.460 so he said, 'What do you do?' 00:00:47.460 --> 00:00:49.330 and I, you know, it was on the phone 00:00:49.330 --> 00:00:53.033 and I started talking to him about this, that, and the other. 00:00:53.033 --> 00:00:55.200 And for about 45 minutes I was rambling 00:00:55.200 --> 00:00:57.080 and this guy was very quiet. 00:00:57.080 --> 00:01:01.887 I didn't realize he wasn't a techie 00:01:01.887 --> 00:01:06.760 and I was going on and on and after 45 minutes I stopped 00:01:06.760 --> 00:01:09.177 and said, "Are you still there? Are you listening?" 00:01:09.177 --> 00:01:11.120 And he said, "Yeah, I'm listening. 00:01:12.020 --> 00:01:15.781 "Can you tell me what do you do again?" (audience laughs) 00:01:15.781 --> 00:01:20.280 And then I realized, how do I summarize this in two words? 00:01:20.280 --> 00:01:24.194 So then I told him, "Hey, I'm building thinking machines." 00:01:24.194 --> 00:01:26.660 And that's when he said, "Why didn't you say that before? 00:01:26.660 --> 00:01:28.420 "It was so easy to say that, right?" 00:01:28.420 --> 00:01:30.130 So that's how the title came by 00:01:30.130 --> 00:01:33.520 and obviously we're not building thinking machines 00:01:33.520 --> 00:01:36.590 but what I'm gonna talk about is towards thinking machines, right? 00:01:36.590 --> 00:01:38.946 So, we have a long way to go. 00:01:38.946 --> 00:01:42.326 So I added the word "towards" later. 00:01:42.326 --> 00:01:45.820 So what I'm gonna talk about is all over the place. 00:01:45.820 --> 00:01:48.705 I'm gonna talk about philosophy, science fiction. 00:01:48.705 --> 00:01:51.436 I'll talk about algorithms 00:01:51.436 --> 00:01:54.660 and I'm gonna talk about deep learning 00:01:54.660 --> 00:01:57.885 and how to think about things beyond deep learning. 00:01:57.885 --> 00:02:02.420 And let me give you a perspective and then we'll start. 00:02:02.420 --> 00:02:04.581 So I'll take questions at the end. 00:02:04.581 --> 00:02:05.660 It's not working. 00:02:09.479 --> 00:02:11.020 It's not working, this. 00:02:39.380 --> 00:02:40.842 That's fine. 00:02:40.842 --> 00:02:45.330 All right, so, I ended my last year's talk on this quotation 00:02:45.330 --> 00:02:47.850 So I thought I'll start on this quotation this time. 00:02:48.600 --> 00:02:50.580 So I like this quotation because it puts a lot 00:02:50.580 --> 00:02:53.719 of things into perspective of what we're doing, 00:02:53.719 --> 00:02:56.061 how our civilization got here 00:02:56.061 --> 00:02:57.490 and where we are headed. 00:02:58.230 --> 00:03:02.886 So it says, "Our technology, our machines, is part of our humanity. 00:03:02.886 --> 00:03:05.920 "We created them to extend ourselves 00:03:05.920 --> 00:03:08.130 "and that is what is unique about human beings!" 00:03:08.130 --> 00:03:11.520 And if you look at chairs, and dogs, and animals, and cats 00:03:11.520 --> 00:03:15.134 they don't create machines to extend themselves. 00:03:15.134 --> 00:03:17.330 They just have instincts and they follow their instincts. 00:03:17.330 --> 00:03:20.020 Right, that's very unique about human civilization. 00:03:20.020 --> 00:03:24.620 We've created Taj Mahal, and space flights, and internet. 00:03:24.620 --> 00:03:26.950 So we've come a very long way. 00:03:26.950 --> 00:03:29.120 So if you think about the tools, right? 00:03:29.120 --> 00:03:33.247 The cavemen had tools and now we have 00:03:33.247 --> 00:03:37.890 a completely robotic assembly line with no humans 00:03:37.890 --> 00:03:40.220 and you could turn the lights off and nothing will happen 00:03:40.220 --> 00:03:41.996 the cars will get produced, right? 00:03:41.996 --> 00:03:44.210 If you look at our transportation 00:03:44.980 --> 00:03:50.080 we have gone from just on-road, bullock carts, 00:03:50.080 --> 00:03:54.180 to massive amounts of transportation that we can do now. 00:03:56.480 --> 00:03:58.980 If you look at our ability to look further 00:03:58.980 --> 00:04:02.170 into space, again... 