1 00:00:18,200 --> 00:00:26,190 PATRICK WINSTON: Welcome to 6034. 2 00:00:26,190 --> 00:00:28,230 I don't know if I can deal with this microphone. 3 00:00:28,230 --> 00:00:29,405 We'll see what happens. 4 00:00:29,405 --> 00:00:30,380 It's going to be a good year. 5 00:00:30,380 --> 00:00:33,530 We've got [INAUDIBLE] a bunch of interesting people. 6 00:00:33,530 --> 00:00:35,790 It's always interesting to see what people named their 7 00:00:35,790 --> 00:00:39,820 children two decades ago. 8 00:00:39,820 --> 00:00:42,510 And I find they were overwhelmed with Emilys. 9 00:00:42,510 --> 00:00:45,420 And there are not too many Peters, Pauls, and Marys, but 10 00:00:45,420 --> 00:00:50,540 enough to call forth a suitable song at some point. 11 00:00:50,540 --> 00:00:55,630 We have lots of Jesses of both genders. 12 00:00:55,630 --> 00:00:57,220 We have a [INAUDIBLE] 13 00:00:57,220 --> 00:00:59,890 of both genders. 14 00:00:59,890 --> 00:01:02,810 And we have a Duncan, where's Duncan? 15 00:01:02,810 --> 00:01:04,080 There you are, Duncan. 16 00:01:04,080 --> 00:01:06,900 You've changed your hairstyle. 17 00:01:06,900 --> 00:01:10,130 I want to assure use that the Thane of Cawdor is not taking 18 00:01:10,130 --> 00:01:12,390 the course this semester. 19 00:01:12,390 --> 00:01:14,890 What I'm going to do is tell you about artificial 20 00:01:14,890 --> 00:01:18,680 intelligence today, and what this subject is about. 21 00:01:18,680 --> 00:01:24,100 There's been about a 10% percent turnover in the roster 22 00:01:24,100 --> 00:01:25,320 in the last 24 hours. 23 00:01:25,320 --> 00:01:29,680 I expect another 10% turnover in the next 24 hours, too. 24 00:01:29,680 --> 00:01:32,600 So I know many of you are sightseers, wanting to know if 25 00:01:32,600 --> 00:01:33,720 this is something you want to do. 26 00:01:33,720 --> 00:01:36,140 So I'm going to tell you about what we're going to do this 27 00:01:36,140 --> 00:01:40,229 semester, and what you'll know when you get out of here. 28 00:01:40,229 --> 00:01:43,620 I'm going to walk you through this outline. 29 00:01:43,620 --> 00:01:45,539 I'm going to start by talking about what artificial 30 00:01:45,539 --> 00:01:48,630 intelligence is, and why we do it. 31 00:01:48,630 --> 00:01:50,160 And then I'll give you a little bit of the history of 32 00:01:50,160 --> 00:01:54,000 artificial intelligence, and conclude with some of the 33 00:01:54,000 --> 00:01:56,570 covenants by which we run the course. 34 00:01:56,570 --> 00:01:58,140 One of which is no laptops, please. 35 00:02:00,900 --> 00:02:05,620 I'll explain why we have these covenants at the end. 36 00:02:05,620 --> 00:02:07,360 So what is it? 37 00:02:07,360 --> 00:02:12,220 Well, it must have something to do with thinking. 38 00:02:12,220 --> 00:02:15,280 So let's start up here, a definition of artificial 39 00:02:15,280 --> 00:02:21,820 intelligence, by saying that it's about thinking, 40 00:02:21,820 --> 00:02:23,850 whatever that is. 41 00:02:23,850 --> 00:02:26,100 My definition of artificial intelligence has 42 00:02:26,100 --> 00:02:28,560 to be rather broad. 43 00:02:28,560 --> 00:02:31,710 So we're going to say it's not only about thinking. 44 00:02:31,710 --> 00:02:39,110 It's also about perception, and it's about action. 45 00:02:42,610 --> 00:02:46,060 And if this were a philosophy class, then I'd stop right 46 00:02:46,060 --> 00:02:49,120 there and just say, in this subject we're going to talk 47 00:02:49,120 --> 00:02:52,370 about problems involving thinking, 48 00:02:52,370 --> 00:02:54,860 perception, and action. 49 00:02:54,860 --> 00:02:57,890 But this is not a philosophy class. 50 00:02:57,890 --> 00:02:59,050 This a Course six class. 51 00:02:59,050 --> 00:03:00,350 It's an engineering school class. 52 00:03:00,350 --> 00:03:02,000 It's an MIT class. 53 00:03:02,000 --> 00:03:03,940 So we need more than that. 54 00:03:03,940 --> 00:03:12,380 And therefore we're going to talk about models that are 55 00:03:12,380 --> 00:03:20,329 targeted at thinking, perception, and action. 56 00:03:20,329 --> 00:03:23,800 And this should not be strange to you, because model making 57 00:03:23,800 --> 00:03:26,900 is what MIT is about. 58 00:03:26,900 --> 00:03:29,540 You run into someone at a bar, or relative asks you what you 59 00:03:29,540 --> 00:03:33,200 do at MIT, the right knee jerk reaction is to say, we learned 60 00:03:33,200 --> 00:03:35,130 how to build models. 61 00:03:35,130 --> 00:03:36,990 That's what we do at MIT. 62 00:03:36,990 --> 00:03:40,570 We build the models using differential equations. 63 00:03:40,570 --> 00:03:43,450 We build models using probabilities. 64 00:03:43,450 --> 00:03:47,829 We build models using physical and computational simulations. 65 00:03:47,829 --> 00:03:50,300 Whatever we do, we build models. 66 00:03:50,300 --> 00:03:55,710 Even in humanities class, MIT approach is to make models 67 00:03:55,710 --> 00:04:00,290 that we can use to explain the past, predict the future, 68 00:04:00,290 --> 00:04:03,050 understand the subject, and control the world. 69 00:04:03,050 --> 00:04:04,290 That's what MIT is about. 70 00:04:04,290 --> 00:04:06,400 And that's what this subject is about, too. 71 00:04:06,400 --> 00:04:10,650 And now, our models are models of thinking. 72 00:04:10,650 --> 00:04:12,060 So you might say, if I take this 73 00:04:12,060 --> 00:04:13,840 classic will I get smarter? 74 00:04:13,840 --> 00:04:14,760 And the answer is yes. 75 00:04:14,760 --> 00:04:15,500 You will get smarter. 76 00:04:15,500 --> 00:04:17,980 Because you'll have better models of your own thinking, 77 00:04:17,980 --> 00:04:21,120 not just the subject matter of the subject, but better models 78 00:04:21,120 --> 00:04:23,840 of your own thinking. 79 00:04:23,840 --> 00:04:25,370 So models targeted at thinking, 80 00:04:25,370 --> 00:04:27,670 perception, and action. 81 00:04:27,670 --> 00:04:29,460 We know that's not quite enough, because in order to 82 00:04:29,460 --> 00:04:33,790 have a model, you have to have representation. 83 00:04:33,790 --> 00:04:37,040 So let's say that artificial intelligence is about 84 00:04:37,040 --> 00:04:52,470 representations that support the making of models to 85 00:04:52,470 --> 00:04:55,040 facilitate an understanding of thinking, 86 00:04:55,040 --> 00:04:57,500 perception, and action. 87 00:04:57,500 --> 00:04:59,409 Now you might say to me, well what's a representation? 88 00:04:59,409 --> 00:05:00,170 And what good can it do? 89 00:05:00,170 --> 00:05:02,780 So I'd like to take a brief moment to tell you about 90 00:05:02,780 --> 00:05:03,540 gyroscopes. 91 00:05:03,540 --> 00:05:06,380 Many of you have friends in mechanical engineering. 92 00:05:06,380 --> 00:05:10,980 One of the best ways embarrass them is to say here's a 93 00:05:10,980 --> 00:05:12,210 bicycle wheel. 94 00:05:12,210 --> 00:05:16,720 And if I spin it, and blow hard on it right here, on the 95 00:05:16,720 --> 00:05:19,110 edge of the wheel, is going to turn over 96 00:05:19,110 --> 00:05:23,160 this way or this way? 97 00:05:23,160 --> 00:05:26,250 I guarantee that what they will do is they'll put their 98 00:05:26,250 --> 00:05:29,400 hand in an arthritic posture called the right hand screw 99 00:05:29,400 --> 00:05:35,870 rule, aptly named because people who use it tend to get 100 00:05:35,870 --> 00:05:41,460 the right answer about 50% of the time. 101 00:05:41,460 --> 00:05:44,560 But we're never going to make that mistake again. 102 00:05:44,560 --> 00:05:45,780 Because we're electrical engineers, 103 00:05:45,780 --> 00:05:46,820 not mechanical engineers. 104 00:05:46,820 --> 00:05:49,300 And we know about representation. 105 00:05:49,300 --> 00:05:51,290 What we're going to do is we're going to think about it 106 00:05:51,290 --> 00:05:52,486 a little bit. 107 00:05:52,486 --> 00:05:54,780 And we're going to use some duct tape to help us think 108 00:05:54,780 --> 00:05:57,730 about just one piece of the wheel. 