1 00:00:09,660 --> 00:00:12,360 PROFESSOR: [INAUDIBLE]. 2 00:00:12,360 --> 00:00:15,310 "Thus Spake Zarathustra" was made famous 3 00:00:15,310 --> 00:00:18,300 and popular by 2001. 4 00:00:18,300 --> 00:00:21,840 And that is music played at this magic moment when some 5 00:00:21,840 --> 00:00:24,220 primate suddenly gets an idea, 6 00:00:24,220 --> 00:00:25,690 presumably one of our ancestors. 7 00:00:28,780 --> 00:00:32,210 So how do we explain all that? 8 00:00:32,210 --> 00:00:34,850 We've got all of the ingredients on the table. 9 00:00:34,850 --> 00:00:37,380 And today I want to talk about various ways of putting those 10 00:00:37,380 --> 00:00:39,200 ingredients together. 11 00:00:39,200 --> 00:00:41,540 So we talked about representations, we've talked 12 00:00:41,540 --> 00:00:43,860 about methods, and today we're going to talk about 13 00:00:43,860 --> 00:00:45,340 architectures. 14 00:00:45,340 --> 00:00:50,410 And by the end of the class you'll know how to put one of 15 00:00:50,410 --> 00:00:51,270 those things together. 16 00:00:51,270 --> 00:00:53,100 Actually, no one knows how to put one of 17 00:00:53,100 --> 00:00:53,900 those things together. 18 00:00:53,900 --> 00:00:56,750 But what you will know is about some alternatives for 19 00:00:56,750 --> 00:01:01,992 putting those things together so as to make something that 20 00:01:01,992 --> 00:01:08,920 is arguably intelligent in the same way we are. 21 00:01:08,920 --> 00:01:10,360 So that is our agenda for today. 22 00:01:14,240 --> 00:01:15,914 We'll also talk a little bit more about stories. 23 00:01:18,475 --> 00:01:26,270 I think it was in 2007 when the Estonians moved a war 24 00:01:26,270 --> 00:01:28,990 memorial from the center of Tallinn off to 25 00:01:28,990 --> 00:01:33,150 a Russian war cemetery. 26 00:01:33,150 --> 00:01:35,690 Prior to that time the Estonians had been building up 27 00:01:35,690 --> 00:01:37,870 their national computer networks because they thought 28 00:01:37,870 --> 00:01:40,346 that computation was the wave of the future-- networks and 29 00:01:40,346 --> 00:01:41,870 all of that. 30 00:01:41,870 --> 00:01:45,920 Shortly after the movement of that war memorial, someone 31 00:01:45,920 --> 00:01:49,604 brought the Estonian national network down-- 32 00:01:49,604 --> 00:01:51,700 a cyber attack. 33 00:01:51,700 --> 00:01:55,930 It was widely believed to be the Russians. 34 00:01:55,930 --> 00:01:58,330 There's a large Russian ethnic population in 35 00:01:58,330 --> 00:01:59,539 Estonia to start with. 36 00:01:59,539 --> 00:02:02,080 And the movement of that war memorial 37 00:02:02,080 --> 00:02:03,730 irritated the Russians. 38 00:02:03,730 --> 00:02:06,960 And so everybody everybody thinks that they did it. 39 00:02:06,960 --> 00:02:07,910 But you know what? 40 00:02:07,910 --> 00:02:12,110 No computer can understand the story I just told. 41 00:02:12,110 --> 00:02:15,610 They can revel through all of the worldwide web finding 42 00:02:15,610 --> 00:02:18,829 information that's relevant to that, but no computer can 43 00:02:18,829 --> 00:02:21,710 understand the story I just told, except one. 44 00:02:21,710 --> 00:02:25,300 You'll see a demonstration of that later on today. 45 00:02:25,300 --> 00:02:28,620 So by the way, if you're interested in understanding 46 00:02:28,620 --> 00:02:31,480 the nature of intelligence, this is, of course, the most 47 00:02:31,480 --> 00:02:35,020 important lecture of the semester. 48 00:02:35,020 --> 00:02:37,530 And I should tell you a little bit about what we're going to 49 00:02:37,530 --> 00:02:38,390 do on Wednesday. 50 00:02:38,390 --> 00:02:43,470 Because for some reason the day before Thanksgiving tends 51 00:02:43,470 --> 00:02:46,510 to be a lecture that's lightly populated, 52 00:02:46,510 --> 00:02:47,980 except in this class. 53 00:02:47,980 --> 00:02:51,820 Because I'm going to talk about the artificial 54 00:02:51,820 --> 00:02:54,340 intelligence business and what can be learned from it about 55 00:02:54,340 --> 00:02:57,579 how to avoid going broke when you start your company. 56 00:02:57,579 --> 00:03:02,760 So for many of you that will be the most important lecture 57 00:03:02,760 --> 00:03:05,680 of the semester. 58 00:03:05,680 --> 00:03:08,070 It all started back in the dawn age of artificial 59 00:03:08,070 --> 00:03:09,810 intelligence. 60 00:03:09,810 --> 00:03:14,020 And really, it all started at Carnegie Mellon, sad to say. 61 00:03:14,020 --> 00:03:16,950 Because the people at Carnegie Mellon, notably Newell and 62 00:03:16,950 --> 00:03:20,790 Simon, were the first to think about sort of a general 63 00:03:20,790 --> 00:03:25,970 purpose way of putting things together so as to build a 64 00:03:25,970 --> 00:03:29,530 structure or architecture in which particular intelligent 65 00:03:29,530 --> 00:03:33,400 systems could be built. 66 00:03:33,400 --> 00:03:36,260 So their idea was called the general problem solver. 67 00:03:45,210 --> 00:03:46,680 A long name for a simple idea. 68 00:03:49,630 --> 00:03:52,890 And the simple idea is that you start your life out in a 69 00:03:52,890 --> 00:03:57,640 current state, call it C. And you want to get 70 00:03:57,640 --> 00:03:59,710 to some goal state. 71 00:03:59,710 --> 00:04:03,940 Call it S. And a way you do that is you measure somehow 72 00:04:03,940 --> 00:04:06,810 the symbolic difference between where you are and 73 00:04:06,810 --> 00:04:08,610 where you want to be. 74 00:04:08,610 --> 00:04:10,085 So that's the difference. 75 00:04:10,085 --> 00:04:14,370 We'll call that difference D. And when you observe that 76 00:04:14,370 --> 00:04:19,200 difference that's enough, they say, in this general approach 77 00:04:19,200 --> 00:04:20,339 to problem solving. 78 00:04:20,339 --> 00:04:23,970 For you to select some operation that will move you 79 00:04:23,970 --> 00:04:26,880 from your current state to some new state, an 80 00:04:26,880 --> 00:04:27,880 intermediate state. 81 00:04:27,880 --> 00:04:35,500 Call it I. So I, or that operator, O, is determined by 82 00:04:35,500 --> 00:04:37,659 the difference, D. 83 00:04:37,659 --> 00:04:40,520 And then, of course, the next thing to do is to measure the 84 00:04:40,520 --> 00:04:43,350 difference between that intermediate state and the 85 00:04:43,350 --> 00:04:47,300 state you want to be in, and choose some operator that's 86 00:04:47,300 --> 00:04:49,440 relevant to reducing that state. 87 00:04:49,440 --> 00:04:53,490 So we'll call that D2, and we'll call this O2. 88 00:04:53,490 --> 00:04:58,950 And D2 is what leads you to 02, and so it goes. 89 00:04:58,950 --> 00:05:00,550 So that's the idea. 90 00:05:00,550 --> 00:05:03,470 And that's often called means-ends analysis. 91 00:05:03,470 --> 00:05:03,620 Why? 92 00:05:03,620 --> 00:05:07,620 Because the end that you want to achieve is being in that 93 00:05:07,620 --> 00:05:13,430 final state, S. And the means is that operator, O. So you 94 00:05:13,430 --> 00:05:15,450 have some notion of where you want to be and the difference 95 00:05:15,450 --> 00:05:17,060 of where you are and where you want to be. 96 00:05:17,060 --> 00:05:20,570 And you pick an operator so as to reduce that difference. 97 00:05:20,570 --> 00:05:22,700 So this is all very abstract. 98 00:05:22,700 --> 00:05:27,020 Let's exercise it in solving a problem that you will all be 99 00:05:27,020 --> 00:05:28,930 faced with here in a day or two. 100 00:05:28,930 --> 00:05:30,910 That is, for many of you-- 101 00:05:30,910 --> 00:05:34,210 most Of you, I hope-- the problem of going home. 102 00:05:34,210 --> 00:05:36,050 So here you are. 103 00:05:36,050 --> 00:05:38,450 You're at MIT. 104 00:05:38,450 --> 00:05:42,205 And where you want to be is over here, at home. 105 00:05:44,860 --> 00:05:49,180 So you measure the difference between MIT and home. 106 00:05:49,180 --> 00:05:51,990 And for many of you it's further than you can go by car 107 00:05:51,990 --> 00:05:54,070 and not so far that you can't go at all. 108 00:05:54,070 --> 00:05:57,180 So what you do is, you say, well, the right operator is 109 00:05:57,180 --> 00:05:59,620 taking an airplane. 