00:04:06.480 --> 00:04:08.680 Since Galileo, we have made a lot of progress. 00:04:08.680 --> 00:04:12.210 Recently we saw the news of Pluto flyby 00:04:12.210 --> 00:04:14.852 so now we're able to send satellites into space. 00:04:14.852 --> 00:04:17.716 If you look at the first computer we built 00:04:17.716 --> 00:04:19.300 and where we are today, right? 00:04:19.300 --> 00:04:22.210 We have a huge data center, and really 00:04:22.210 --> 00:04:24.160 if you look at the whole thing in perspective 00:04:24.160 --> 00:04:26.498 we have made an enormous amount of progress 00:04:26.498 --> 00:04:29.867 in the last so many centuries, right? 00:04:29.867 --> 00:04:32.580 So if you look just at the technical part 00:04:32.580 --> 00:04:35.300 the IT kind of intelligent machines 00:04:35.300 --> 00:04:36.670 we're not talking about mixies 00:04:36.670 --> 00:04:40.460 and other things, just look at what AI 00:04:40.460 --> 00:04:43.622 and deep learning and all this stuff has produced. 00:04:43.622 --> 00:04:45.900 Today's machines can play chess. 00:04:45.900 --> 00:04:47.750 And there's no human on the planet 00:04:47.750 --> 00:04:50.412 who can play chess better than the machine. 00:04:50.412 --> 00:04:54.708 I want to take a pause and think about where we are. 00:04:54.708 --> 00:04:56.741 There's no human on the planet 00:04:56.741 --> 00:04:58.880 who can play chess better than a machine. 00:05:00.240 --> 00:05:02.600 There's no human on the planet 00:05:02.600 --> 00:05:05.020 who can play Jeopardy better than a machine. 00:05:06.960 --> 00:05:10.970 And recently, Google came out with automatic cars 00:05:10.970 --> 00:05:13.840 so the machines can drive cars and record show 00:05:13.840 --> 00:05:18.896 that these cars are better than humans under ideal conditions 00:05:18.896 --> 00:05:21.440 And they have much less accident rates 00:05:21.440 --> 00:05:25.030 and all the accidents happened because of other humans drivers. 00:05:25.030 --> 00:05:26.350 They're not because of cars. 00:05:27.910 --> 00:05:29.998 And recently you also saw 00:05:29.998 --> 00:05:32.730 how machines are able to create pictures, right? 00:05:32.730 --> 00:05:34.420 So this is one of the things 00:05:34.420 --> 00:05:37.753 that we saw what deep learning is internally doing. 00:05:37.753 --> 00:05:39.300 And now think about all this. 00:05:39.300 --> 00:05:44.060 Just think about where machines have gone today. 00:05:44.060 --> 00:05:45.860 How many things they can do 00:05:45.860 --> 00:05:48.259 which are way beyond our imagination 00:05:48.259 --> 00:05:49.760 that machines could have done. 00:05:51.140 --> 00:05:55.040 So obviously there's a lot they've done. 00:05:55.040 --> 00:05:57.181 But can they do the following? 00:05:57.181 --> 00:05:59.244 We would want to stress the limits 00:05:59.244 --> 00:06:02.405 so one of the holy grails of AI 00:06:02.405 --> 00:06:06.032 is to have a machine have a conversation with a human being. 00:06:06.032 --> 00:06:07.858 We all know the Turing test 00:06:07.858 --> 00:06:11.109 and the repercussions of this will be huge. 00:06:11.109 --> 00:06:14.468 If you think about how we talk to the internet today 00:06:14.468 --> 00:06:18.970 we carefully craft three-word, four-word queries, right? 00:06:18.970 --> 00:06:21.950 And you know, we allow the internet to make mistakes. 00:06:21.950 --> 00:06:25.110 We craft the queries again, we take the suggestions or not. 00:06:25.110 --> 00:06:28.636 We talk to the internet like we're talking to a three-year-old. 00:06:28.636 --> 00:06:33.200 Now in the day and age needs of massive data computers, NLP 00:06:33.200 --> 00:06:37.210 and all this deep-learning stuff, imagine what a shameful thing it is 00:06:37.210 --> 00:06:40.091 to talk to a computer like a 3-year-old. 00:06:40.091 --> 00:06:43.990 So it's got the capacity of thousands of people 00:06:43.990 --> 00:06:46.395 but it can't understand language. 