109 00:05:57,730 --> 00:06:01,270 So I want you to just think about that piece of the wheel 110 00:06:01,270 --> 00:06:03,050 as the wheel comes flying over the top, and I 111 00:06:03,050 --> 00:06:04,810 blow on it like that. 112 00:06:04,810 --> 00:06:08,090 What's going to happen to that one piece? 113 00:06:08,090 --> 00:06:10,520 It's going to go off that way, right? 114 00:06:10,520 --> 00:06:13,400 And the next piece is going to go off that way too. 115 00:06:13,400 --> 00:06:17,610 So when it comes over, it has to go that way. 116 00:06:17,610 --> 00:06:19,440 Let me do some ground f here just to be sure. 117 00:06:22,200 --> 00:06:23,450 It's very powerful feeling. 118 00:06:26,760 --> 00:06:28,730 Try it. 119 00:06:28,730 --> 00:06:29,540 We need a demonstration. 120 00:06:29,540 --> 00:06:33,010 I don't anybody think that I'm cheating, here. 121 00:06:33,010 --> 00:06:39,840 So let's just twist it one way or the other. 122 00:06:39,840 --> 00:06:42,100 So that's powerful pull, isn't it. 123 00:06:42,100 --> 00:06:46,820 Alex is now never going to get the gyroscope wrong, because 124 00:06:46,820 --> 00:06:49,870 he's got the right representation. 125 00:06:49,870 --> 00:06:52,130 So much of what you're going to accumulate in this subject 126 00:06:52,130 --> 00:06:54,560 is a suite of representations that will help you to build 127 00:06:54,560 --> 00:06:56,990 programs that are intelligent. 128 00:06:56,990 --> 00:07:00,330 But I want to give you a second example, one a little 129 00:07:00,330 --> 00:07:01,300 bit more computational. 130 00:07:01,300 --> 00:07:03,990 But one of which was very familiar to you by the time 131 00:07:03,990 --> 00:07:06,690 you went to first grade, in most cases. 132 00:07:06,690 --> 00:07:09,610 It's the problem of the farmer, the fox, the goose, 133 00:07:09,610 --> 00:07:11,020 and the grain. 134 00:07:11,020 --> 00:07:14,070 There's a river, a leaky rowboat that can only carry 135 00:07:14,070 --> 00:07:17,340 the farmer, and one of his four possessions. 136 00:07:17,340 --> 00:07:18,080 So what's the right 137 00:07:18,080 --> 00:07:21,220 representation for this problem? 138 00:07:21,220 --> 00:07:24,370 It might be a picture of the farmer. 139 00:07:24,370 --> 00:07:29,570 It might be a poem about the situation, perhaps a haiku. 140 00:07:29,570 --> 00:07:33,240 We know that those are not the right representation. 141 00:07:33,240 --> 00:07:38,340 Somehow, we get the sense that the right representation most 142 00:07:38,340 --> 00:07:42,440 involve something about the location of the participants 143 00:07:42,440 --> 00:07:44,270 in this scenario. 144 00:07:44,270 --> 00:07:48,480 So we might draw a picture that looks like this. 145 00:07:48,480 --> 00:07:54,150 There's the scenario, and here in glorious green, 146 00:07:54,150 --> 00:07:58,450 representing our algae infested rivers is the river. 147 00:07:58,450 --> 00:08:05,306 And here's the farmer, the fox, the goose, and the grain. 148 00:08:05,306 --> 00:08:07,900 An initial situation. 149 00:08:07,900 --> 00:08:10,743 Now there are other situations like this one, for example. 150 00:08:13,360 --> 00:08:19,200 We have the river, and the farmer, and the 151 00:08:19,200 --> 00:08:21,590 goose is on that side. 152 00:08:21,590 --> 00:08:26,160 And the fox and the grain is on that side. 153 00:08:26,160 --> 00:08:31,520 And we know that the farmer can execute a movement from 154 00:08:31,520 --> 00:08:34,929 one situation to another. 155 00:08:34,929 --> 00:08:36,980 So now we're getting somewhere where with the problem. 156 00:08:36,980 --> 00:08:39,120 This is at MIT approach to the farmer, fox, 157 00:08:39,120 --> 00:08:40,010 goose, and grain problem. 158 00:08:40,010 --> 00:08:41,669 It might have stumped you when you were a little kid. 159 00:08:45,370 --> 00:08:46,870 How many such situations are there? 160 00:08:50,180 --> 00:08:52,632 What do you think, Tanya? 161 00:08:52,632 --> 00:08:56,280 It looks to me like all four of individuals can be on one 162 00:08:56,280 --> 00:08:58,740 side or the other. 163 00:08:58,740 --> 00:09:01,480 So for every position the farmer can be, each of the 164 00:09:01,480 --> 00:09:03,530 other things can be on either side of the river. 165 00:09:06,060 --> 00:09:09,500 So it would be two to the fourth she says aggressively 166 00:09:09,500 --> 00:09:12,060 and without hesitation. 167 00:09:12,060 --> 00:09:14,020 Yes, two to the fourth, 16 possibilities. 168 00:09:14,020 --> 00:09:17,590 So we could actually draw out the entire graph. 169 00:09:17,590 --> 00:09:19,770 It's small enough. 170 00:09:19,770 --> 00:09:24,620 There's another position over here with the farmer, fox, 171 00:09:24,620 --> 00:09:25,832 goose, and grain. 172 00:09:25,832 --> 00:09:28,560 And in fact that's the one we want. 173 00:09:28,560 --> 00:09:32,660 And if we draw out the entire graph, it looks like this. 174 00:09:35,910 --> 00:09:40,250 This is a graph of the situations and the allowed 175 00:09:40,250 --> 00:09:42,960 connections between them. 176 00:09:42,960 --> 00:09:47,280 Why are there not 16? 177 00:09:47,280 --> 00:09:48,420 Because the other-- 178 00:09:48,420 --> 00:09:49,170 how many have I got? 179 00:09:49,170 --> 00:09:49,625 Four? 180 00:09:49,625 --> 00:09:51,720 10? 181 00:09:51,720 --> 00:09:55,600 The others are situations in which somebody gets eaten. 182 00:09:55,600 --> 00:09:58,490 So we don't want to go to any of those places. 183 00:09:58,490 --> 00:10:00,850 So having got the representation, something 184 00:10:00,850 --> 00:10:02,750 magical has happened. 185 00:10:02,750 --> 00:10:04,030 We've got our constraints exposed. 186 00:10:18,420 --> 00:10:19,860 And that's why we build representations. 187 00:10:19,860 --> 00:10:23,080 That's whey you algebra in high school, because algebraic 188 00:10:23,080 --> 00:10:25,410 notation exposes the constraints that make it 189 00:10:25,410 --> 00:10:29,630 possible to actually figure out how many customers you get 190 00:10:29,630 --> 00:10:31,510 for the number of advertisements you place in 191 00:10:31,510 --> 00:10:32,760 the newspaper. 192 00:10:34,420 --> 00:10:36,870 So artificial intelligence is about constraints exposed by 193 00:10:36,870 --> 00:10:39,760 representations that support models targeted to thinking-- 194 00:10:39,760 --> 00:10:41,450 actually there's one more thing, too. 195 00:10:41,450 --> 00:10:42,800 Not quite done. 196 00:10:42,800 --> 00:10:48,500 Because after all, in the end, we have to build programs. 197 00:10:48,500 --> 00:11:04,820 So it's about algorithms enabled by constraints exposed 198 00:11:04,820 --> 00:11:09,510 by representations that model targeted thinking, perception, 199 00:11:09,510 --> 00:11:11,000 and action. 200 00:11:11,000 --> 00:11:15,520 So these algorithms, or we might call them just as well 201 00:11:15,520 --> 00:11:18,620 procedures, or we might call them just as well methods, 202 00:11:18,620 --> 00:11:20,030 whatever you like. 203 00:11:20,030 --> 00:11:21,980 These are the stuff of what artificial intelligence is 204 00:11:21,980 --> 00:11:24,860 about-- methods, algorithms, representations. 205 00:11:24,860 --> 00:11:27,470 I'd like to give you one more example. 206 00:11:27,470 --> 00:11:29,570 It's something we call, in artificial intelligence, 207 00:11:29,570 --> 00:11:31,770 generated test. 208 00:11:31,770 --> 00:11:33,640 And it's such a simple idea, you'll never hear it again in 209 00:11:33,640 --> 00:11:34,840 this subject. 210 00:11:34,840 --> 00:11:37,130 But it's an idea you need to add to your repertoire of 211 00:11:37,130 --> 00:11:40,780 problem solving methods, techniques, procedures, and 212 00:11:40,780 --> 00:11:42,900 algorithms. 213 00:11:42,900 --> 00:11:44,150 So here's how it works. 214 00:11:48,380 --> 00:11:52,370 Maybe I can explain to best by starting off with an example. 215 00:11:52,370 --> 00:11:55,590 Here's a tree leaf I picked off a tree on 216 00:11:55,590 --> 00:11:56,760 the way over to class. 