110 00:05:59,620 --> 00:06:04,125 So there is the operator, take an airplane. 111 00:06:09,090 --> 00:06:13,600 And this is the difference, D. And the difference, D, being 112 00:06:13,600 --> 00:06:17,630 sufficiently large, you take the plane. 113 00:06:17,630 --> 00:06:20,260 Trouble is, if you happen to be sitting here in 114 00:06:20,260 --> 00:06:21,420 [? 10-250 ?] 115 00:06:21,420 --> 00:06:23,490 there's no way you can take an airplane, because they don't 116 00:06:23,490 --> 00:06:25,460 fit in here. 117 00:06:25,460 --> 00:06:27,040 So you've got another problem, and that is 118 00:06:27,040 --> 00:06:29,330 to get to the airplane. 119 00:06:29,330 --> 00:06:32,560 So the distance between here and Logan is such that the 120 00:06:32,560 --> 00:06:38,045 right way to do that is to take the MBTA. 121 00:06:40,980 --> 00:06:43,310 And that's determined because you're working on this 122 00:06:43,310 --> 00:06:45,840 difference reduction right here, the difference from 123 00:06:45,840 --> 00:06:48,070 being at MIT and being at the airport. 124 00:06:48,070 --> 00:06:50,610 So that difference dictates that you take the MBTA. 125 00:06:50,610 --> 00:06:53,909 So you see, you're re-cursing. 126 00:06:53,909 --> 00:06:56,860 But you know there are no MBTA cars in here either. 127 00:06:56,860 --> 00:07:00,210 So there's still a difference like so. 128 00:07:00,210 --> 00:07:05,310 And that difference dictates that you walk. 129 00:07:05,310 --> 00:07:10,670 So you've got D1, D2, and D3. 130 00:07:10,670 --> 00:07:14,140 And by the time you've excised the operators relevant to 131 00:07:14,140 --> 00:07:16,690 those three differences, you're at Logan. 132 00:07:16,690 --> 00:07:19,705 Then you take the airplane, you get over to your hometown, 133 00:07:19,705 --> 00:07:22,960 and your faced with the smaller difference of getting 134 00:07:22,960 --> 00:07:25,660 from that airport to where you actually want to go. 135 00:07:25,660 --> 00:07:29,150 So that's the general problem- solver idea. 136 00:07:29,150 --> 00:07:34,440 It was such an exciting idea at the time. 137 00:07:34,440 --> 00:07:36,850 It was such an exciting idea at the time because people 138 00:07:36,850 --> 00:07:39,380 would say to themselves, ah! 139 00:07:39,380 --> 00:07:42,570 This is a general purpose problem solver, so we can set 140 00:07:42,570 --> 00:07:46,740 it onto the problem of making itself smarter. 141 00:07:46,740 --> 00:07:49,460 And so there was a kind of imagined chain reaction that 142 00:07:49,460 --> 00:07:51,080 would take place. 143 00:07:51,080 --> 00:07:55,659 And the developers of this architecture warned the public 144 00:07:55,659 --> 00:07:59,190 that within 10 years-- that is to say, by about 1970-- 145 00:07:59,190 --> 00:08:03,760 computers would be generally as smart as people. 146 00:08:03,760 --> 00:08:05,870 And a lot of people made fun of them for that prediction. 147 00:08:05,870 --> 00:08:09,300 But it was actually scientists attempting to be responsible. 148 00:08:09,300 --> 00:08:12,750 Because they thought something, a quite serious 149 00:08:12,750 --> 00:08:15,220 dislocation was coming along, and that people should know 150 00:08:15,220 --> 00:08:16,400 that it was coming. 151 00:08:16,400 --> 00:08:19,030 And so they felt it was their responsibility in that age of 152 00:08:19,030 --> 00:08:22,810 scientific responsibility to warn the public. 153 00:08:22,810 --> 00:08:28,770 It didn't turn out that way, because the problem of 154 00:08:28,770 --> 00:08:32,049 collecting the differences and finding the operators, that's 155 00:08:32,049 --> 00:08:34,340 outside the scope of the architecture. 156 00:08:34,340 --> 00:08:38,500 So this is the problem that has to be solved by a human 157 00:08:38,500 --> 00:08:40,570 before this architecture can be used. 158 00:08:40,570 --> 00:08:44,370 You have to have identified the differences that you might 159 00:08:44,370 --> 00:08:51,490 encounter and the operators that you might use, and build 160 00:08:51,490 --> 00:08:54,740 this table which relates the two together. 161 00:08:54,740 --> 00:08:57,150 So maybe that one, that one, some off-diagonal 162 00:08:57,150 --> 00:08:58,490 elements, and so on. 163 00:08:58,490 --> 00:09:02,600 But building that table turned out to be a hard job. 164 00:09:02,600 --> 00:09:05,210 So not surprisingly, the idea evolved. 165 00:09:05,210 --> 00:09:08,940 And eventually the folks at Carnegie who developed the 166 00:09:08,940 --> 00:09:10,790 general problem solver-- 167 00:09:10,790 --> 00:09:14,170 most notably Newell and his students-- 168 00:09:14,170 --> 00:09:17,800 developed a newer, fresher, more elaborate architecture 169 00:09:17,800 --> 00:09:19,050 called SOAR. 170 00:09:23,610 --> 00:09:24,760 And here's how SOAR works. 171 00:09:24,760 --> 00:09:26,050 First of all, what does SOAR mean? 172 00:09:31,860 --> 00:09:33,090 It doesn't mean anything. 173 00:09:33,090 --> 00:09:45,750 It used to mean State Operator And Result. 174 00:09:49,660 --> 00:09:52,360 But for some reason the proponents of the SOAR 175 00:09:52,360 --> 00:09:55,450 architecture decided they don't like that acronym, and 176 00:09:55,450 --> 00:09:59,540 have asserted that SOAR is merely a label that shouldn't 177 00:09:59,540 --> 00:10:02,010 be thought of as an acronym. 178 00:10:02,010 --> 00:10:05,940 In any event, SOAR consists of various parts. 179 00:10:05,940 --> 00:10:10,750 It has a long-term memory. 180 00:10:10,750 --> 00:10:12,790 It has a short-term memory. 181 00:10:16,050 --> 00:10:21,030 And it has connections to the outside world, maybe a vision 182 00:10:21,030 --> 00:10:24,950 system and an action system. 183 00:10:24,950 --> 00:10:29,380 But most of the activity of the SOAR problem-solving 184 00:10:29,380 --> 00:10:33,750 architecture takes place in a short-term memory. 185 00:10:33,750 --> 00:10:36,300 So you can view the contents of the long-term memory as 186 00:10:36,300 --> 00:10:38,000 shuttling in and out of short-term memory. 187 00:10:38,000 --> 00:10:40,950 So you can see right away that this mechanism, this 188 00:10:40,950 --> 00:10:45,800 architecture, is heavily influenced by certain 189 00:10:45,800 --> 00:10:49,060 cognitive psychology experiments having to do with 190 00:10:49,060 --> 00:10:51,400 how much you can hold in your short-term memory-- 191 00:10:51,400 --> 00:10:53,530 nonsense syllables and all that sort of thing that was 192 00:10:53,530 --> 00:10:56,570 popular back in those days. 193 00:10:56,570 --> 00:10:59,530 So this was an architecture devised primarily by 194 00:10:59,530 --> 00:11:00,380 psychologists. 195 00:11:00,380 --> 00:11:04,170 And it had amongst its features a short-term memory 196 00:11:04,170 --> 00:11:05,420 and a long-term memory. 197 00:11:08,330 --> 00:11:10,580 So that's part 1 of this architecture. 198 00:11:10,580 --> 00:11:12,770 So what's in the long-term memory? 199 00:11:12,770 --> 00:11:25,890 Well, assertions and rules, AKA productions. 200 00:11:30,830 --> 00:11:35,390 A production being the Carnegie vernacular for rule. 201 00:11:35,390 --> 00:11:38,060 It's just the rule-based stuff like you saw on almost the 202 00:11:38,060 --> 00:11:39,930 first day of class. 203 00:11:39,930 --> 00:11:43,650 So the whole thing is a gigantic rule-based system 204 00:11:43,650 --> 00:11:48,230 with assertions and rules the shuttle back and forth from 205 00:11:48,230 --> 00:11:50,160 long-term memory into short-term memory where 206 00:11:50,160 --> 00:11:52,740 processing takes place. 207 00:11:52,740 --> 00:11:55,350 The third thing that comes to mind when you think of SOAR 208 00:11:55,350 --> 00:11:57,710 architecture is they had an elaborate preference system. 