00:06:46.395 --> 00:06:48.200 So we need to change that. 00:06:48.200 --> 00:06:50.650 Now imagine beyond keywords what can happen. 00:06:50.650 --> 00:06:52.247 We can do question-answering 00:06:52.247 --> 00:06:54.630 but how do we do question-answering today? 00:06:54.630 --> 00:06:57.640 We have created Yahoo Answers, we have created Quora 00:06:57.640 --> 00:07:01.140 and people who type questions, we do a match. 00:07:01.140 --> 00:07:04.875 Between the questions and the answers 00:07:04.875 --> 00:07:07.600 and then we again do retrieval. 00:07:07.600 --> 00:07:09.640 We're still not answering questions. 00:07:10.520 --> 00:07:11.980 Now think about conversations. 00:07:11.980 --> 00:07:14.160 Conversation is an even more complex thing. 00:07:14.160 --> 00:07:17.528 If it works out, what are the repercussions? 00:07:17.528 --> 00:07:20.160 I don't want to study physics from my physics teacher. 00:07:20.160 --> 00:07:22.840 I want to study it from Einstein or Feynman. 00:07:23.710 --> 00:07:28.260 We already know all the language and the knowledge of these people. 00:07:28.260 --> 00:07:32.020 Can we not have a persona of a person, Feynman or Einstein 00:07:32.020 --> 00:07:34.780 and have a conversation with that person? 00:07:35.540 --> 00:07:38.520 So, just imagine the future of what will happen 00:07:38.520 --> 00:07:41.550 if we are able to just have conversations with the machines. 00:07:41.550 --> 00:07:43.590 So, there's a long way to go between 00:07:43.590 --> 00:07:46.679 keyword search and conversations. 00:07:46.679 --> 00:07:48.987 Can we discover a cure for cancer? 00:07:48.987 --> 00:07:51.000 There are a lot of diseases out there. 00:07:51.000 --> 00:07:52.820 Now, obviously there is a lot 00:07:52.820 --> 00:07:54.840 of research pharma companies are doing. 00:07:54.840 --> 00:07:56.940 There's a lot of new initiatives in how 00:07:56.940 --> 00:08:00.160 to use the high-end machine learning in pharma research. 00:08:00.990 --> 00:08:06.140 But my contention is that I believe that the cure for a lot 00:08:06.140 --> 00:08:08.647 of diseases is already out there. 00:08:08.647 --> 00:08:11.470 In all the medical literature, if somebody 00:08:11.470 --> 00:08:14.450 could actually read them, hold that knowledge 00:08:14.450 --> 00:08:19.133 in the brain, in RAM, and do interconnections 00:08:19.133 --> 00:08:21.739 we should be able to find a lot of things. 00:08:21.739 --> 00:08:22.860 But what is the problem? 00:08:22.860 --> 00:08:25.920 A single human expert, even in one field 00:08:25.920 --> 00:08:28.730 cannot keep up with that quest of knowledge, right? 00:08:28.730 --> 00:08:31.720 We'll forget some things, we won't read certain papers. 00:08:31.720 --> 00:08:34.059 And therefore, it's the other problem. 00:08:34.059 --> 00:08:37.190 We have too much knowledge and our individual brains 00:08:37.190 --> 00:08:41.378 are not capable of forming those connections in the... 00:08:41.378 --> 00:08:44.198 Because we can't even read that many documents, right? 00:08:44.198 --> 00:08:46.040 But if a machine could do it 00:08:46.040 --> 00:08:48.050 the way NLP has progressed 00:08:48.050 --> 00:08:50.690 can we not find cures or new medicine? 00:08:52.600 --> 00:08:54.940 Can I crack the next IIT Entrance Exam? 00:08:58.140 --> 00:09:01.560 You're laughing today, but you never know. 00:09:01.560 --> 00:09:03.776 Five years from now, what will happen? 00:09:03.776 --> 00:09:08.740 And we should hope that if Watson is a test of intelligence 00:09:08.740 --> 00:09:11.550 if Igloo is a test of intelligence 00:09:11.550 --> 00:09:14.191 could this not be a test of intelligence? 