217 00:11:56,760 --> 00:11:59,950 I hope it's not the last of the species. 218 00:11:59,950 --> 00:12:01,540 What is it, what kind of tree? 219 00:12:05,680 --> 00:12:06,720 I don't know. 220 00:12:06,720 --> 00:12:09,870 I never did learn my trees, or my colors, or my 221 00:12:09,870 --> 00:12:11,910 multiplication tables. 222 00:12:11,910 --> 00:12:17,060 So I have to go back to this book, the Audubon Society 223 00:12:17,060 --> 00:12:18,870 Field Guide to North American Trees. 224 00:12:18,870 --> 00:12:21,160 And how would I solve the problem? 225 00:12:21,160 --> 00:12:21,790 It's pretty simple. 226 00:12:21,790 --> 00:12:26,490 I just turn the pages one at a time, until I find something 227 00:12:26,490 --> 00:12:28,700 that looks like this leaf. 228 00:12:28,700 --> 00:12:33,190 And then I discover it's a sycamore, or something. 229 00:12:33,190 --> 00:12:35,930 MIT's full of them. 230 00:12:35,930 --> 00:12:39,660 So when I do that, I do something very intuitive, very 231 00:12:39,660 --> 00:12:41,070 natural, something you do all the time. 232 00:12:41,070 --> 00:12:42,550 But we're going to give it a name. 233 00:12:42,550 --> 00:12:43,850 We're going to call it generate and test. 234 00:12:55,510 --> 00:12:59,790 And generate and test method consists of generating some 235 00:12:59,790 --> 00:13:04,260 possible solutions, feeding them into a box that tests 236 00:13:04,260 --> 00:13:10,760 them, and then out the other side comes mostly failures. 237 00:13:10,760 --> 00:13:14,890 But every once in a while we get something that succeeds 238 00:13:14,890 --> 00:13:17,240 and pleases us. 239 00:13:17,240 --> 00:13:18,960 That's what I did with the leaf. 240 00:13:18,960 --> 00:13:21,340 But now you have a name for it. 241 00:13:21,340 --> 00:13:25,490 Once you have a name for something, you 242 00:13:25,490 --> 00:13:26,860 get power over it. 243 00:13:26,860 --> 00:13:28,970 You can start to talk about it. 244 00:13:28,970 --> 00:13:32,180 So I can say, if you're doing a generate and test approach 245 00:13:32,180 --> 00:13:37,280 to a problem, you better build a generator with certain 246 00:13:37,280 --> 00:13:40,210 properties that make generators good. 247 00:13:40,210 --> 00:13:42,940 For example, they should not be redundant. 248 00:13:42,940 --> 00:13:46,210 They shouldn't give you the same solution twice. 249 00:13:46,210 --> 00:13:49,680 They should be informable. 250 00:13:49,680 --> 00:13:52,620 They should be able to absorb information such as, this is a 251 00:13:52,620 --> 00:13:53,650 deciduous tree. 252 00:13:53,650 --> 00:13:56,260 Don't bother looking at the conifers. 253 00:13:56,260 --> 00:13:58,130 So once you have a name for something, you can start 254 00:13:58,130 --> 00:13:59,370 talking about. 255 00:13:59,370 --> 00:14:02,580 And that vocabulary gives you power. 256 00:14:02,580 --> 00:14:06,610 So we call this the Rumpelstiltskin Principle 257 00:14:06,610 --> 00:14:10,530 perhaps The first of our powerful ideas for the day. 258 00:14:10,530 --> 00:14:12,180 This subject is full of powerful ideas. 259 00:14:12,180 --> 00:14:14,300 There will be some in every class. 260 00:14:14,300 --> 00:14:16,480 Rumpelstiltskin Principle says that once you can name 261 00:14:16,480 --> 00:14:18,250 something, you get power over it. 262 00:14:18,250 --> 00:14:19,220 You know what that little thing is on 263 00:14:19,220 --> 00:14:20,470 the end of your shoelace? 264 00:14:23,650 --> 00:14:24,280 It's interesting. 265 00:14:24,280 --> 00:14:26,480 She's gesturing like mad. 266 00:14:26,480 --> 00:14:28,920 That's something we'll talk about later, too-- 267 00:14:28,920 --> 00:14:32,070 motor stuff, and how it helps us think. 268 00:14:32,070 --> 00:14:33,310 What is it? 269 00:14:33,310 --> 00:14:35,500 No one knows? 270 00:14:35,500 --> 00:14:36,780 It's an ag something, right? 271 00:14:36,780 --> 00:14:38,910 It's an aglet, very good. 272 00:14:38,910 --> 00:14:41,460 So once you have the name, you can start to talk about. 273 00:14:41,460 --> 00:14:44,320 You can say the purpose of an aglet is pretty much like the 274 00:14:44,320 --> 00:14:45,590 whipping on the end of a rope. 275 00:14:45,590 --> 00:14:48,140 It keeps the thing from unwinding. 276 00:14:48,140 --> 00:14:51,140 Now you have a place to hang that knowledge. 277 00:14:51,140 --> 00:14:54,730 So we're talking about this frequently from now into the 278 00:14:54,730 --> 00:14:56,680 rest of the semester, the power of being 279 00:14:56,680 --> 00:14:58,010 able to name things. 280 00:14:58,010 --> 00:15:01,850 Symbolic labels give us power over concepts. 281 00:15:01,850 --> 00:15:04,290 While we're here I should also say that this is a very simple 282 00:15:04,290 --> 00:15:06,570 idea, generate and test. 283 00:15:06,570 --> 00:15:09,770 And you might be tempted to say to someone, we learned 284 00:15:09,770 --> 00:15:11,470 about generate and test today. 285 00:15:11,470 --> 00:15:14,290 But it's a trivial idea. 286 00:15:14,290 --> 00:15:17,030 The word trivial is a word I would like you to purge from 287 00:15:17,030 --> 00:15:22,340 your vocabulary, because it's a very dangerous label. 288 00:15:22,340 --> 00:15:25,970 The reason it's dangerous is because there's a difference 289 00:15:25,970 --> 00:15:27,360 between trivial and simple. 290 00:15:27,360 --> 00:15:30,310 What is it? 291 00:15:30,310 --> 00:15:33,320 What's the difference between labeling something as trivial 292 00:15:33,320 --> 00:15:34,570 and calling it simple? 293 00:15:34,570 --> 00:15:35,820 Yes? 294 00:15:39,440 --> 00:15:40,450 Exactly so. 295 00:15:40,450 --> 00:15:44,950 He says that simple can be powerful, and trivial makes it 296 00:15:44,950 --> 00:15:48,830 sound like it's not only simple, but of little worth. 297 00:15:48,830 --> 00:15:53,060 So many MIT people miss opportunities, because they 298 00:15:53,060 --> 00:15:55,270 have a tendency to think that ideas aren't important unless 299 00:15:55,270 --> 00:15:57,660 they're complicated. 300 00:15:57,660 --> 00:15:59,965 But the most simple ideas in artificial intelligence are 301 00:15:59,965 --> 00:16:02,120 often the most powerful. 302 00:16:02,120 --> 00:16:04,130 We could teach an artificial intelligence course to you 303 00:16:04,130 --> 00:16:06,180 that would be so full of mathematics it would make a 304 00:16:06,180 --> 00:16:09,150 Course 18 professor gag. 305 00:16:09,150 --> 00:16:11,900 But those ideas would be merely gratuitously 306 00:16:11,900 --> 00:16:15,590 complicated, and gratuitously mathematical, and gratuitously 307 00:16:15,590 --> 00:16:16,940 not simple. 308 00:16:16,940 --> 00:16:18,790 Simple ideas are often the most powerful. 309 00:16:22,460 --> 00:16:24,570 So where are we so far? 310 00:16:24,570 --> 00:16:27,165 We talked about the definition. 311 00:16:27,165 --> 00:16:29,660 We talked about an example of a method. 312 00:16:29,660 --> 00:16:32,560 Showed you a representation, and perhaps also talked about 313 00:16:32,560 --> 00:16:34,220 the first idea, too. 314 00:16:34,220 --> 00:16:36,690 You've got the representation right, you're 315 00:16:36,690 --> 00:16:38,440 often almost done. 316 00:16:38,440 --> 00:16:41,030 Because with this representation, they can 317 00:16:41,030 --> 00:16:44,580 immediately see that there are just two solutions to this 318 00:16:44,580 --> 00:16:48,930 problem, something that wouldn't have occurred to us 319 00:16:48,930 --> 00:16:50,630 when we were little kids, and didn't think to draw the 320 00:16:50,630 --> 00:16:51,880 [? state ?] diagram. 321 00:16:56,640 --> 00:16:57,890 There's still one more thing. 322 00:17:01,410 --> 00:17:04,740 In the past, and in other places, artificial 323 00:17:04,740 --> 00:17:09,108 intelligence is often taught as purely about reasoning. 324 00:17:09,108 --> 00:17:12,515 But we solve problems with our eyes, as well as 325 00:17:12,515 --> 00:17:15,420 our symbolic apparatus. 