209 00:12:04,530 --> 00:12:06,610 You recall that when we talked about rule-based systems 210 00:12:06,610 --> 00:12:08,560 there's always a question of what do you do when more than 211 00:12:08,560 --> 00:12:10,300 one rule would work? 212 00:12:10,300 --> 00:12:13,300 You have to have some way of breaking those ties. 213 00:12:13,300 --> 00:12:15,430 The SOAR architecture has an elaborate 214 00:12:15,430 --> 00:12:18,770 subsystem for doing that. 215 00:12:18,770 --> 00:12:20,580 But I said that these are the first three things you think, 216 00:12:20,580 --> 00:12:21,760 and maybe that's not right. 217 00:12:21,760 --> 00:12:25,050 Because the next thing you think about is perhaps a 218 00:12:25,050 --> 00:12:29,890 better thing to identify with the SOAR architecture. 219 00:12:29,890 --> 00:12:31,435 And that's the idea of problem spaces. 220 00:12:38,620 --> 00:12:40,160 And that's the idea that if you're going to solve a 221 00:12:40,160 --> 00:12:44,220 problem you have to develop a space and do a search through 222 00:12:44,220 --> 00:12:45,760 that space. 223 00:12:45,760 --> 00:12:48,280 Just like we did when we talked about how we can get 224 00:12:48,280 --> 00:12:50,580 from here to home. 225 00:12:50,580 --> 00:12:54,310 There's a space of places, that's our problem space. 226 00:12:54,310 --> 00:12:56,430 We can do a search through that space to find a way to 227 00:12:56,430 --> 00:12:58,400 get from one place to another. 228 00:12:58,400 --> 00:13:03,540 That's the sort of thing that SOAR is focused on. 229 00:13:03,540 --> 00:13:07,520 Finally, the fifth element that you tend to think about 230 00:13:07,520 --> 00:13:11,150 when you think about SOAR is the idea of universal 231 00:13:11,150 --> 00:13:12,400 subgoaling. 232 00:13:21,440 --> 00:13:23,540 And that's the idea that whenever you can't think of 233 00:13:23,540 --> 00:13:27,420 want to do next, that becomes your next problem that 234 00:13:27,420 --> 00:13:30,590 deserves it's own problem space and its own set of 235 00:13:30,590 --> 00:13:35,210 differences and operators, and rules and assertions. 236 00:13:35,210 --> 00:13:37,970 So you start off on a high level, then you have to solve 237 00:13:37,970 --> 00:13:40,560 problems at a lower level, just like you did up there 238 00:13:40,560 --> 00:13:43,430 with a general problem solver. 239 00:13:43,430 --> 00:13:46,090 So if you have these two architectures you can begin to 240 00:13:46,090 --> 00:13:49,280 say, well, what are they centered on? 241 00:13:49,280 --> 00:13:53,860 And this architecture, this general problem solver, is 242 00:13:53,860 --> 00:13:55,970 centered on the idea that everything is 243 00:13:55,970 --> 00:14:00,150 about problem solving. 244 00:14:00,150 --> 00:14:04,505 Because the problem solving hypothesis-- 245 00:14:09,630 --> 00:14:11,020 no one gave it that name. 246 00:14:11,020 --> 00:14:12,680 But that's what it was. 247 00:14:12,680 --> 00:14:14,770 And this architecture did get its name. 248 00:14:14,770 --> 00:14:20,860 And it was said always, by Newell, to be based on what he 249 00:14:20,860 --> 00:14:22,875 called the symbol system hypothesis. 250 00:14:33,660 --> 00:14:36,760 The hypothesis that what we are as humans is symbol 251 00:14:36,760 --> 00:14:38,100 manipulators. 252 00:14:38,100 --> 00:14:43,980 And we can uncover how that all works by giving people 253 00:14:43,980 --> 00:14:47,960 crypto-arithmetic problems and having them talk out loud, by 254 00:14:47,960 --> 00:14:49,850 thinking about what happens when you try to remember 255 00:14:49,850 --> 00:14:53,280 nonsense syllables, by all that sort of stuff that was en 256 00:14:53,280 --> 00:14:57,120 vogue in terms of psychology experiments in the day when 257 00:14:57,120 --> 00:15:00,110 this architecture was first articulated. 258 00:15:00,110 --> 00:15:01,700 But when you look at architectures you can sort of 259 00:15:01,700 --> 00:15:05,472 see where they come from and what their antecedents are. 260 00:15:05,472 --> 00:15:12,290 It has a short-term memory and a long-term memory, because 261 00:15:12,290 --> 00:15:15,730 Newell and his associates were cognitive scientists. 262 00:15:15,730 --> 00:15:21,140 It has assertions and rules and preferences, because 263 00:15:21,140 --> 00:15:25,170 Newell and his associates were also AI people. 264 00:15:25,170 --> 00:15:29,950 And it has problem spaces and universal subgoaling because 265 00:15:29,950 --> 00:15:32,520 those are ideas that had been work out in a more primitive 266 00:15:32,520 --> 00:15:37,370 form already in the general problem-solver architecture. 267 00:15:37,370 --> 00:15:39,740 So that's a glimpse of what SOAR looked like 268 00:15:39,740 --> 00:15:40,750 in its early days. 269 00:15:40,750 --> 00:15:42,350 It's been very highly developed by a 270 00:15:42,350 --> 00:15:44,280 lot of smart people. 271 00:15:44,280 --> 00:15:47,420 So although it's symbol centered, they've attached to 272 00:15:47,420 --> 00:15:50,710 it things having to do with emotion and perception, but 273 00:15:50,710 --> 00:15:53,290 generally with the view that the first thing to do when 274 00:15:53,290 --> 00:15:57,020 faced with this perception is to get it out of there and get 275 00:15:57,020 --> 00:15:58,830 it into a symbolic form. 276 00:15:58,830 --> 00:16:05,180 That's sort of the bias that the architecture comes with. 277 00:16:05,180 --> 00:16:09,770 So those are two architectures that are heavily biased toward 278 00:16:09,770 --> 00:16:11,570 thinking that the important part of what we 279 00:16:11,570 --> 00:16:14,600 do is problem solving. 280 00:16:14,600 --> 00:16:17,020 But the most important, perhaps-- 281 00:16:17,020 --> 00:16:18,820 at least from an MIT perspective-- 282 00:16:18,820 --> 00:16:21,770 of these problem-solving oriented ways of thinking 283 00:16:21,770 --> 00:16:26,150 about the world, is Marvin Minsky's architecture, which 284 00:16:26,150 --> 00:16:36,860 he articulates in his book "The Emotion Machine." 285 00:16:36,860 --> 00:16:39,820 And Marvin is not just concerned with problem 286 00:16:39,820 --> 00:16:42,080 solving, but also with how problem solving 287 00:16:42,080 --> 00:16:43,620 might come in layers. 288 00:16:43,620 --> 00:16:46,960 So let me show you an example of the sort of problem what 289 00:16:46,960 --> 00:16:49,570 motivates some of Marvin's thinking. 290 00:16:56,770 --> 00:17:08,599 So you can read that, it's a short little vignette. 291 00:17:14,387 --> 00:17:17,708 You have no trouble understanding it, right? 292 00:17:17,708 --> 00:17:19,098 No. 293 00:17:19,098 --> 00:17:20,960 It's not difficult for us humans. 294 00:17:20,960 --> 00:17:22,950 Awfully tough for a computer. 295 00:17:22,950 --> 00:17:27,760 In part, because the thinking you need, your ability to 296 00:17:27,760 --> 00:17:31,210 understand that story, requires you to think on many 297 00:17:31,210 --> 00:17:34,110 levels at the same time. 298 00:17:34,110 --> 00:17:37,030 First of all, there's a sort of, at the 299 00:17:37,030 --> 00:17:39,490 bottom, instinctive reaction. 300 00:17:46,950 --> 00:17:48,650 You see where there's instinctive reaction? 301 00:17:48,650 --> 00:17:53,000 That's the part where she hears a sound 302 00:17:53,000 --> 00:17:53,850 and terns her head. 303 00:17:53,850 --> 00:17:54,400 That's instinct, right? 304 00:17:54,400 --> 00:17:57,190 That's practically built in. 305 00:17:57,190 --> 00:17:59,030 But then what she sees is a car. 306 00:17:59,030 --> 00:18:02,310 So that's something that we don't have wired in. 307 00:18:02,310 --> 00:18:05,700 It would be unlikely that we've evolved in the last 100 308 00:18:05,700 --> 00:18:08,500 years to have an instinctive appreciation of cars 309 00:18:08,500 --> 00:18:09,980 barrelling down the road. 310 00:18:09,980 --> 00:18:11,120 So the next level in Marvin's 311 00:18:11,120 --> 00:18:13,365 architecture is learned reaction. 