00:09:14.191 --> 00:09:17.560 The ability of AI system to be able 00:09:17.560 --> 00:09:20.970 to actually solve an IIT paper and get a rank 1. 00:09:23.610 --> 00:09:26.987 Can I search all the video scenes 00:09:26.987 --> 00:09:28.610 which only have a goal shot 00:09:28.610 --> 00:09:31.600 in the football videos and nothing else. 00:09:31.600 --> 00:09:34.219 I don't want to watch the rest of it. 00:09:34.219 --> 00:09:36.040 A lot of balls going here and there. 00:09:36.040 --> 00:09:37.850 I just wanna see the goal shots. 00:09:37.850 --> 00:09:39.965 Today I cannot do that. 00:09:39.965 --> 00:09:42.140 Can my machines be intelligent enough 00:09:42.140 --> 00:09:44.430 the vision part, that can actually find 00:09:44.430 --> 00:09:46.330 this is a goal, this is a goal, this is a goal 00:09:46.330 --> 00:09:48.563 the rest of it is something else. 00:09:48.563 --> 00:09:50.920 So we can imagine the applications out there. 00:09:51.820 --> 00:09:54.018 We were talking about sarcasm a lot 00:09:54.018 --> 00:09:57.370 and we all understand sarcasm is a very hard thing to do. 00:09:57.370 --> 00:10:01.542 And imagine if you could detect sarcasm, what else can you do? 00:10:01.542 --> 00:10:03.440 You're writing an email to your boss 00:10:03.440 --> 00:10:06.732 you're angry, you have written a sarcastic comment 00:10:06.732 --> 00:10:09.260 and Gmail says, "Hey, are you sure about this?" 00:10:09.260 --> 00:10:12.090 In the heat of the moment (audience laughs) 00:10:13.000 --> 00:10:14.580 can I put it this way? 00:10:15.520 --> 00:10:17.580 So, like, today we do attachments 00:10:17.580 --> 00:10:19.820 can we detect sarcasm and things like that? 00:10:20.730 --> 00:10:25.500 And to me the holy grail of AI is not really 00:10:25.500 --> 00:10:28.265 all these big things, but a very simple thing. 00:10:28.265 --> 00:10:31.080 Can a machine find a joke funny? 00:10:32.210 --> 00:10:33.879 Now there are a lot of... 00:10:33.879 --> 00:10:36.009 I don't know if you guys watch Star Trek 00:10:36.009 --> 00:10:39.850 but Data, in 300 years, 400 years from now 00:10:39.850 --> 00:10:40.990 is an android. 00:10:40.990 --> 00:10:43.280 He is capable of all these other things. 00:10:43.280 --> 00:10:46.840 He's a great supercomputer in a human form 00:10:46.840 --> 00:10:48.700 but he's still struggling with humor. 00:10:48.700 --> 00:10:50.620 That's how hard the problem is. 00:10:51.560 --> 00:10:54.240 So obviously we have a long way to go. 00:10:54.240 --> 00:10:56.380 We have come a long way and we have a long way to go. 00:10:56.380 --> 00:10:59.290 So this talk is really about the way forward. 00:11:00.960 --> 00:11:04.199 So, what do we imagine the future to be? 00:11:04.199 --> 00:11:06.340 We want something like this. 00:11:07.230 --> 00:11:09.350 Good and bad, hopefully good. 00:11:09.350 --> 00:11:10.380 We want a Jarvis, right? 00:11:10.380 --> 00:11:11.740 We all want a Jarvis 00:11:11.740 --> 00:11:13.360 who'll takes care of the chores 00:11:13.360 --> 00:11:16.870 and get rid of whatever 00:11:16.870 --> 00:11:19.350 and we all want a Jarvis right? 00:11:19.350 --> 00:11:22.531 So if you watch these movies again 00:11:22.531 --> 00:11:24.556 after watching this talk 00:11:24.556 --> 00:11:26.560 you'll have a very different perspective 00:11:26.560 --> 00:11:29.708 on what we need to do to get here. 00:11:29.708 --> 00:11:31.680 It's not gonna happen just because we're 00:11:31.680 --> 00:11:35.040 gonna make more and more Hollywood movies like this. 00:11:35.335 --> 00:11:38.552 I mean, Asimov wrote "I, Robot" in the 70s 00:11:38.705 --> 00:11:40.531 and we're still not there. 