326 00:17:15,420 --> 00:17:18,858 And you solved that problem with your eyes. 327 00:17:18,858 --> 00:17:22,839 So I like to reinforce that by giving you a little puzzle. 328 00:17:22,839 --> 00:17:24,630 Let's see, who's here? 329 00:17:30,370 --> 00:17:32,420 I don't see [? Kambe, ?] but I'll bet he's from Africa. 330 00:17:32,420 --> 00:17:34,610 Is anyone from Africa? 331 00:17:34,610 --> 00:17:37,180 No one's from Africa? 332 00:17:37,180 --> 00:17:38,960 No? 333 00:17:38,960 --> 00:17:40,210 Well so much the better-- 334 00:17:46,090 --> 00:17:47,760 because they would know the answer to the puzzle. 335 00:17:47,760 --> 00:17:49,230 Here's the puzzle. 336 00:17:49,230 --> 00:17:53,270 How many countries in Africa does the Equator cross? 337 00:17:56,690 --> 00:17:58,290 Would anybody be willing to stake their 338 00:17:58,290 --> 00:17:59,750 life on their answer? 339 00:18:02,520 --> 00:18:03,770 Probably not. 340 00:18:06,910 --> 00:18:16,550 Well, now let me repeat the question. 341 00:18:16,550 --> 00:18:19,475 How many countries in Africa does the Equator cross? 342 00:18:21,995 --> 00:18:23,040 Yeah, six. 343 00:18:23,040 --> 00:18:25,970 What happened is a miracle. 344 00:18:25,970 --> 00:18:29,770 The miracle is that I have communicated with you through 345 00:18:29,770 --> 00:18:34,660 language, and your language system commanded your visual 346 00:18:34,660 --> 00:18:38,590 system to execute a program that involves scanning across 347 00:18:38,590 --> 00:18:41,390 that line, counting as you go. 348 00:18:41,390 --> 00:18:43,770 And then your vision system came back to your language 349 00:18:43,770 --> 00:18:45,950 system and said, six. 350 00:18:45,950 --> 00:18:47,575 And that is a miracle. 351 00:18:47,575 --> 00:18:50,600 And without understanding that miracle, we'll never have a 352 00:18:50,600 --> 00:18:53,330 full understanding of the nature of intelligence. 353 00:18:53,330 --> 00:18:55,720 But that kind of problem solving is the kind of problem 354 00:18:55,720 --> 00:18:57,950 solving I wish we could teach you a lot about it. 355 00:18:57,950 --> 00:19:00,420 But we can't teach you about stuff we don't understand. 356 00:19:00,420 --> 00:19:01,500 We [INAUDIBLE] 357 00:19:01,500 --> 00:19:02,750 for that. 358 00:19:05,540 --> 00:19:08,930 That's a little bit about the definition and some examples. 359 00:19:08,930 --> 00:19:10,460 What's it for? 360 00:19:10,460 --> 00:19:12,530 We can deal with that very quickly. 361 00:19:12,530 --> 00:19:16,440 If we're engineers, it's for building smarter programs. 362 00:19:16,440 --> 00:19:20,690 It's about building a tool kit of representations and methods 363 00:19:20,690 --> 00:19:22,660 that make it possible to build smarter programs. 364 00:19:22,660 --> 00:19:25,660 And you will find, these days, that you can't build a big 365 00:19:25,660 --> 00:19:28,520 system without having embedded in it somewhere the ideas that 366 00:19:28,520 --> 00:19:31,230 we talk about in the subject. 367 00:19:31,230 --> 00:19:33,420 If you're a scientist, there's a somewhat different 368 00:19:33,420 --> 00:19:33,970 motivation. 369 00:19:33,970 --> 00:19:36,420 But it amounts to studying the same sorts of things. 370 00:19:36,420 --> 00:19:40,800 If you're a scientist, you're interested in what it is that 371 00:19:40,800 --> 00:19:43,800 enables us to build a computational account of 372 00:19:43,800 --> 00:19:45,110 intelligence. 373 00:19:45,110 --> 00:19:46,910 That's the part that I do. 374 00:19:46,910 --> 00:19:49,620 But most this subject is going to be about the other part, 375 00:19:49,620 --> 00:19:52,440 the part that makes it possible for you to build 376 00:19:52,440 --> 00:19:53,510 smarter programs. 377 00:19:53,510 --> 00:19:56,990 And some of it will be about what it is that makes us 378 00:19:56,990 --> 00:20:02,900 different from the chimpanzees with whom we share an enormous 379 00:20:02,900 --> 00:20:05,130 fraction of our DNA. 380 00:20:05,130 --> 00:20:08,010 It used to be thought that we share 95% of our DNA with 381 00:20:08,010 --> 00:20:09,840 chimpanzees. 382 00:20:09,840 --> 00:20:13,660 Then it went up to 98. 383 00:20:13,660 --> 00:20:15,470 Thank God it stopped about there. 384 00:20:15,470 --> 00:20:16,770 Then it actually went back a little bit. 385 00:20:16,770 --> 00:20:23,350 I think we're back down to 94. 386 00:20:23,350 --> 00:20:25,190 How about if we talk a little bit now about the history of 387 00:20:25,190 --> 00:20:28,940 AI, so we can see how we got to where we are today? 388 00:20:28,940 --> 00:20:31,040 This will also be a history of AI that tells you a little bit 389 00:20:31,040 --> 00:20:32,420 about what you'll learn in this course. 390 00:20:36,610 --> 00:20:42,730 It all started with Lady Lovelace, the world's first 391 00:20:42,730 --> 00:20:47,750 programmer, Who wrote programs about 100 years before there 392 00:20:47,750 --> 00:20:50,500 were computers to run them. 393 00:20:50,500 --> 00:20:54,140 But it's interesting that even in 1842, people were hassling 394 00:20:54,140 --> 00:20:57,200 her about whether computers could get really smart. 395 00:20:57,200 --> 00:21:03,370 And she said, "The analytical engine has no pretensions to 396 00:21:03,370 --> 00:21:05,500 originate anything. 397 00:21:05,500 --> 00:21:10,080 It can do whatever we know how to order it to perform." 398 00:21:10,080 --> 00:21:13,600 Screwball idea that persists to this day. 399 00:21:13,600 --> 00:21:15,730 Nevertheless, that was the origin of it all. 400 00:21:15,730 --> 00:21:17,330 That was the beginning of the discussions. 401 00:21:17,330 --> 00:21:22,050 And then nothing much happened until about 1950, when Alan 402 00:21:22,050 --> 00:21:24,110 Turing wrote his famous paper, which 403 00:21:24,110 --> 00:21:27,090 introduced the Turing test. 404 00:21:27,090 --> 00:21:29,340 Of course, Alan Turing had previously won the Second 405 00:21:29,340 --> 00:21:34,090 World War by breaking the German code, the Ultra Code, 406 00:21:34,090 --> 00:21:36,240 for which the British government rewarded him by 407 00:21:36,240 --> 00:21:38,180 driving him to suicide, because he happened to be 408 00:21:38,180 --> 00:21:41,040 homosexual. 409 00:21:41,040 --> 00:21:44,260 But Turing wrote his paper in 1950, and that was the first 410 00:21:44,260 --> 00:21:48,720 milestone after Lady Lovelace's comment in 1842. 411 00:21:48,720 --> 00:21:53,140 And then the modern era really began with a paper written by 412 00:21:53,140 --> 00:21:56,120 Marvin Minsky in 1960, titled "Steps Toward Artificial 413 00:21:56,120 --> 00:21:59,530 Intelligence." And it wasn't a long after that Jim 414 00:21:59,530 --> 00:22:01,490 [? Slagle, ?] 415 00:22:01,490 --> 00:22:03,830 a nearly blind graduate student, wrote a program that 416 00:22:03,830 --> 00:22:07,270 did symbolic integration. 417 00:22:07,270 --> 00:22:09,850 Not adding up area under a curve, but doing symbolic 418 00:22:09,850 --> 00:22:11,870 integration just like you learn to do in high school 419 00:22:11,870 --> 00:22:14,470 when you're a freshman. 420 00:22:14,470 --> 00:22:16,560 Now on Monday, we're going to talk about this program. 421 00:22:16,560 --> 00:22:18,730 And you're going to understand exactly how it works. 422 00:22:18,730 --> 00:22:21,570 And you can write one yourself. 423 00:22:21,570 --> 00:22:23,730 And we're going to reach way back in time to look at that 424 00:22:23,730 --> 00:22:27,400 program because, in one day discussing it, talking about 425 00:22:27,400 --> 00:22:29,500 it, will be in itself a miniature artificial 426 00:22:29,500 --> 00:22:30,360 intelligence course. 427 00:22:30,360 --> 00:22:35,250 Because it's so rich with important ideas. 428 00:22:35,250 --> 00:22:39,310 So that's the dawn age, early dawn age. 429 00:22:39,310 --> 00:22:49,810 This was the age of speculation, and this was the 430 00:22:49,810 --> 00:22:51,060 dawn age in here. 431 00:22:53,820 --> 00:22:57,600 So in that early dawn age , the integration program took 432 00:22:57,600 --> 00:22:58,390 the world by storm. 