312 00:18:22,000 --> 00:18:25,530 So that's the part about thinking about the car. 313 00:18:25,530 --> 00:18:28,320 Now spread throughout there-- well, let's see where is a 314 00:18:28,320 --> 00:18:31,640 particularly good example. 315 00:18:31,640 --> 00:18:34,750 She decides to sprint across the road. 316 00:18:34,750 --> 00:18:37,380 So that's where she's solving a problem. 317 00:18:37,380 --> 00:18:39,610 So that's the deliberative thinking level. 318 00:18:47,770 --> 00:18:55,630 It doesn't stop there, because later on she reflects on her 319 00:18:55,630 --> 00:18:57,670 impulsive decision. 320 00:18:57,670 --> 00:18:59,950 So she thinks not only about stuff that's happening out 321 00:18:59,950 --> 00:19:03,020 there in the world, but she also thinks about stuff that's 322 00:19:03,020 --> 00:19:04,970 going on in here. 323 00:19:04,970 --> 00:19:09,580 So that's a level which we can call reflective thinking. 324 00:19:16,340 --> 00:19:20,630 Well, you know, it doesn't stop there, because she also 325 00:19:20,630 --> 00:19:24,690 considers, in another part of the story, something about 326 00:19:24,690 --> 00:19:27,690 being uneasy about arriving late. 327 00:19:27,690 --> 00:19:29,790 So she's not only just thinking about events that are 328 00:19:29,790 --> 00:19:32,890 going on in her mind right now, but events that are going 329 00:19:32,890 --> 00:19:36,610 on right now relative to plans she's made. 330 00:19:36,610 --> 00:19:39,660 Some Marvin calls that the self-reflecting layer. 331 00:19:48,820 --> 00:19:52,280 But that isn't the whole thing either, because toward the end 332 00:19:52,280 --> 00:19:54,560 of the story she starts to worry about what her friends 333 00:19:54,560 --> 00:19:55,970 would think of her. 334 00:19:55,970 --> 00:19:58,580 So there's a kind of reflective thinking in a more 335 00:19:58,580 --> 00:20:00,410 social context. 336 00:20:00,410 --> 00:20:02,590 So he calls that self-conscious thinking. 337 00:20:11,280 --> 00:20:15,270 So as the Carnegie folks think, the SOAR architecture 338 00:20:15,270 --> 00:20:19,830 focuses mostly on problem solving, Minsky's "Emotion 339 00:20:19,830 --> 00:20:25,980 Machine" book considers not just thinking, but thinking on 340 00:20:25,980 --> 00:20:28,700 many layers. 341 00:20:28,700 --> 00:20:34,740 And the blocker to doing any of that can be said to be the 342 00:20:34,740 --> 00:20:38,700 development of common sense, which computers, alas, have 343 00:20:38,700 --> 00:20:40,430 never had much of. 344 00:20:40,430 --> 00:20:47,170 So this could be said to be based on the common sense 345 00:20:47,170 --> 00:20:48,420 hypothesis. 346 00:20:53,180 --> 00:20:55,530 And the common sense hypothesis holds that in order 347 00:20:55,530 --> 00:20:57,590 to do all of that stuff, you have to have 348 00:20:57,590 --> 00:20:59,650 common sense like people. 349 00:20:59,650 --> 00:21:01,420 And if you have to have common sense like people, you have to 350 00:21:01,420 --> 00:21:04,700 think about how much of that is there and how 351 00:21:04,700 --> 00:21:06,460 can we go get it? 352 00:21:06,460 --> 00:21:13,390 And so this spawned a lot of activity in the media lab 353 00:21:13,390 --> 00:21:15,680 amongst people influenced by Marvin, having to do with 354 00:21:15,680 --> 00:21:17,340 gathering common sense. 355 00:21:17,340 --> 00:21:20,420 The open mind project, the work of Henry Lieberman and 356 00:21:20,420 --> 00:21:24,260 others, having to do with the gathering of common sense from 357 00:21:24,260 --> 00:21:29,330 the world wide web as a way of populating systems that would 358 00:21:29,330 --> 00:21:32,470 lay the foundation for doing this kind of layered thinking. 359 00:21:36,130 --> 00:21:43,290 So that is a brief survey of some mechanisms, some older 360 00:21:43,290 --> 00:21:47,210 than others, but all but GPS-- 361 00:21:47,210 --> 00:21:48,070 GPS too. 362 00:21:48,070 --> 00:21:50,630 Let's face it, it's hard to think of solving any problem 363 00:21:50,630 --> 00:21:52,020 without means-ends analysis being involved. 364 00:21:52,020 --> 00:21:57,380 So GPS isn't wrong, it's just not the only tool you need to 365 00:21:57,380 --> 00:21:59,430 think about what to do. 366 00:21:59,430 --> 00:22:03,950 So these are early, and late, and still-current. 367 00:22:03,950 --> 00:22:07,260 But it's not the only thing there is, because there have 368 00:22:07,260 --> 00:22:09,850 been reactions against this problem-solving way of 369 00:22:09,850 --> 00:22:12,740 thinking about the development of intelligence. 370 00:22:12,740 --> 00:22:16,120 And the most prominent of those counter currents, of 371 00:22:16,120 --> 00:22:19,900 those alternative ideas, belongs to Rod Brooks and his 372 00:22:19,900 --> 00:22:22,130 subsumption architecture. 373 00:22:22,130 --> 00:22:24,950 So along about the early-- 374 00:22:24,950 --> 00:22:30,100 along about the years surrounding 1990, Brooks 375 00:22:30,100 --> 00:22:31,350 became upset-- 376 00:22:33,750 --> 00:22:35,000 subsumption-- 377 00:22:39,190 --> 00:22:41,610 because robots couldn't do much. 378 00:22:41,610 --> 00:22:43,980 They would turn them on at night, and then the next 379 00:22:43,980 --> 00:22:45,750 morning they'd come in the laboratory and they would have 380 00:22:45,750 --> 00:22:51,820 moved 25 feet, nicely avoiding a table perhaps. 381 00:22:51,820 --> 00:22:55,500 Not doing very much and taking a long time to do it. 382 00:22:55,500 --> 00:22:58,070 So he had decided that it's because people were thinking 383 00:22:58,070 --> 00:23:00,590 in the wrong way. 384 00:23:00,590 --> 00:23:04,770 In those days people thought that the way you build a robot 385 00:23:04,770 --> 00:23:07,970 is you build a vision system, and then you build a reasoning 386 00:23:07,970 --> 00:23:11,340 system, and then you build an action system. 387 00:23:11,340 --> 00:23:15,830 And it can do almost nothing, but it does something. 388 00:23:15,830 --> 00:23:18,980 So you improve the vision system, and improve the 389 00:23:18,980 --> 00:23:22,530 reasoning system, and improve the actual system. 390 00:23:22,530 --> 00:23:25,090 And now you've broken it, because all the stuff you used 391 00:23:25,090 --> 00:23:29,000 to be able to do doesn't work anymore. 392 00:23:29,000 --> 00:23:30,490 So what's the alternative? 393 00:23:30,490 --> 00:23:33,320 Well, the alternative, as articulated by Brooks, is to 394 00:23:33,320 --> 00:23:35,900 turn this idea on its side. 395 00:23:35,900 --> 00:23:40,570 So instead of having an encapsulated vision system, an 396 00:23:40,570 --> 00:23:42,960 encapsulated reasoning system, and an encapsulated action 397 00:23:42,960 --> 00:23:47,390 system, what you have is layers that are focused not on 398 00:23:47,390 --> 00:23:50,820 the sensing and the reasoning and the action, but layers 399 00:23:50,820 --> 00:23:55,260 that are specialized to dealing with the world. 400 00:23:55,260 --> 00:23:59,000 So in Brook's way of thinking about things, at the lowest 401 00:23:59,000 --> 00:24:01,880 level you might have a system that's capable of-- 402 00:24:01,880 --> 00:24:06,580 well, before we get to that, avoiding objects. 403 00:24:06,580 --> 00:24:08,820 And maybe the next level up is the wandering layer. 404 00:24:11,460 --> 00:24:14,835 And maybe the next level up after that is explore. 405 00:24:17,990 --> 00:24:20,140 And maybe the next level up after that is seek. 406 00:24:24,060 --> 00:24:26,550 Now in the old days when people took 6001 I had no 407 00:24:26,550 --> 00:24:29,930 trouble getting an answer the question, what does this 408 00:24:29,930 --> 00:24:32,940 remind you of in 6001? 409 00:24:32,940 --> 00:24:34,530 It doesn't remind you of anything in 6001 since you 410 00:24:34,530 --> 00:24:35,740 haven't taken it. 411 00:24:35,740 --> 00:24:38,780 But it viewed, as a generalization of a 412 00:24:38,780 --> 00:24:42,620 programming idea, what is the programming idea? 413 00:24:42,620 --> 00:24:45,740 There are only a few powerful ideas in program, and this is 414 00:24:45,740 --> 00:24:48,180 a generalization of one of them. 415 00:24:48,180 --> 00:24:49,070 What is it? 416 00:24:49,070 --> 00:24:50,400 Do you have a name? 417 00:24:50,400 --> 00:24:51,412 Yes, Andrew? 