00:11:41.397 --> 00:11:44.098 It's not gonna happen because we keep doing "data science" 00:11:45.297 --> 00:11:47.840 And that's one of the reasons I wanted to do this talk 00:11:47.840 --> 00:11:49.533 'cause a lot of people keep 00:11:50.004 --> 00:11:52.273 thinking "data science is the end of the world" 00:11:52.273 --> 00:11:54.024 but there's a lot more to data science 00:11:54.024 --> 00:11:57.085 and I want to see how we can go beyond data science 00:11:57.478 --> 00:11:59.004 - and this is not data science. 00:11:59.014 --> 00:12:00.558 This is artificial intelligence. 00:12:00.558 --> 00:12:02.496 Right? So I want to draw the distinction 00:12:02.512 --> 00:12:04.646 and say how we can move beyond data science 00:12:05.058 --> 00:12:06.328 - nothing wrong with it - 00:12:06.523 --> 00:12:08.275 but it's, it's a done deal. 00:12:08.873 --> 00:12:11.408 Right? We have software you can download, 00:12:11.408 --> 00:12:13.166 you can put up whatever you want, 00:12:13.481 --> 00:12:16.650 it's a done deal. Data science has been packaged, already. 00:12:17.522 --> 00:12:20.217 Right? If you look at Microsoft Azure, 00:12:20.217 --> 00:12:22.276 or some of these other softwares, right? 00:12:22.301 --> 00:12:23.712 It has already been packaged 00:12:24.053 --> 00:12:25.963 All you have to do is download the right 00:12:25.963 --> 00:12:28.361 software, put your data in the right format, 00:12:28.361 --> 00:12:32.875 and you're done. Right? So there's nothing "great" about data science anymore. 00:12:33.840 --> 00:12:36.289 Sorry about that, but, you know, 00:12:36.533 --> 00:12:39.301 we need to jolt ourselves out of this comfort zone, and say 00:12:39.301 --> 00:12:40.941 "okay, we are all data scientists" 00:12:40.941 --> 00:12:42.053 - that's not it, right? 00:12:42.102 --> 00:12:44.557 How do we get here? How will data science get here? 00:12:45.133 --> 00:12:49.181 Alright. So, we'll get here by asking a lot of deeper questions. 00:12:49.364 --> 00:12:51.299 Right? Not the questions like 00:12:51.778 --> 00:12:54.955 "Why is this customer returning from Flipkart?", right? or 00:12:54.955 --> 00:12:58.766 "Who's -- what is the next product to recommend to somebody?", or 00:12:59.102 --> 00:13:02.025 "Which movie you're going to ask?" These are not the questions 00:13:02.457 --> 00:13:04.639 that'll take us to the next stage. Right? 00:13:04.639 --> 00:13:07.149 So the question that'll take us to the next stage is 00:13:07.456 --> 00:13:10.297 "what is learning?" Fundamentally, philosophically. 00:13:10.540 --> 00:13:13.867 "What is learning?" We see that we are learning, children are learning, 00:13:13.867 --> 00:13:16.244 everybody is going to school, we all are learning. 00:13:16.244 --> 00:13:18.356 We think that machine learning is learning, 00:13:18.466 --> 00:13:20.497 but what is learning really, right? 00:13:20.642 --> 00:13:21.994 "What is understanding?" 00:13:22.444 --> 00:13:26.539 What does that mean? What does the word "mean" mean? 00:13:28.069 --> 00:13:31.387 What is thinking? We keep saying 00:13:31.387 --> 00:13:32.882 "Oh -- I'm thinking about this" 00:13:32.882 --> 00:13:34.796 What are you doing when you're thinking? 00:13:34.967 --> 00:13:37.945 So, today I'm going to show you an equation of thinking. 00:13:37.977 --> 00:13:40.842 Okay? So, it'll be fun -- I don't claim 00:13:40.842 --> 00:13:42.764 this is - THE - equation of thinking, 00:13:42.764 --> 00:13:45.072 but I'm trying to get to that plot point 00:13:45.072 --> 00:13:46.928 where we start thinking about thinking, 00:13:47.865 --> 00:13:49.172 and not just think. 00:13:50.296 --> 00:13:54.483 "What is creativity?" Now, creativity is, 00:13:54.483 --> 00:13:57.815 if you look at an artist, or a musician, or even a scientist, 00:13:57.815 --> 00:13:59.356 we create new inventions 00:13:59.356 --> 00:14:01.417 out of the knowledge we have, 00:14:01.417 --> 00:14:06.327 and innovation is a manifestation of the knowledge in a certain form. 00:14:06.327 --> 00:14:09.405 Right? A poet creates, a musician creates -- 00:14:09.405 --> 00:14:11.203 so what is creativity? 