433 00:22:58,390 --> 00:23:00,280 Because not everybody knows how to do integration. 434 00:23:00,280 --> 00:23:03,580 And someone, everyone, thought that if we can do integration 435 00:23:03,580 --> 00:23:05,420 today, the rest of intelligence will be figured 436 00:23:05,420 --> 00:23:06,882 out tomorrow. 437 00:23:06,882 --> 00:23:10,580 Too bad for our side it didn't work out that way. 438 00:23:10,580 --> 00:23:14,200 Here's another dawn age program, the Eliza 439 00:23:14,200 --> 00:23:15,220 [? thing ?]. 440 00:23:15,220 --> 00:23:17,590 But I imagine you'd prefer a demonstration to 441 00:23:17,590 --> 00:23:19,455 just reading it, right? 442 00:23:19,455 --> 00:23:22,870 Do you prefer a demonstration? 443 00:23:22,870 --> 00:23:24,120 Let's see if we can demonstrate it. 444 00:24:08,250 --> 00:24:11,400 This is left over from a hamentashen debate of a couple 445 00:24:11,400 --> 00:24:13,411 of years ago. 446 00:24:13,411 --> 00:24:15,050 How do you spell hamentashen, anybody know? 447 00:24:19,450 --> 00:24:20,600 I sure hope that's right. 448 00:24:20,600 --> 00:24:21,390 It doesn't matter. 449 00:24:21,390 --> 00:24:22,640 Something interesting will come. 450 00:24:25,920 --> 00:24:28,580 OK, your choice. 451 00:24:28,580 --> 00:24:31,310 Teal? 452 00:24:31,310 --> 00:24:32,560 Burton House? 453 00:24:35,890 --> 00:24:37,140 Teal. 454 00:24:47,810 --> 00:24:49,710 So that's dawn age AI. 455 00:24:49,710 --> 00:24:53,700 And no one ever took that stuff seriously, except that 456 00:24:53,700 --> 00:24:54,900 it was a fun [INAUDIBLE] 457 00:24:54,900 --> 00:24:57,980 project level thing to work out some matching 458 00:24:57,980 --> 00:24:59,810 programs, and so on. 459 00:24:59,810 --> 00:25:01,310 The integration program was serious. 460 00:25:01,310 --> 00:25:03,240 This one wasn't. 461 00:25:03,240 --> 00:25:06,940 This was serious, programs that do geometric analogy, 462 00:25:06,940 --> 00:25:09,020 problems of the kind you find on intelligence tests. 463 00:25:09,020 --> 00:25:09,990 Do you have the answer to this? 464 00:25:09,990 --> 00:25:11,340 A is to B as C is to what? 465 00:25:14,260 --> 00:25:15,610 That would be 2, I guess. 466 00:25:18,270 --> 00:25:19,520 What's the second best answer? 467 00:25:22,070 --> 00:25:24,750 And the theories of the program that solve these 468 00:25:24,750 --> 00:25:28,990 problems are pretty much identical to what you just 469 00:25:28,990 --> 00:25:30,460 figured out. 470 00:25:30,460 --> 00:25:35,970 In the first case you deleted the inside figure. 471 00:25:35,970 --> 00:25:38,280 And the second case is, the reason you got four is because 472 00:25:38,280 --> 00:25:44,340 you deleted the outside part and grew the inside part. 473 00:25:44,340 --> 00:25:46,670 There's another one. 474 00:25:46,670 --> 00:25:49,510 I think this was the hardest one it got, or the easiest one 475 00:25:49,510 --> 00:25:50,110 it didn't get. 476 00:25:50,110 --> 00:25:51,300 I've forgotten. 477 00:25:51,300 --> 00:25:54,170 A is to B as C is to 3. 478 00:25:57,490 --> 00:26:00,180 In the late dawn age, we began to turn our attention from 479 00:26:00,180 --> 00:26:03,010 purely symbolic reasoning to thinking a little bit about 480 00:26:03,010 --> 00:26:04,600 perceptual apparatus. 481 00:26:04,600 --> 00:26:07,410 And programs were written that could figure out the nature of 482 00:26:07,410 --> 00:26:11,800 shapes and forms, such as that. 483 00:26:11,800 --> 00:26:14,200 And it's interesting that those programs had the same 484 00:26:14,200 --> 00:26:19,920 kind of difficulty with this that you do. 485 00:26:19,920 --> 00:26:22,410 Because now, having deleted all the edges, everything 486 00:26:22,410 --> 00:26:23,840 becomes ambiguous. 487 00:26:23,840 --> 00:26:26,790 And it may be a series of platforms, or it may be a 488 00:26:26,790 --> 00:26:27,615 series of-- 489 00:26:27,615 --> 00:26:30,520 can you see the saw blade sticking up if you go through 490 00:26:30,520 --> 00:26:31,770 the reversal? 491 00:26:33,790 --> 00:26:35,750 Programs were written that could learn from a small 492 00:26:35,750 --> 00:26:37,600 number of examples. 493 00:26:37,600 --> 00:26:41,490 Many people think of computer learning as involving leading 494 00:26:41,490 --> 00:26:46,030 some neural net to submission with thousands of trials. 495 00:26:46,030 --> 00:26:48,280 Programs were written in the early dawn age that learned 496 00:26:48,280 --> 00:26:51,980 that an arch is something that has to have the flat part on 497 00:26:51,980 --> 00:26:55,180 top, and the two sides can't touch, and the top may or may 498 00:26:55,180 --> 00:26:56,430 not be a wedge. 499 00:26:59,630 --> 00:27:01,880 In the late dawn age, though, the most important thing, 500 00:27:01,880 --> 00:27:06,302 perhaps, was what you look at with me on Wednesday next. 501 00:27:06,302 --> 00:27:09,120 It's a rule-based expert systems. 502 00:27:09,120 --> 00:27:14,740 And a program was written at Stanford that did diagnosis of 503 00:27:14,740 --> 00:27:16,520 bacterial infections of the blood. 504 00:27:16,520 --> 00:27:21,040 It turned out to do it better than most doctors, most 505 00:27:21,040 --> 00:27:22,860 general practitioners. 506 00:27:22,860 --> 00:27:26,990 It was never used, curiously enough. 507 00:27:26,990 --> 00:27:31,360 Because nobody cares what your problem actually is. 508 00:27:31,360 --> 00:27:33,420 They just give you a broad spectrum antibiotic that'll 509 00:27:33,420 --> 00:27:34,670 kill everything. 510 00:27:36,700 --> 00:27:39,700 But this late dawn age system, the so-called [INAUDIBLE] 511 00:27:39,700 --> 00:27:46,440 system, was the system that launched a thousand companies, 512 00:27:46,440 --> 00:27:48,960 because people started building expert systems built 513 00:27:48,960 --> 00:27:50,040 on that technology. 514 00:27:50,040 --> 00:27:52,130 Here's that you don't know you used, or that was 515 00:27:52,130 --> 00:27:53,430 used on your behalf. 516 00:27:53,430 --> 00:27:56,570 If you go through, for example, the Atlanta airport, 517 00:27:56,570 --> 00:27:59,380 your airplane is parked by a rule-based expert system that 518 00:27:59,380 --> 00:28:03,430 knows how to park aircraft effectively. 519 00:28:03,430 --> 00:28:09,841 It saves Delta Airlines about to $0.5 million a day of jet 520 00:28:09,841 --> 00:28:13,780 fuel by being all smarter about how to park them. 521 00:28:13,780 --> 00:28:15,760 So that's an example of an expert system that does a 522 00:28:15,760 --> 00:28:17,160 little bit of good for a lot of people. 523 00:28:19,680 --> 00:28:20,600 There's Deep Blue. 524 00:28:20,600 --> 00:28:27,930 That takes us to the next stage beyond the age of expert 525 00:28:27,930 --> 00:28:30,410 systems, and the business age. 526 00:28:30,410 --> 00:28:33,250 It takes us into this age here, which I call the 527 00:28:33,250 --> 00:28:43,260 bulldozer age, because this is the time when people began to 528 00:28:43,260 --> 00:28:46,390 see that we had at our disposal 529 00:28:46,390 --> 00:28:48,700 unlimited amounts of computing. 530 00:28:48,700 --> 00:28:50,880 And frequently you can substitute computing for 531 00:28:50,880 --> 00:28:52,930 intelligence. 532 00:28:52,930 --> 00:28:56,420 So no one would say that Deep Blue does anything like what a 533 00:28:56,420 --> 00:28:59,230 human chess master does. 534 00:28:59,230 --> 00:29:04,180 But nevertheless, Deep Blue, by processing data like a 535 00:29:04,180 --> 00:29:06,700 bulldozer processes gravel, was able to 536 00:29:06,700 --> 00:29:07,950 beat the world champion. 537 00:29:11,910 --> 00:29:13,100 So what's the right way? 538 00:29:13,100 --> 00:29:15,390 That's the age we're in right now. 539 00:29:15,390 --> 00:29:18,420 I will of course be inducing programs for those ages as we 540 00:29:18,420 --> 00:29:19,940 go through the subject. 