418 00:24:51,412 --> 00:24:52,620 STUDENT: Layers of abstraction? 419 00:24:52,620 --> 00:24:54,210 PROFESSOR: Layers of abstraction, 420 00:24:54,210 --> 00:24:56,760 and abstraction barriers. 421 00:24:56,760 --> 00:24:58,500 That nails it pretty well. 422 00:24:58,500 --> 00:25:03,690 Because each of these guys can have its own vision, action, 423 00:25:03,690 --> 00:25:04,940 and reasoning system. 424 00:25:09,980 --> 00:25:12,870 And if you think of these as abstraction boundaries, then 425 00:25:12,870 --> 00:25:15,035 when you got this thing working you don't screw with 426 00:25:15,035 --> 00:25:16,460 it anymore. 427 00:25:16,460 --> 00:25:18,020 You build this layer on top. 428 00:25:18,020 --> 00:25:20,240 And it may reach down in here from time to time, but it 429 00:25:20,240 --> 00:25:22,990 doesn't fundamentally change it. 430 00:25:22,990 --> 00:25:25,790 Brooks was inspired in part by the way our brains are 431 00:25:25,790 --> 00:25:26,980 constructed. 432 00:25:26,980 --> 00:25:29,340 All that old stuff that we share with pigs is down in 433 00:25:29,340 --> 00:25:32,090 there deep, and we put the neocortex over it. 434 00:25:32,090 --> 00:25:34,660 So it looks layered in a way that would make 435 00:25:34,660 --> 00:25:36,960 [? Gerry Sussman ?] proud. 436 00:25:36,960 --> 00:25:42,610 So this then is the way that Brooks looks at the world, and 437 00:25:42,610 --> 00:25:47,310 it's characterized by a few features just like SOAR is. 438 00:25:47,310 --> 00:25:50,040 One of those features is no representation. 439 00:25:58,740 --> 00:26:03,920 So this is a detail that's probably right at the level 440 00:26:03,920 --> 00:26:07,620 that Brooks was operating, and very questionable when you get 441 00:26:07,620 --> 00:26:10,310 above the level that Brooks was operating. 442 00:26:10,310 --> 00:26:15,810 But before I go on, let me say what the hypothesis is. 443 00:26:15,810 --> 00:26:17,695 The hypothesis is the creature hypothesis. 444 00:26:25,550 --> 00:26:30,320 It's the hypothesis that once you can get a machine to act 445 00:26:30,320 --> 00:26:33,190 as smart as an insect, then the rest will be easy. 446 00:26:35,960 --> 00:26:36,930 Well, how do you get a creature to 447 00:26:36,930 --> 00:26:38,200 be smart as an insect? 448 00:26:38,200 --> 00:26:39,760 Maybe you don't need representation. 449 00:26:39,760 --> 00:26:42,330 We focused on representation in this course, so you can see 450 00:26:42,330 --> 00:26:43,580 there's a little stress--- 451 00:26:50,680 --> 00:26:52,210 Next thing is, what do you do if you don't have a 452 00:26:52,210 --> 00:26:54,490 representation? 453 00:26:54,490 --> 00:26:55,110 Let's see. 454 00:26:55,110 --> 00:26:57,430 Your representation makes a model possible. 455 00:26:57,430 --> 00:27:02,250 Models make it possible to predict, to understand, to 456 00:27:02,250 --> 00:27:03,780 explain, and to control. 457 00:27:03,780 --> 00:27:06,630 So if you don't have one what can you possibly do? 458 00:27:06,630 --> 00:27:14,470 Brooks' answer is, you use the world instead of a model. 459 00:27:21,500 --> 00:27:23,790 So everything you do is reactive. 460 00:27:23,790 --> 00:27:25,210 You don't have anything in your head that is 461 00:27:25,210 --> 00:27:27,990 a map of this room. 462 00:27:27,990 --> 00:27:30,930 But maybe I don't need one because I can get around that 463 00:27:30,930 --> 00:27:33,720 table by constantly observing it. 464 00:27:33,720 --> 00:27:35,200 And we don't have to fill up the memory with that 465 00:27:35,200 --> 00:27:36,710 information, I can just react to it. 466 00:27:40,560 --> 00:27:46,010 So no representation, use the world instead of a model, and 467 00:27:46,010 --> 00:27:50,630 the mechanisms in their purest form are just 468 00:27:50,630 --> 00:27:51,880 finite-state machines. 469 00:27:59,350 --> 00:28:02,940 So with that, Brooks was able to do things that people were 470 00:28:02,940 --> 00:28:06,350 never able to do before. 471 00:28:06,350 --> 00:28:08,330 And what's the modern [? instantiation ?] 472 00:28:08,330 --> 00:28:10,390 of this architecture? 473 00:28:10,390 --> 00:28:14,010 Now, according to Brooks, in use in 5 million homes in the 474 00:28:14,010 --> 00:28:15,260 United States? 475 00:28:18,009 --> 00:28:18,950 STUDENT: The Roomba? 476 00:28:18,950 --> 00:28:23,900 PROFESSOR: It's the Roomba The Roomba robot is, by Brooks' 477 00:28:23,900 --> 00:28:25,850 account, approximately the thirteenth 478 00:28:25,850 --> 00:28:27,960 business plan of iRobot. 479 00:28:27,960 --> 00:28:32,740 And it's the one that made it big, because the Rumba vacuum 480 00:28:32,740 --> 00:28:35,610 cleaner has been very successful. 481 00:28:35,610 --> 00:28:38,215 Would you like to see a movie of its processor? 482 00:28:40,960 --> 00:28:45,520 So this is a film made some time ago that shows, in some 483 00:28:45,520 --> 00:28:49,370 sense, the summa of that architecture. 484 00:28:49,370 --> 00:28:53,300 What I want you to imagine very briefly is a robot that 485 00:28:53,300 --> 00:28:57,600 wanders around in the halls and rooms of the old 486 00:28:57,600 --> 00:28:58,290 [? Tech Square ?] 487 00:28:58,290 --> 00:29:01,440 clinking the Coke cans. 488 00:29:01,440 --> 00:29:04,920 Okay, you all got an image of that in your mind? 489 00:29:04,920 --> 00:29:07,260 Because I want you to compare the image you now have of that 490 00:29:07,260 --> 00:29:10,550 robot that's wandering around collecting the Coke can, with 491 00:29:10,550 --> 00:29:11,800 the actual movie. 492 00:29:21,693 --> 00:29:22,180 [VIDEO PLAYBACK] 493 00:29:22,180 --> 00:29:25,520 -Herbert, the soda-can collecting mobile robot. 494 00:29:25,520 --> 00:29:29,360 He was built at the MIT AI lab in 1989. 495 00:29:29,360 --> 00:29:32,606 Work was done by John Cannell under the supervision of 496 00:29:32,606 --> 00:29:34,820 Rodney Brooks. 497 00:29:34,820 --> 00:29:38,800 Herbert is a robot controlled by subsumption architecture. 498 00:29:38,800 --> 00:29:43,700 This is a collection of small behaviors that influence the 499 00:29:43,700 --> 00:29:45,920 overall activities of the robot. 500 00:29:45,920 --> 00:29:48,310 There are no centralized controllers 501 00:29:48,310 --> 00:29:51,380 and no world model. 502 00:29:51,380 --> 00:29:55,890 -Herbert navigates by using a number of infrared proximity 503 00:29:55,890 --> 00:29:58,610 censors around its body and basically 504 00:29:58,610 --> 00:30:01,350 following walls and corridors. 505 00:30:01,350 --> 00:30:04,710 It can also look for the can through a laser light striper. 506 00:30:04,710 --> 00:30:07,370 Right now it's come out of the door of an office, followed 507 00:30:07,370 --> 00:30:12,350 along the wall, and then its laser light striper has seen a 508 00:30:12,350 --> 00:30:15,400 can on top of the desk in front of it. 509 00:30:15,400 --> 00:30:20,060 When this happens the robots and deploys its arm. 510 00:30:20,060 --> 00:30:21,560 You can see the arm going out now. 511 00:30:24,180 --> 00:30:26,470 -The arm has a number of censors itself. 512 00:30:26,470 --> 00:30:30,130 There are fingertip censors, a break beam in the jaws, and 513 00:30:30,130 --> 00:30:34,340 two infrared proximity sensors on the front of the hand. 514 00:30:34,340 --> 00:30:36,960 -It grabs cans in a stereotype fashion. 515 00:30:36,960 --> 00:30:40,720 First, it lowers down to find a surface somewhere, then it 516 00:30:40,720 --> 00:30:42,630 bounces along the surface until it 517 00:30:42,630 --> 00:30:44,630 sees the can in front. 