00:14:11.203 --> 00:14:14.777 And the last question I have, here is 00:14:14.777 --> 00:14:16.881 "What is consciousness?" Right? 00:14:16.881 --> 00:14:20.210 So, ultimately, if you look at movies like "I, Robot", 00:14:20.210 --> 00:14:23.730 the word "I" from the robot is not really about 00:14:23.730 --> 00:14:27.110 the robot's great abilities at mundane tasks, 00:14:27.110 --> 00:14:29.092 but really it's about the "I" in it. 00:14:29.667 --> 00:14:34.409 "I am a conscious being", and now what are the consequences. 00:14:34.480 --> 00:14:36.232 Right? So what is consciousness, and 00:14:36.249 --> 00:14:39.188 can we have sentient machines at the end of the day, right? 00:14:39.519 --> 00:14:41.860 So, we won't go there today, maybe we'll see 00:14:41.860 --> 00:14:43.718 if we have time we'll watch a video, 00:14:44.247 --> 00:14:46.949 but I'll try to cover the bottom three and 00:14:46.949 --> 00:14:50.381 see if we can find something interesting. 00:14:50.381 --> 00:14:53.827 So, learning. Learning is one of the most basic things, 00:14:53.827 --> 00:14:55.388 we all do learning all the time. 00:14:55.686 --> 00:14:58.219 -- at least we all claim to be learning all the time. 00:14:58.478 --> 00:15:01.539 So, really, I'm going to use language and 00:15:01.539 --> 00:15:03.662 not vision at first, but language as 00:15:03.662 --> 00:15:07.122 the basis for all the examples. 99:59:59.999 --> 99:59:59.999 So, learning really is many, many things: 99:59:59.999 --> 99:59:59.999 the first thing we learn, so, you know, 99:59:59.999 --> 99:59:59.999 the greatest example of a machine learning system, or an A.I. system 99:59:59.999 --> 99:59:59.999 is a human child. 99:59:59.999 --> 99:59:59.999 And all you have to do is just observe 99:59:59.999 --> 99:59:59.999 how a baby is growing up, how he's picking language, how he's 99:59:59.999 --> 99:59:59.999 picking walking, how he's picking swimming, how he's picking tantrums, right? 99:59:59.999 --> 99:59:59.999 And you learn so much about A.I. because you're looking at the real A.I. 99:59:59.999 --> 99:59:59.999 So what is learning? I want to use that example 99:59:59.999 --> 99:59:59.999 and see how we pick up language. 99:59:59.999 --> 99:59:59.999 If I use the word -- if I start -- imagine you're reading a novel, 99:59:59.999 --> 99:59:59.999 or imagine words are coming at you one at a time: 99:59:59.999 --> 99:59:59.999 you see the word "united" - what do you think the next word would be? 99:59:59.999 --> 99:59:59.999 Right? "United States", "United Something" , whatever. 99:59:59.999 --> 99:59:59.999 then, [MIC CUTS], predicting. When we're learning, 99:59:59.999 --> 99:59:59.999 we are also simultaneously predicting. 99:59:59.999 --> 99:59:59.999 And this is one of the flaws in current machine learning: 99:59:59.999 --> 99:59:59.999 that we keep thinking that learning is separate, prediction is separate. 99:59:59.999 --> 99:59:59.999 We'll learn first, then we'll score. 99:59:59.999 --> 99:59:59.999 Right? But the human brain is not like that. 99:59:59.999 --> 99:59:59.999 We don't learn for sixty years and suddenly we start behaving. 99:59:59.999 --> 99:59:59.999 We're constantly learning and we're constantly applying that learning, 99:59:59.999 --> 99:59:59.999 and that is one of the fundamental reasons why, 99:59:59.999 --> 99:59:59.999 you know, I call the current model of machine learning 99:59:59.999 --> 99:59:59.999 like the [inaudible] which is never going to become a data-flow architecture 99:59:59.999 --> 99:59:59.999 ever, right? So that is one of the problems. 99:59:59.999 --> 99:59:59.999 So imagine what we're doing now, we are predicting what will come next.