541 00:29:19,940 --> 00:29:22,950 There is a question of what age we're in right now. 542 00:29:22,950 --> 00:29:24,840 And it's always dangerous to name an age when 543 00:29:24,840 --> 00:29:26,640 you're in it, I guess. 544 00:29:26,640 --> 00:29:30,660 I like to call at the age of the right way. 545 00:29:30,660 --> 00:29:33,270 And this is an age when we begin to realize that that 546 00:29:33,270 --> 00:29:36,750 definition up there is actually a little incomplete, 547 00:29:36,750 --> 00:29:41,630 because much of our intelligence has to do not 548 00:29:41,630 --> 00:29:47,920 with thinking, perception, and action acting separately, but 549 00:29:47,920 --> 00:29:55,510 with loops that tie all those together. 550 00:29:55,510 --> 00:29:57,490 We had one example with Africa. 551 00:29:57,490 --> 00:30:02,270 Here's another example drawn from a program that has been 552 00:30:02,270 --> 00:30:04,960 under development, and continues to be, in my 553 00:30:04,960 --> 00:30:06,210 laboratory. 554 00:30:11,490 --> 00:30:13,500 We're going to ask the system to imagine something. 555 00:30:18,670 --> 00:30:19,610 SYSTEM: OK. 556 00:30:19,610 --> 00:30:22,406 I will imagine that a ball falls into a bowl. 557 00:30:33,098 --> 00:30:33,584 OK. 558 00:30:33,584 --> 00:30:38,930 I will imagine that a man runs into a woman. 559 00:30:38,930 --> 00:30:40,240 PATRICK WINSTON: You see, it does the best that it can if 560 00:30:40,240 --> 00:30:43,440 it doesn't have a good memory of what these situations 561 00:30:43,440 --> 00:30:45,005 actually involve. 562 00:30:45,005 --> 00:30:48,130 But having imagined the scene it can then-- 563 00:30:48,130 --> 00:30:48,620 SYSTEM: Yes. 564 00:30:48,620 --> 00:30:52,110 I have learned from experience that contact between a man and 565 00:30:52,110 --> 00:30:56,240 a woman appeared because a man runs into a woman. 566 00:30:56,240 --> 00:30:58,110 PATRICK WINSTON: Having imagined the scene, it can 567 00:30:58,110 --> 00:31:01,740 then read the answers using its visual apparatus on the 568 00:31:01,740 --> 00:31:03,580 scene that it imagined. 569 00:31:03,580 --> 00:31:06,530 So just like what you did with Africa, only now it's working 570 00:31:06,530 --> 00:31:08,790 with its own visual memory, using visual programs. 571 00:31:12,026 --> 00:31:12,507 SYSTEM: OK. 572 00:31:12,507 --> 00:31:17,330 I will imagine that a man gives a ball to a man. 573 00:31:17,330 --> 00:31:21,050 PATRICK WINSTON: I know this looks like slugs, but they're 574 00:31:21,050 --> 00:31:22,300 actually distinguished professors. 575 00:31:26,500 --> 00:31:27,750 It always does the best it can. 576 00:31:30,465 --> 00:31:31,451 SYSTEM: OK. 577 00:31:31,451 --> 00:31:33,600 I will imagine that a man flies. 578 00:31:39,560 --> 00:31:41,480 PATRICK WINSTON: It's the best that it can do. 579 00:31:47,870 --> 00:31:51,920 So that concludes our discussion of the history. 580 00:31:51,920 --> 00:31:54,240 And I've provided you with a little bit of a glimpse of 581 00:31:54,240 --> 00:31:57,880 what we're going to look at as the semester unfolds. 582 00:31:57,880 --> 00:32:00,410 Yes, Chris? 583 00:32:00,410 --> 00:32:03,880 CHRIS: Is it actually a demonstration of something? 584 00:32:03,880 --> 00:32:06,200 Does it have a large database of videos? 585 00:32:06,200 --> 00:32:08,676 PATRICK WINSTON: No, it has a small database videos. 586 00:32:08,676 --> 00:32:13,480 CHRIS: But it's intelligently picking among them based on-- 587 00:32:13,480 --> 00:32:16,140 PATRICK WINSTON: Based on their content. 588 00:32:16,140 --> 00:32:20,570 So if you say imagine that a student gave a ball to another 589 00:32:20,570 --> 00:32:23,170 student, it imagines that. 590 00:32:23,170 --> 00:32:26,430 You say, now does the other student have the ball? 591 00:32:26,430 --> 00:32:27,590 Does the other student take the ball? 592 00:32:27,590 --> 00:32:30,030 It can answer those questions because it can review the same 593 00:32:30,030 --> 00:32:33,250 video and see the take as well as the give in the same video. 594 00:32:35,760 --> 00:32:42,840 So now we have to think about why we ought to be optimistic 595 00:32:42,840 --> 00:32:44,460 about the future. 596 00:32:44,460 --> 00:32:47,210 Because we've had a long history here, and we haven't 597 00:32:47,210 --> 00:32:48,830 solved the problem. 598 00:32:48,830 --> 00:32:51,350 But one reason why we can feel optimistic about future is 599 00:32:51,350 --> 00:32:54,580 because all of our friends have been on the march. 600 00:32:54,580 --> 00:32:57,950 And our friends include the cognitive psychologists, the 601 00:32:57,950 --> 00:32:58,930 [? developmental ?] 602 00:32:58,930 --> 00:33:02,200 psychologists, the linguists, sometimes the philosophers, 603 00:33:02,200 --> 00:33:03,450 and especially the paleoanthropologists. 604 00:33:06,040 --> 00:33:09,430 Because it is becoming increasingly clear why we're 605 00:33:09,430 --> 00:33:12,780 actually different from the chimpanzees, and how we got to 606 00:33:12,780 --> 00:33:14,690 be that way. 607 00:33:14,690 --> 00:33:19,910 The high school idea is that we evolved through slow, 608 00:33:19,910 --> 00:33:23,540 gradual, and continuous improvement. 609 00:33:23,540 --> 00:33:25,730 But that doesn't seem to be the way it happened. 610 00:33:25,730 --> 00:33:30,210 There are some characteristics of our species that are 611 00:33:30,210 --> 00:33:34,010 informative when it comes to guiding the activities of 612 00:33:34,010 --> 00:33:36,120 people like me. 613 00:33:36,120 --> 00:33:37,950 And here's what the story seems to be 614 00:33:37,950 --> 00:33:40,600 from the fossil record. 615 00:33:40,600 --> 00:33:45,330 First of all, we humans have been around for maybe 200,000 616 00:33:45,330 --> 00:33:49,210 years in our present anatomical form. 617 00:33:49,210 --> 00:33:53,140 If someone walked through the door right now from 200,000 618 00:33:53,140 --> 00:34:00,990 years ago, I imagine they would be dirty, 619 00:34:00,990 --> 00:34:03,750 but other than that-- 620 00:34:03,750 --> 00:34:05,100 probably naked, too-- 621 00:34:05,100 --> 00:34:08,070 other than that, you wouldn't be able to tell the 622 00:34:08,070 --> 00:34:09,320 difference, especially at MIT. 623 00:34:15,840 --> 00:34:21,900 And so the ensuing 150,000 years was a period in which we 624 00:34:21,900 --> 00:34:25,159 humans didn't actually amount to much. 625 00:34:25,159 --> 00:34:30,070 But somehow, shortly before 50,000 years ago, some small 626 00:34:30,070 --> 00:34:34,500 group of us developed a capability that separated us 627 00:34:34,500 --> 00:34:37,520 from all other species. 628 00:34:37,520 --> 00:34:40,409 It was an accident of evolution. 629 00:34:40,409 --> 00:34:43,310 And these accidents may or may not happen, but it happened to 630 00:34:43,310 --> 00:34:45,199 produce us. 631 00:34:45,199 --> 00:34:47,920 It's also the case that we probably necked down as a 632 00:34:47,920 --> 00:34:50,600 species to a few thousand, or maybe even a few hundred 633 00:34:50,600 --> 00:34:55,219 individuals, something which made these accidental changes, 634 00:34:55,219 --> 00:34:57,210 accidental evolutionary products, 635 00:34:57,210 --> 00:35:00,500 more capable of sticking. 636 00:35:00,500 --> 00:35:02,820 This leads us to speculate on what it was that happened 637 00:35:02,820 --> 00:35:04,750 50,000 years ago. 638 00:35:04,750 --> 00:35:10,520 And paleoanthropologists, Noam Chomsky, a lot of people 639 00:35:10,520 --> 00:35:12,580 reached similar conclusions. 640 00:35:12,580 --> 00:35:15,050 And that conclusion is-- 641 00:35:15,050 --> 00:35:16,700 I'll quote Chomsky. 642 00:35:16,700 --> 00:35:18,990 He's the voice of authority. 643 00:35:18,990 --> 00:35:22,020 "It seems that shortly before 50,000 years ago, some small 644 00:35:22,020 --> 00:35:26,040 group of us acquired the ability to take two concepts, 645 00:35:26,040 --> 00:35:29,020 and combine them to make a third concept, without 646 00:35:29,020 --> 00:35:33,540 disturbing the original two concepts, without limit." And 647 00:35:33,540 --> 00:35:37,110 from a perspective of an AI person like me, what Chomsky 648 00:35:37,110 --> 00:35:39,490 seems to be saying is, we learned how to begin to 649 00:35:39,490 --> 00:35:42,270 describe things, in a way that was intimately 650 00:35:42,270 --> 00:35:43,790 connected with language. 