518 00:30:44,630 --> 00:30:48,495 It uses the hand-based IRs to re-center the arm by rotating 519 00:30:48,495 --> 00:30:52,380 the robot's body until the can comes between the jaws of the 520 00:30:52,380 --> 00:30:57,960 gripper, at which point the break-beam senses the can. 521 00:30:57,960 --> 00:31:00,050 -After acquiring the can, Herbert will have tucked the 522 00:31:00,050 --> 00:31:03,371 arm back into its normal traveling configuration and 523 00:31:03,371 --> 00:31:04,621 attempt to go home. 524 00:31:13,740 --> 00:31:15,560 -Since it has no central [? arm presentation, ?] it 525 00:31:15,560 --> 00:31:17,400 doesn't have any map of where it came from. 526 00:31:17,400 --> 00:31:21,550 Instead, it has an algorithm which uses a magnetic compass 527 00:31:21,550 --> 00:31:25,000 to determine every time it comes through a door, will it 528 00:31:25,000 --> 00:31:26,290 be able to find the door? 529 00:31:26,290 --> 00:31:29,770 It basically has a policy of always going north every time 530 00:31:29,770 --> 00:31:33,220 it exits the door. 531 00:31:33,220 --> 00:31:36,280 -So now the can is being tucked away. 532 00:31:36,280 --> 00:31:39,040 As the robot turns you'll see a red stripe from the laser 533 00:31:39,040 --> 00:31:40,290 range finder. 534 00:31:46,240 --> 00:31:47,410 And now it's using the [INAUDIBLE] 535 00:31:47,410 --> 00:31:51,530 IR to navigate back, find the door, and go through the door 536 00:31:51,530 --> 00:31:52,780 with its prize. 537 00:32:07,616 --> 00:32:08,120 [END VIDEO PLAYBACK] 538 00:32:08,120 --> 00:32:09,700 PROFESSOR: And there, if you were paying attention, you saw 539 00:32:09,700 --> 00:32:12,840 a little glimpse of John Cannell who was the student to 540 00:32:12,840 --> 00:32:13,700 develop that system. 541 00:32:13,700 --> 00:32:16,950 So that was a tour de force. 542 00:32:16,950 --> 00:32:18,070 That was a magic moment. 543 00:32:18,070 --> 00:32:19,970 That was when you open the champagne. 544 00:32:19,970 --> 00:32:22,430 It's not what you expected, of course, because when I say 545 00:32:22,430 --> 00:32:25,280 imagine a robot wandering around in [? Tech Square ?] 546 00:32:25,280 --> 00:32:28,160 picking up Coke cans, that leaves open a huge envelope of 547 00:32:28,160 --> 00:32:30,120 possible hallucinations. 548 00:32:30,120 --> 00:32:33,930 And usually or hallucinations about these things are-- 549 00:32:33,930 --> 00:32:38,590 we imagine things to be more fluid, more natural, and more 550 00:32:38,590 --> 00:32:39,780 impressive than they actually are. 551 00:32:39,780 --> 00:32:43,680 But that was impressive, because no robot came close to 552 00:32:43,680 --> 00:32:45,655 doing anything like that before. 553 00:32:49,030 --> 00:32:50,740 More to be said about that during the 554 00:32:50,740 --> 00:32:53,060 business lecture on Wednesday. 555 00:32:56,000 --> 00:32:58,810 So that's the subsumption architecture. 556 00:32:58,810 --> 00:33:06,230 By the way, maybe at this point we can say something 557 00:33:06,230 --> 00:33:10,020 about how the other architectures relate to what 558 00:33:10,020 --> 00:33:12,460 Minsky was talking about. 559 00:33:12,460 --> 00:33:18,270 What's this deliberative thinking layer correspond to? 560 00:33:18,270 --> 00:33:24,370 That's what SOAR is about, and maybe GPS. 561 00:33:27,020 --> 00:33:30,200 So what's subsumption about? 562 00:33:30,200 --> 00:33:31,860 It's about stuff down here. 563 00:33:38,400 --> 00:33:42,685 It's about instinctive reaction and learned reaction. 564 00:33:55,290 --> 00:33:59,120 But shoot, what about Minsky's other layers? 565 00:33:59,120 --> 00:34:02,180 If we're going to be building systems that are as smart as 566 00:34:02,180 --> 00:34:04,390 those things then we have to worry a little bit about that 567 00:34:04,390 --> 00:34:05,710 sort of thing too. 568 00:34:05,710 --> 00:34:09,080 So that brings us to the genesis architecture. 569 00:34:09,080 --> 00:34:12,679 And now let me give you the standard caution that should 570 00:34:12,679 --> 00:34:15,500 be early in the presentation of any academic. 571 00:34:15,500 --> 00:34:19,940 I will sometimes say "I," and what I mean is "we." And 572 00:34:19,940 --> 00:34:23,540 sometimes I'll say "we," and what I mean is "they." This 573 00:34:23,540 --> 00:34:27,739 was a system that was developed mostly by students 574 00:34:27,739 --> 00:34:31,280 of mine who persuaded me, after a great deal of time, 575 00:34:31,280 --> 00:34:34,420 that they were thinking the right kinds of thoughts. 576 00:34:34,420 --> 00:34:38,469 But here's how the genesis architecture works. 577 00:34:38,469 --> 00:34:43,940 As no surprise, given recent discussions, it's all centered 578 00:34:43,940 --> 00:34:45,190 on language. 579 00:34:50,429 --> 00:34:54,880 And the language part of the genesis system has two roles, 580 00:34:54,880 --> 00:34:59,910 one of which is to guide, and marshal, and interact with the 581 00:34:59,910 --> 00:35:01,160 perceptual systems. 582 00:35:07,230 --> 00:35:14,645 And the other is to enable the description of events. 583 00:35:36,980 --> 00:35:37,610 That's how it works. 584 00:35:37,610 --> 00:35:41,372 So is perception important? 585 00:35:41,372 --> 00:35:43,290 I don't know. 586 00:35:43,290 --> 00:35:47,130 I might ask you a question like, is there anybody sitting 587 00:35:47,130 --> 00:35:50,440 in the front row wearing blue jeans? 588 00:35:50,440 --> 00:35:52,410 And it's hard for you to resist, under those 589 00:35:52,410 --> 00:35:55,370 circumstances, your eyes from going over there and answering 590 00:35:55,370 --> 00:35:56,350 the question. 591 00:35:56,350 --> 00:35:58,180 Your eyes answer the question. 592 00:35:58,180 --> 00:36:02,780 No symbol processing system is involved, except in so far as 593 00:36:02,780 --> 00:36:05,600 my language system has communicated with their 594 00:36:05,600 --> 00:36:08,540 language system, which drives your motor system and your 595 00:36:08,540 --> 00:36:10,690 vision system to go over there and answer the 596 00:36:10,690 --> 00:36:13,490 question for you. 597 00:36:13,490 --> 00:36:16,980 But it's not just the real stuff that the language system 598 00:36:16,980 --> 00:36:18,040 directs your attention to. 599 00:36:18,040 --> 00:36:20,970 It's also the imagined stuff. 600 00:36:20,970 --> 00:36:22,160 It's been a long semester. 601 00:36:22,160 --> 00:36:25,010 Have I told you the story about my table saw? 602 00:36:25,010 --> 00:36:25,580 Probably not. 603 00:36:25,580 --> 00:36:26,120 Here's the deal. 604 00:36:26,120 --> 00:36:27,630 I bought a table saw. 605 00:36:27,630 --> 00:36:29,560 It's a wonderful table saw. 606 00:36:29,560 --> 00:36:30,940 I was installing it with a friend of mine 607 00:36:30,940 --> 00:36:33,190 who's a cabinet maker. 608 00:36:33,190 --> 00:36:37,910 He said, never wear gloves when you operate the saw. 609 00:36:37,910 --> 00:36:39,950 "Why?" I said. 610 00:36:39,950 --> 00:36:41,900 Before he could answer the question I figured it out. 611 00:36:41,900 --> 00:36:43,540 Can you figure out why you never wear gloves when you 612 00:36:43,540 --> 00:36:45,450 operate a table saw? 613 00:36:45,450 --> 00:36:46,930 You know what a table saw is, right? 614 00:36:46,930 --> 00:36:48,650 It's a table with a spinning blade in the middle. 615 00:36:48,650 --> 00:36:51,640 And you use it to cut wood. 616 00:36:51,640 --> 00:36:55,095 Why should you never wear gloves? 617 00:36:55,095 --> 00:36:55,582 Yes? 618 00:36:55,582 --> 00:36:56,560 STUDENT: Well-- 619 00:36:56,560 --> 00:36:57,770 STUDENT: --Well, you know the answer. 620 00:36:57,770 --> 00:36:59,170 Ha, that's not fair. 621 00:36:59,170 --> 00:37:00,230 That's old Brett up there. 622 00:37:00,230 --> 00:37:03,180 He's heard the story too many times. 623 00:37:03,180 --> 00:37:05,042 Yes, Andrew, you got it. 624 00:37:05,042 --> 00:37:06,810 STUDENT: I've been told the answer before. 625 00:37:06,810 --> 00:37:07,960 PROFESSOR: You've been told the answer. 626 00:37:07,960 --> 00:37:09,740 How about somebody who hasn't been told the answer. 