651 00:35:43,790 --> 00:35:46,360 And that, in the end, is what separates us from the 652 00:35:46,360 --> 00:35:49,150 chimpanzees. 653 00:35:49,150 --> 00:35:50,790 So you might say, well let's just study language. 654 00:35:50,790 --> 00:35:53,580 No, you can't do that, because we think with our eyes. 655 00:35:53,580 --> 00:35:55,510 So language does two things. 656 00:35:55,510 --> 00:35:57,790 Number one, it enables us to make descriptions. 657 00:35:57,790 --> 00:36:00,040 Descriptions enable us to tell stories. 658 00:36:00,040 --> 00:36:02,400 And storytelling and story understanding is what all of 659 00:36:02,400 --> 00:36:03,730 education is about. 660 00:36:03,730 --> 00:36:04,810 That's going up. 661 00:36:04,810 --> 00:36:09,370 And going down enables us to marshal the resources of our 662 00:36:09,370 --> 00:36:12,950 perceptual systems, and even command our perceptual systems 663 00:36:12,950 --> 00:36:17,450 to imagine things we've never seen. 664 00:36:17,450 --> 00:36:20,120 So here's an example. 665 00:36:20,120 --> 00:36:22,420 Imagine running down the street with a 666 00:36:22,420 --> 00:36:24,640 full bucket of water. 667 00:36:24,640 --> 00:36:25,890 What happens? 668 00:36:27,810 --> 00:36:29,100 Your leg gets wet. 669 00:36:29,100 --> 00:36:30,780 The water sloshes out. 670 00:36:30,780 --> 00:36:34,600 You'll never find that fact anywhere on the web. 671 00:36:34,600 --> 00:36:37,470 You've probably never been told that that's what happens 672 00:36:37,470 --> 00:36:39,910 when you run down the street with a full bucket of water. 673 00:36:39,910 --> 00:36:43,400 But you easily imagine this scenario, and you know what's 674 00:36:43,400 --> 00:36:43,990 going to happen. 675 00:36:43,990 --> 00:36:47,150 There was internal imagination simulation. 676 00:36:47,150 --> 00:36:49,030 We're never going to understand human intelligence 677 00:36:49,030 --> 00:36:51,670 until we can understand that. 678 00:36:51,670 --> 00:36:52,540 Here's another example. 679 00:36:52,540 --> 00:36:53,670 Imagine running down the street with a 680 00:36:53,670 --> 00:36:55,620 full bucket of nickels? 681 00:36:55,620 --> 00:36:56,870 What happens? 682 00:36:59,460 --> 00:37:00,415 Nickels weigh a lot. 683 00:37:00,415 --> 00:37:01,395 You're going to be bent over. 684 00:37:01,395 --> 00:37:03,850 You're going to stagger. 685 00:37:03,850 --> 00:37:05,690 But nobody ever told you that. 686 00:37:05,690 --> 00:37:08,080 You won't find it anywhere on the web. 687 00:37:08,080 --> 00:37:10,720 So language is at the center of things because it enables 688 00:37:10,720 --> 00:37:14,380 storytelling going up, and marshalling the resources of 689 00:37:14,380 --> 00:37:18,240 the perceptual apparatus, going down. 690 00:37:18,240 --> 00:37:20,950 And that's where we're going to finish the subject the 691 00:37:20,950 --> 00:37:23,320 semester, by trying to understand more about that 692 00:37:23,320 --> 00:37:25,950 phenomenon. 693 00:37:25,950 --> 00:37:28,150 So that concludes everything I wanted to say about the 694 00:37:28,150 --> 00:37:30,520 material and the subject. 695 00:37:30,520 --> 00:37:32,290 Now I want to turn my attention a little bit to how 696 00:37:32,290 --> 00:37:34,090 we are going to operate the subject. 697 00:37:34,090 --> 00:37:36,915 Because there are many characteristics of the subject 698 00:37:36,915 --> 00:37:38,715 that are confusing. 699 00:37:43,610 --> 00:37:46,840 First of all, we have four kinds of 700 00:37:46,840 --> 00:37:48,180 activities in the course. 701 00:38:17,930 --> 00:38:22,870 And each of these has a different purpose. 702 00:38:39,580 --> 00:38:40,850 So I did the lectures. 703 00:38:40,850 --> 00:38:43,910 And the lectures are supposed to be an hour about 704 00:38:43,910 --> 00:38:46,740 introducing the material and the big picture. 705 00:38:46,740 --> 00:38:48,880 They're about powerful ideas. 706 00:38:52,138 --> 00:38:56,060 They're about the experience side of the course. 707 00:38:56,060 --> 00:38:59,130 Let me step aside and make a remark. 708 00:38:59,130 --> 00:39:00,390 MIT is about two things. 709 00:39:00,390 --> 00:39:06,200 It's about skill building, and it's about big ideas. 710 00:39:06,200 --> 00:39:10,120 So you can build a skill at home, or at Dartmouth, or at 711 00:39:10,120 --> 00:39:13,830 Harvard, or Princeton, or all those kinds of places. 712 00:39:13,830 --> 00:39:16,280 But the experience you can only get at MIT. 713 00:39:16,280 --> 00:39:18,000 I know everybody there is to know in artificial 714 00:39:18,000 --> 00:39:18,940 intelligence. 715 00:39:18,940 --> 00:39:20,190 I can tell you about how they think. 716 00:39:20,190 --> 00:39:21,250 I can tell you about how I think. 717 00:39:21,250 --> 00:39:22,620 And that's something you're not going to 718 00:39:22,620 --> 00:39:25,120 get any other place. 719 00:39:25,120 --> 00:39:29,640 So that's my role, as I see it, in giving these lectures. 720 00:39:29,640 --> 00:39:34,160 Recitations are four buttressing and expanding on 721 00:39:34,160 --> 00:39:37,430 the material, and providing a venue that's small enough for 722 00:39:37,430 --> 00:39:38,680 discussion. 723 00:39:40,540 --> 00:39:42,200 Mega recitations are [? a usual ?] 724 00:39:42,200 --> 00:39:43,380 components of the course. 725 00:39:43,380 --> 00:39:46,230 They're taught at the same hour on Fridays. 726 00:39:46,230 --> 00:39:48,100 Mark Seifter, my graduate student, 727 00:39:48,100 --> 00:39:49,510 will be teaching those. 728 00:39:49,510 --> 00:39:53,140 And those are wrapped around past quiz problems. 729 00:39:53,140 --> 00:39:55,700 And Mark will show you how to work them. 730 00:39:55,700 --> 00:40:00,050 It's very important component to the subject. 731 00:40:00,050 --> 00:40:02,400 And finally the tutorials are about helping 732 00:40:02,400 --> 00:40:04,890 you with the homework. 733 00:40:04,890 --> 00:40:09,160 So you might say to me, well, do I really 734 00:40:09,160 --> 00:40:11,590 need to go to class? 735 00:40:11,590 --> 00:40:15,170 I like to say that the answer is, only if you like to pass 736 00:40:15,170 --> 00:40:17,870 the subject. 737 00:40:17,870 --> 00:40:19,920 But you are MIT students. 738 00:40:19,920 --> 00:40:22,190 And MIT people always like to look at the data. 739 00:40:26,150 --> 00:40:29,640 So this is a scattergram we made after the subject was 740 00:40:29,640 --> 00:40:33,010 taught last fall, which shows the relationship between 741 00:40:33,010 --> 00:40:38,680 attendance at lectures and the grades awarded in the course. 742 00:40:38,680 --> 00:40:40,820 And if you're not sure what that all means, here's the 743 00:40:40,820 --> 00:40:42,070 regression line. 744 00:40:46,160 --> 00:40:50,400 So that information is a little suspect for two 745 00:40:50,400 --> 00:40:55,520 reasons, one of which is we asked people to self report on 746 00:40:55,520 --> 00:40:58,330 how many lectures they thought they attended. 747 00:40:58,330 --> 00:41:02,340 And our mechanism for assigning these numerical 748 00:41:02,340 --> 00:41:03,590 grades is a little weird. 749 00:41:06,560 --> 00:41:08,470 And there's a third thing, too, and that is, one must 750 00:41:08,470 --> 00:41:12,860 never confuse correlation with cause. 751 00:41:12,860 --> 00:41:15,150 You can think of other explanations for why that 752 00:41:15,150 --> 00:41:21,230 trend line goes up, different from whether it has something 753 00:41:21,230 --> 00:41:25,840 to do with lectures producing good grades. 754 00:41:25,840 --> 00:41:28,070 You might ask how I feel about the people up there on the 755 00:41:28,070 --> 00:41:30,580 other upper left hand corner. 756 00:41:30,580 --> 00:41:32,970 There are one or two people who were near the top of the 757 00:41:32,970 --> 00:41:37,190 subject who didn't go to class at all. 758 00:41:37,190 --> 00:41:40,890 And I have mixed feelings about that. 759 00:41:40,890 --> 00:41:41,550 You're adults. 760 00:41:41,550 --> 00:41:43,460 It's your call. 761 00:41:43,460 --> 00:41:46,490 On the other hand, I wish that if that's what you do 762 00:41:46,490 --> 00:41:49,780 habitually in all the subjects you take at MIT, that you 763 00:41:49,780 --> 00:41:52,450 would resign and go somewhere else, and let somebody else 764 00:41:52,450 --> 00:41:53,780 take their slot. 765 00:41:53,780 --> 00:41:57,050 Because you're not benefiting from the powerful ideas, and 766 00:41:57,050 --> 00:41:58,930 the other kinds of things that involve 767 00:41:58,930 --> 00:42:01,150 interaction with faculty. 768 00:42:01,150 --> 00:42:02,350 So it can be done. 769 00:42:02,350 --> 00:42:03,310 But I don't recommend it. 770 00:42:03,310 --> 00:42:06,970 By the way, all of the four activities that we have here 771 00:42:06,970 --> 00:42:08,440 show similar regression lines. 772 00:42:11,280 --> 00:42:12,930 But what about that five point scale? 773 00:42:12,930 --> 00:42:14,940 Let me explain how that works to you. 774 00:42:14,940 --> 00:42:18,060 We love to have people ask us what the class 775 00:42:18,060 --> 00:42:20,430 average is on a quiz. 776 00:42:20,430 --> 00:42:23,480 Because that's when we get to use our blank stare. 777 00:42:23,480 --> 00:42:26,180 Because we have no idea what the class average 778 00:42:26,180 --> 00:42:28,740 ever is on any quiz. 779 00:42:28,740 --> 00:42:29,990 Here's what we do. 780 00:42:46,110 --> 00:42:48,660 Like everybody else, we start off with a score 781 00:42:48,660 --> 00:42:50,680 from zero to 100. 782 00:42:50,680 --> 00:42:54,750 But then we say to ourselves, what score would you get if 783 00:42:54,750 --> 00:42:57,260 you had a thorough understanding of the material? 784 00:42:57,260 --> 00:43:00,400 And we say, well, for this particular exam, it's this 785 00:43:00,400 --> 00:43:02,480 number right here. 786 00:43:02,480 --> 00:43:04,630 And what score would you get if you had a good 787 00:43:04,630 --> 00:43:05,500 understanding of the material? 788 00:43:05,500 --> 00:43:07,340 That's that score. 789 00:43:07,340 --> 00:43:10,040 And what happens if you're down here is that you're 790 00:43:10,040 --> 00:43:12,670 following off the edge of the range in which we think you 791 00:43:12,670 --> 00:43:14,940 need to do more work. 792 00:43:14,940 --> 00:43:18,830 So what we do is, we say that if you're in this range here-- 793 00:43:18,830 --> 00:43:21,520 following MIT convention with GPAs and stuff, 794 00:43:21,520 --> 00:43:23,440 that gets you a five. 795 00:43:23,440 --> 00:43:25,300 If you're in this range down here, there's a 796 00:43:25,300 --> 00:43:28,170 sharp drop off to four. 797 00:43:28,170 --> 00:43:29,710 If you're in this range down here, there's a 798 00:43:29,710 --> 00:43:33,070 sharp fall off to three. 799 00:43:33,070 --> 00:43:34,950 So that means if you're in the middle of one of those 800 00:43:34,950 --> 00:43:37,790 plateaus there's no point in arguing with this. 801 00:43:37,790 --> 00:43:40,550 Because it's not going to do you any good. 802 00:43:40,550 --> 00:43:43,020 We have these boundaries where we think performance break 803 00:43:43,020 --> 00:43:45,430 points are. 804 00:43:45,430 --> 00:43:47,870 So you say, well that seems a little harsh. 805 00:43:47,870 --> 00:43:49,850 Blah, blah, blah, blah, blah, and start arguing. 806 00:43:49,850 --> 00:43:53,560 But then we will come back with a second major innovation 807 00:43:53,560 --> 00:43:54,860 we have in the course. 808 00:43:54,860 --> 00:44:00,300 That is that your grade is calculated in several parts. 809 00:44:00,300 --> 00:44:06,970 Part one is the max of your grade on Q1, and 810 00:44:06,970 --> 00:44:09,830 part one of the final. 811 00:44:09,830 --> 00:44:12,780 So in other words, you get two shots at everything. 812 00:44:12,780 --> 00:44:18,080 So if you have complete glorious undeniable horrible F 813 00:44:18,080 --> 00:44:21,540 on the first quiz, it gets erased on the final if you do 814 00:44:21,540 --> 00:44:23,910 well on that part of the final. 815 00:44:23,910 --> 00:44:25,590 So each quiz has a corresponding 816 00:44:25,590 --> 00:44:27,080 mirror on the final. 817 00:44:27,080 --> 00:44:32,680 You get the max of the score you got on those two pieces. 818 00:44:32,680 --> 00:44:36,540 And now you say to me, I'm an MIT student. 819 00:44:36,540 --> 00:44:38,066 I have a lot of guts. 820 00:44:38,066 --> 00:44:39,740 I'm only going to take the final. 821 00:44:44,020 --> 00:44:46,060 It has been done. 822 00:44:46,060 --> 00:44:49,400 We don't recommend it. 823 00:44:49,400 --> 00:44:51,970 And the reason we don't recommend it is that we don't 824 00:44:51,970 --> 00:44:53,730 expect everybody to do all of the final. 825 00:44:53,730 --> 00:44:55,740 So there would be a lot of time pressure if you had to do 826 00:44:55,740 --> 00:45:00,200 all of the final, all five parts of the final. 827 00:45:00,200 --> 00:45:02,890 So we have four quizzes. 828 00:45:02,890 --> 00:45:05,590 And the final has a fifth part because there's some material 829 00:45:05,590 --> 00:45:08,710 that we teach you after the last date on which we can give 830 00:45:08,710 --> 00:45:11,010 you a final by institute rules. 831 00:45:11,010 --> 00:45:12,050 But that's roughly how it works. 832 00:45:12,050 --> 00:45:16,690 And you can read about more of the details in the FAQ on the 833 00:45:16,690 --> 00:45:17,940 subject homepage. 834 00:45:19,910 --> 00:45:21,050 So now we're almost done. 835 00:45:21,050 --> 00:45:24,190 I just want to talk a little bit about how we're going to 836 00:45:24,190 --> 00:45:27,840 communicate with you in the next few days, while we're 837 00:45:27,840 --> 00:45:29,710 getting ourselves organized. 838 00:45:29,710 --> 00:45:32,560 So, number one-- if I could ask the TAs to help me pass 839 00:45:32,560 --> 00:45:34,570 these out-- 840 00:45:34,570 --> 00:45:37,890 we need to schedule you into tutorials. 841 00:45:37,890 --> 00:45:39,970 So we're going to ask you to fill out this form, and give 842 00:45:39,970 --> 00:45:41,220 it to us before you leave. 843 00:45:44,250 --> 00:45:48,910 So you'll be hearing from us once we do the sort. 844 00:45:48,910 --> 00:45:58,100 There's the issue of whether we're going to have ordinary 845 00:45:58,100 --> 00:46:00,500 recitation and a mega recitation this week. 846 00:46:00,500 --> 00:46:01,460 So pay attention. 847 00:46:01,460 --> 00:46:03,630 Otherwise, you're going to be stranded in a classroom with 848 00:46:03,630 --> 00:46:04,990 nothing to do. 849 00:46:04,990 --> 00:46:07,450 We're not going to have any regular recitations this week. 850 00:46:10,010 --> 00:46:11,410 Are we having regular recitation this week, 851 00:46:11,410 --> 00:46:11,720 [INAUDIBLE]? 852 00:46:11,720 --> 00:46:14,270 No. 853 00:46:14,270 --> 00:46:19,890 We may, and probably will, have a mega recitation this 854 00:46:19,890 --> 00:46:24,480 week that's devoted to a Python review. 855 00:46:24,480 --> 00:46:27,810 Now we know that there are many of you who are 856 00:46:27,810 --> 00:46:31,250 celebrating a religious holiday on Friday, and so we 857 00:46:31,250 --> 00:46:34,680 will be putting a lot of resources online so you can 858 00:46:34,680 --> 00:46:38,060 get that review in another way. 859 00:46:38,060 --> 00:46:42,740 We probably will have a Python review on Friday. 860 00:46:42,740 --> 00:46:45,640 And we ask that you look at our home page for further 861 00:46:45,640 --> 00:46:49,520 information about that as the week progresses. 862 00:46:49,520 --> 00:46:50,640 So that's all folks. 863 00:46:50,640 --> 00:46:52,470 That concludes what we're going to do today. 864 00:46:52,470 --> 00:46:56,750 And as soon as you give us your form, we're through.