627 00:37:09,740 --> 00:37:10,130 Yes? 628 00:37:10,130 --> 00:37:11,920 STUDENT: Because the gloves might get caught. 629 00:37:11,920 --> 00:37:14,410 PROFESSOR: Because the glove might get caught and pull your 630 00:37:14,410 --> 00:37:15,670 hand into the blade. 631 00:37:15,670 --> 00:37:16,920 And then what happens? 632 00:37:19,140 --> 00:37:19,740 It's horrible. 633 00:37:19,740 --> 00:37:23,360 You're hand gets mangled and your fingers get cut off, and 634 00:37:23,360 --> 00:37:26,300 this happens a lot to professionals. 635 00:37:26,300 --> 00:37:28,790 It won't actually happen with that table saw that I bought, 636 00:37:28,790 --> 00:37:33,130 because its play detects flesh and stops the blade, which 637 00:37:33,130 --> 00:37:35,250 then leads to stopping the blade and having the blade 638 00:37:35,250 --> 00:37:38,140 retreat into the table in about two microseconds-- 639 00:37:38,140 --> 00:37:39,590 two milliseconds. 640 00:37:39,590 --> 00:37:42,460 So, in general though, it's a bad idea, and you always have 641 00:37:42,460 --> 00:37:46,490 to suppose that the mechanism isn't working anyway in order 642 00:37:46,490 --> 00:37:48,320 to use good safety practice. 643 00:37:48,320 --> 00:37:50,170 But here's an example of something that nobody ever 644 00:37:50,170 --> 00:37:53,560 told you that he was able to figure out, by imagining what 645 00:37:53,560 --> 00:37:56,815 would happen and reading the answers off of the scene that 646 00:37:56,815 --> 00:37:58,420 he imagined. 647 00:37:58,420 --> 00:38:02,260 So nobody ever says many of the things that we know, but 648 00:38:02,260 --> 00:38:03,000 we know them anyway. 649 00:38:03,000 --> 00:38:04,330 Here's another example. 650 00:38:04,330 --> 00:38:05,650 Imagine running down the street with a 651 00:38:05,650 --> 00:38:06,550 full bucket of water. 652 00:38:06,550 --> 00:38:09,190 What happens? 653 00:38:09,190 --> 00:38:12,110 The water splashes out and gets your leg wet, right? 654 00:38:12,110 --> 00:38:14,500 You won't find that in Open Minds database. 655 00:38:14,500 --> 00:38:16,630 Nobody ever said that over the web. 656 00:38:16,630 --> 00:38:17,620 It's not written down anywhere. 657 00:38:17,620 --> 00:38:19,040 But you know it. 658 00:38:19,040 --> 00:38:22,710 Because you, we human beings have the capacity to imagine 659 00:38:22,710 --> 00:38:27,090 perceptual things and read the answers to questions off of 660 00:38:27,090 --> 00:38:30,940 our imaginations with that perceptual apparatus. 661 00:38:30,940 --> 00:38:35,020 So that's a very important connection down there. 662 00:38:35,020 --> 00:38:38,150 And then if you've got the ability to describe events, 663 00:38:38,150 --> 00:38:40,930 then you've got the ability to tell and understand stories. 664 00:38:45,180 --> 00:38:48,030 And if you can do that, then you can start to get a handle 665 00:38:48,030 --> 00:38:56,160 on culture, both macro and micro. 666 00:38:58,790 --> 00:39:04,500 And by macro culture I mean the country you grew up in, 667 00:39:04,500 --> 00:39:06,320 the religion you grew up with. 668 00:39:06,320 --> 00:39:09,290 And by micro I mean your family and personal 669 00:39:09,290 --> 00:39:11,680 experience, and all shades in between. 670 00:39:14,570 --> 00:39:18,520 So I don't know, what inspires me and my associates to think 671 00:39:18,520 --> 00:39:19,250 in these terms? 672 00:39:19,250 --> 00:39:21,800 We talked about a little bit of it last time when I talked 673 00:39:21,800 --> 00:39:27,060 about evolution and the apparent flowering of our 674 00:39:27,060 --> 00:39:30,140 species about 50,000 years ago, at 675 00:39:30,140 --> 00:39:32,300 which time we got something. 676 00:39:32,300 --> 00:39:34,940 And I believe that what we got-- 677 00:39:34,940 --> 00:39:37,260 and this is the characterization of this 678 00:39:37,260 --> 00:39:40,540 particular hypothesis-- 679 00:39:40,540 --> 00:39:45,130 what we got is the ability to tell stories 680 00:39:45,130 --> 00:39:46,980 and understand them. 681 00:39:46,980 --> 00:39:52,540 So if we want to label this representation, it's the label 682 00:39:52,540 --> 00:39:55,840 strong story hypothesis. 683 00:40:02,200 --> 00:40:03,530 So what's the weak story hypothesis? 684 00:40:03,530 --> 00:40:06,540 The weak story hypothesis is, this is important. 685 00:40:06,540 --> 00:40:11,260 The strong story hypothesis is, this is all there is. 686 00:40:11,260 --> 00:40:14,190 But is there any other evidence of this is really, 687 00:40:14,190 --> 00:40:15,670 really, really important? 688 00:40:15,670 --> 00:40:20,520 So I've queried Krishna here before the class starts, and 689 00:40:20,520 --> 00:40:23,560 he tells me I haven't told you about the following 690 00:40:23,560 --> 00:40:24,850 experiment. 691 00:40:24,850 --> 00:40:28,300 This, in my way of thinking, is the most important series 692 00:40:28,300 --> 00:40:32,720 of experiments ever done in cognitive psychology, 693 00:40:32,720 --> 00:40:35,550 developmental psychology, actually. 694 00:40:35,550 --> 00:40:39,090 So here's how we get started. 695 00:40:43,380 --> 00:40:48,070 There's a rectangular room, if you're a person. 696 00:40:48,070 --> 00:40:51,490 If you're a rat, it's a rectangular box. 697 00:40:51,490 --> 00:40:53,720 All the walls are painted white. 698 00:40:53,720 --> 00:40:55,610 Are you with me so far? 699 00:40:55,610 --> 00:41:02,360 So now, in each corner there's a basket, or cloth, or 700 00:41:02,360 --> 00:41:10,320 something in which or under which you can put some food. 701 00:41:10,320 --> 00:41:15,055 Now, you put the food there while the rat watches you. 702 00:41:18,550 --> 00:41:26,540 And then you give the rat a little spin to disorient it. 703 00:41:26,540 --> 00:41:28,790 All right? 704 00:41:28,790 --> 00:41:34,130 So then, the rest stops and goes for the food. 705 00:41:34,130 --> 00:41:37,440 And you can keep track of where the rat goes. 706 00:41:37,440 --> 00:41:41,070 And the rat goes with approximate equal probability 707 00:41:41,070 --> 00:41:44,760 predominantly to those two corners. 708 00:41:44,760 --> 00:41:46,096 So I'd have bet you didn't know that 709 00:41:46,096 --> 00:41:48,040 rats were that smart. 710 00:41:48,040 --> 00:41:51,070 So they understand the rectangular nature of the room 711 00:41:51,070 --> 00:41:52,880 and they don't go to the diagonal corners where the 712 00:41:52,880 --> 00:41:55,090 food cannot be. 713 00:41:55,090 --> 00:41:58,390 So are these genius rats? 714 00:41:58,390 --> 00:42:00,050 Or maybe we're just rats with big brains. 715 00:42:00,050 --> 00:42:02,550 Because we do the same thing. 716 00:42:02,550 --> 00:42:08,220 So if you repeat this experiment and replace the rat 717 00:42:08,220 --> 00:42:14,470 with a small child, and then you put a toy in there instead 718 00:42:14,470 --> 00:42:17,520 of food, and the rat-- 719 00:42:17,520 --> 00:42:18,550 not the rat. 720 00:42:18,550 --> 00:42:24,600 The child is usually held in a parent's arms, usually the 721 00:42:24,600 --> 00:42:26,780 child's mother-- 722 00:42:26,780 --> 00:42:28,840 usually because they think that if they participate in 723 00:42:28,840 --> 00:42:31,320 these experiments up there at Harvard their kid will get 724 00:42:31,320 --> 00:42:34,290 into Harvard some day. 725 00:42:34,290 --> 00:42:37,015 So the kid goes to a diagonal corner just like a rat. 726 00:42:39,860 --> 00:42:46,790 And then the next thing you do is, you try an adult, maybe an 727 00:42:46,790 --> 00:42:48,890 MIT student. 728 00:42:48,890 --> 00:42:50,680 That way you can use food again. 729 00:42:55,650 --> 00:42:58,330 And you get the same result. 730 00:42:58,330 --> 00:43:00,680 Who could be surprised? 731 00:43:00,680 --> 00:43:04,570 So rats, children, and human adults, pretty much all the 732 00:43:04,570 --> 00:43:11,450 same with respect to this experiment, until you paint 733 00:43:11,450 --> 00:43:13,440 one wall blue. 734 00:43:13,440 --> 00:43:16,370 Rats are not colorblind, in case you're wondering. 735 00:43:16,370 --> 00:43:19,130 Then what happens? 736 00:43:19,130 --> 00:43:25,930 Well, if you pay one wall blue the rat still goes with equal 737 00:43:25,930 --> 00:43:29,560 probability to the two diagonal corners. 738 00:43:29,560 --> 00:43:33,370 If you paint one wall blue, the child still goes to the 739 00:43:33,370 --> 00:43:37,230 two diagonal corners with approximate equal probability. 740 00:43:37,230 --> 00:43:42,940 It's only us genius human adults who go 741 00:43:42,940 --> 00:43:44,190 only to that corner. 742 00:43:46,870 --> 00:43:48,285 So this invites a couple of questions. 743 00:43:51,690 --> 00:43:57,510 One of which is, when does a child become an adult? 744 00:43:57,510 --> 00:43:58,760 Any ideas? 745 00:44:00,990 --> 00:44:04,185 [INAUDIBLE], what do you think? 746 00:44:04,185 --> 00:44:06,610 STUDENT: [INAUDIBLE]. 747 00:44:06,610 --> 00:44:08,630 PROFESSOR: You can pick a number greater than 1 748 00:44:08,630 --> 00:44:14,620 and less than 10. 749 00:44:14,620 --> 00:44:16,024 [INAUDIBLE], what do you think? 750 00:44:16,024 --> 00:44:16,860 STUDENT: Five? 751 00:44:16,860 --> 00:44:19,550 PROFESSOR: It's a pretty good guess. 752 00:44:19,550 --> 00:44:21,920 Do you have siblings at that age? 753 00:44:21,920 --> 00:44:23,980 It's a surprise but, why is it five? 754 00:44:23,980 --> 00:44:25,480 Is it because-- 755 00:44:25,480 --> 00:44:26,910 what does it relate to? 756 00:44:26,910 --> 00:44:30,450 Is there any correlate to the onset of that ability? 757 00:44:30,450 --> 00:44:32,450 You might try everything, as [INAUDIBLE] does, because 758 00:44:32,450 --> 00:44:35,250 she's extremely careful. 759 00:44:35,250 --> 00:44:40,340 So she's tried gender, she's tried the onset of language, 760 00:44:40,340 --> 00:44:44,430 the appreciation of music, handedness, and there's only 761 00:44:44,430 --> 00:44:46,530 one thing that matters. 762 00:44:46,530 --> 00:44:51,640 And that is that the child becomes adult at that time 763 00:44:51,640 --> 00:44:54,440 when they start to use the word left and right when they 764 00:44:54,440 --> 00:44:57,450 describe the world. 765 00:44:57,450 --> 00:44:59,700 Now I said that very carefully because they understand left 766 00:44:59,700 --> 00:45:03,050 and right at an earlier age, but they only have started to 767 00:45:03,050 --> 00:45:05,530 use the words left and right when they describe the world 768 00:45:05,530 --> 00:45:10,900 at the time that they begin to break this symmetry and go to 769 00:45:10,900 --> 00:45:13,560 the correct corner. 770 00:45:13,560 --> 00:45:16,140 Now for the next element of this I need something to read. 771 00:45:16,140 --> 00:45:17,390 Has anyone got a textbook handy? 772 00:45:22,720 --> 00:45:27,680 Ah, "China, an Illustrated History." 773 00:45:27,680 --> 00:45:28,930 Now I need a volunteer. 774 00:45:30,875 --> 00:45:31,320 OK. 775 00:45:31,320 --> 00:45:32,440 Andrew, you want to do this? 776 00:45:32,440 --> 00:45:33,690 So here's what you're going to do. 777 00:45:37,080 --> 00:45:38,270 You can stay there. 778 00:45:38,270 --> 00:45:40,520 But you need to stand up. 779 00:45:40,520 --> 00:45:42,750 So what I'm going to do is, I'm going to read you a 780 00:45:42,750 --> 00:45:44,310 passage from this book. 781 00:45:44,310 --> 00:45:45,710 And I want you to say it back to me at the 782 00:45:45,710 --> 00:45:47,260 same time I read it. 783 00:45:47,260 --> 00:45:51,610 It's as if you're doing simultaneous translation, 784 00:45:51,610 --> 00:45:53,405 except it's English to English. 785 00:45:56,970 --> 00:45:59,160 This things got words I can't pronounce. 786 00:45:59,160 --> 00:46:01,080 OK, are you ready to go? 787 00:46:01,080 --> 00:46:02,220 All right. 788 00:46:02,220 --> 00:46:05,330 "When overwhelmed by the magnitude of the problems he 789 00:46:05,330 --> 00:46:08,730 tackled, he began to suspect that others were plotting 790 00:46:08,730 --> 00:46:14,600 against him or secretly ridiculing him." 791 00:46:14,600 --> 00:46:15,420 Thank you very much. 792 00:46:15,420 --> 00:46:16,560 That's great. 793 00:46:16,560 --> 00:46:17,880 So you see, he could do it. 794 00:46:17,880 --> 00:46:18,880 Some people can't do it. 795 00:46:18,880 --> 00:46:21,360 At least it take a little practice. 796 00:46:21,360 --> 00:46:22,780 But he did it. 797 00:46:22,780 --> 00:46:25,080 And guess what I've done to him? 798 00:46:25,080 --> 00:46:29,080 I've reduced his intelligence to that of a rat. 799 00:46:29,080 --> 00:46:36,930 Because if you do this experiment with an adult human 800 00:46:36,930 --> 00:46:38,850 who's doing this simultaneous English to English 801 00:46:38,850 --> 00:46:40,940 translation, they go with equal 802 00:46:40,940 --> 00:46:43,910 probability to the two corners. 803 00:46:43,910 --> 00:46:45,210 So what's happened? 804 00:46:45,210 --> 00:46:47,720 What's happened is you've jambed 805 00:46:47,720 --> 00:46:50,930 their language processor. 806 00:46:50,930 --> 00:46:54,070 And when their language processor is jambed they can't 807 00:46:54,070 --> 00:46:57,970 put the blue wall together with the rectangular shape. 808 00:46:57,970 --> 00:47:00,290 So it seems to be that language is the mediator of 809 00:47:00,290 --> 00:47:02,520 exactly the combinators you need in order to build 810 00:47:02,520 --> 00:47:04,250 descriptions. 811 00:47:04,250 --> 00:47:06,870 Because they can't even put those things together when 812 00:47:06,870 --> 00:47:09,870 their language processor is jambed by the simultaneous 813 00:47:09,870 --> 00:47:13,670 translation phenomenon. 814 00:47:13,670 --> 00:47:17,310 So that brings us to the two gold star ideas of the day. 815 00:47:17,310 --> 00:47:29,730 One is, if you want to make yourself smarter you want to 816 00:47:29,730 --> 00:47:30,690 do those things-- 817 00:47:30,690 --> 00:47:32,760 look, listen, draw, and talk. 818 00:47:32,760 --> 00:47:36,100 Because those are the particular mechanisms that 819 00:47:36,100 --> 00:47:40,440 surround this area down here, which is the center of what we 820 00:47:40,440 --> 00:47:43,120 do-- which is the center of our thinking. 821 00:47:43,120 --> 00:47:45,140 So why do you take notes in class? 822 00:47:45,140 --> 00:47:47,350 Now because you'll ever look at them before, but because it 823 00:47:47,350 --> 00:47:51,980 forces the engagement of your linguistic-- 824 00:47:51,980 --> 00:47:53,390 of your linguistic, your motor, 825 00:47:53,390 --> 00:47:54,820 and your visual apparatus. 826 00:47:54,820 --> 00:47:56,520 And that makes you smarter, because it's 827 00:47:56,520 --> 00:47:59,520 exercising that stuff. 828 00:47:59,520 --> 00:48:02,560 The second thing you can say, in conclusion, especially from 829 00:48:02,560 --> 00:48:12,355 this experiment, is beware of fast talkers. 830 00:48:16,820 --> 00:48:19,260 Why do you want to be aware of fast talkers. 831 00:48:19,260 --> 00:48:22,430 It's not because they will talk you into anything. 832 00:48:22,430 --> 00:48:24,940 It's because that when they talk fast they're jambing your 833 00:48:24,940 --> 00:48:28,000 language processor and you can't thing. 834 00:48:28,000 --> 00:48:32,600 That's why you want to be aware of fast talkers. 835 00:48:32,600 --> 00:48:34,850 Because if they jamb your language processor you won't 836 00:48:34,850 --> 00:48:38,200 thinking and you'll buy that car, or you'll buy that drink, 837 00:48:38,200 --> 00:48:40,925 or you'll do any manner of things that people who want 838 00:48:40,925 --> 00:48:45,272 you to do those things have learned to do by talking to 839 00:48:45,272 --> 00:48:47,960 jamb your processor. 840 00:48:47,960 --> 00:48:49,620 So that completes what we're going to do today. 841 00:48:49,620 --> 00:48:51,290 And I'll give you a demonstration of some of this 842 00:48:51,290 --> 00:48:52,540 stuff on another occasion.