1 00:00:00,090 --> 00:00:02,490 The following content is provided under a Creative 2 00:00:02,490 --> 00:00:04,030 Commons license. 3 00:00:04,030 --> 00:00:06,330 Your support will help MIT OpenCourseWare 4 00:00:06,330 --> 00:00:10,690 continue to offer high quality educational resources for free. 5 00:00:10,690 --> 00:00:13,320 To make a donation or view additional materials 6 00:00:13,320 --> 00:00:17,280 from hundreds of MIT courses, visit MIT OpenCourseWare 7 00:00:17,280 --> 00:00:19,980 at ocw.mit.edu. 8 00:00:19,980 --> 00:00:21,150 ABBY NOYCE: OK. 9 00:00:21,150 --> 00:00:22,590 So, the plan. 10 00:00:22,590 --> 00:00:25,200 The plan is that today, we are going to talk about-- 11 00:00:25,200 --> 00:00:27,720 we talked yesterday about models of working memory. 12 00:00:27,720 --> 00:00:29,610 We're going to talk today about some 13 00:00:29,610 --> 00:00:32,630 of the neuro stuff underlying working 14 00:00:32,630 --> 00:00:34,380 memory, what we know about how all of this 15 00:00:34,380 --> 00:00:36,810 is actually implemented. 16 00:00:36,810 --> 00:00:39,000 And then, if we have time, we're going 17 00:00:39,000 --> 00:00:41,250 to segue into talking a little bit 18 00:00:41,250 --> 00:00:45,134 about longer-term representations of knowledge 19 00:00:45,134 --> 00:00:46,800 to get ready for next week, when we talk 20 00:00:46,800 --> 00:00:50,100 about longer-term memory, and we talk 21 00:00:50,100 --> 00:00:52,410 about decision making and working with the knowledge 22 00:00:52,410 --> 00:00:56,620 that you have, and things like that. 23 00:00:56,620 --> 00:01:01,450 So just as a refresher, this is Baddeley's model 24 00:01:01,450 --> 00:01:03,820 that we talked about yesterday. 25 00:01:03,820 --> 00:01:06,640 Baddeley and Hitch said working memory is not 26 00:01:06,640 --> 00:01:08,110 a unified capability. 27 00:01:08,110 --> 00:01:10,270 Working memory has different pieces. 28 00:01:10,270 --> 00:01:12,370 In particular, they pointed out that you 29 00:01:12,370 --> 00:01:16,450 have a visual-spatial capability that 30 00:01:16,450 --> 00:01:19,510 seems to be separate from your auditory-phonological 31 00:01:19,510 --> 00:01:20,680 capability. 32 00:01:20,680 --> 00:01:24,130 You can do tasks that use both these pieces of working memory 33 00:01:24,130 --> 00:01:27,160 without one interfering with the other. 34 00:01:27,160 --> 00:01:30,760 And we said that there is this central control executive 35 00:01:30,760 --> 00:01:35,110 process that coordinates both of these buffers, 36 00:01:35,110 --> 00:01:38,680 the phonological buffer and the visual-spatial buffer, 37 00:01:38,680 --> 00:01:42,550 decides what information goes into and out of those buffers, 38 00:01:42,550 --> 00:01:46,000 can work with it, do tasks on it. 39 00:01:46,000 --> 00:01:48,220 So remember, yesterday we had the-- 40 00:01:48,220 --> 00:01:50,740 picture a D, rotate it to the right, picture a 4 41 00:01:50,740 --> 00:01:52,010 on top of it. 42 00:01:52,010 --> 00:01:54,640 So that's, you're using your visual-spatial sketchpad 43 00:01:54,640 --> 00:01:56,770 to store that information, but what's 44 00:01:56,770 --> 00:01:58,630 actually manipulating it and moving 45 00:01:58,630 --> 00:02:01,580 it around is that central executive process. 46 00:02:01,580 --> 00:02:05,080 So that's an example of where one is working on the other. 47 00:02:05,080 --> 00:02:11,920 Central executive can also post up from long-term memory. 48 00:02:11,920 --> 00:02:14,530 It also takes into account the sensory input 49 00:02:14,530 --> 00:02:16,540 that you're getting. 50 00:02:16,540 --> 00:02:19,390 Generally it's in charge of-- 51 00:02:19,390 --> 00:02:21,500 with some quotes around that-- 52 00:02:21,500 --> 00:02:26,200 what is actually in your mind at any given point in time. 53 00:02:26,200 --> 00:02:33,305 AUDIENCE: [INAUDIBLE] 54 00:02:33,305 --> 00:02:34,430 ABBY NOYCE: Yeah, well, OK. 55 00:02:34,430 --> 00:02:37,110 So the central executive doesn't necessarily 56 00:02:37,110 --> 00:02:40,770 get to control how things move in and out of long term 57 00:02:40,770 --> 00:02:42,990 memory the way that it controls what's 58 00:02:42,990 --> 00:02:46,320 going in and out of those two sensory buffers. 59 00:02:46,320 --> 00:02:49,320 So the relationship between them is different, 60 00:02:49,320 --> 00:02:52,080 which is kind of what I was trying to signify by drawing it 61 00:02:52,080 --> 00:02:53,800 slightly differently. 62 00:02:53,800 --> 00:02:55,010 It made sense. 63 00:02:55,010 --> 00:02:55,510 It did. 64 00:02:58,330 --> 00:02:59,370 OK. 65 00:02:59,370 --> 00:03:02,490 So, we've got this lovely model, which you might notice 66 00:03:02,490 --> 00:03:06,480 says nothing at all about the brain. 67 00:03:06,480 --> 00:03:08,610 So what are we talking about when 68 00:03:08,610 --> 00:03:10,530 we say you've got these storage buffers? 69 00:03:10,530 --> 00:03:13,290 What's in them? 70 00:03:13,290 --> 00:03:16,980 So, we generally believe that working memory-- 71 00:03:16,980 --> 00:03:20,790 memory when you're actively holding information, 72 00:03:20,790 --> 00:03:22,680 working with information-- 73 00:03:22,680 --> 00:03:26,100 is what's called activity based memory. 74 00:03:26,100 --> 00:03:29,180 So it depends on sustained activation 75 00:03:29,180 --> 00:03:32,970 in a population of neurons, a group of neurons that 76 00:03:32,970 --> 00:03:36,420 keep firing in a particular way in order to hold 77 00:03:36,420 --> 00:03:40,350 that representation in mind. 78 00:03:40,350 --> 00:03:42,480 Having it kind of actively maintained 79 00:03:42,480 --> 00:03:45,234 in a group of neurons like that makes it right there. 80 00:03:45,234 --> 00:03:46,900 It's easy to work with this information. 81 00:03:46,900 --> 00:03:50,790 You don't have to pull up a new representation. 82 00:03:50,790 --> 00:03:52,256 But it's also less permanent. 83 00:03:52,256 --> 00:03:53,880 There's only so long that these neurons 84 00:03:53,880 --> 00:03:56,850 are going to keep firing in a particular pattern. 85 00:03:56,850 --> 00:03:59,790 Longer term memory, like we'll talk to about next week, 86 00:03:59,790 --> 00:04:03,600 actually changes the waiting of synapses between neurons 87 00:04:03,600 --> 00:04:06,810 to put patterns into the way your brain is 88 00:04:06,810 --> 00:04:10,260 connected versus merely in the way it's acting right now. 89 00:04:13,970 --> 00:04:16,519 The reason it's less permanent is because as soon 90 00:04:16,519 --> 00:04:20,779 as these neurons stop being told to stay active in some way, 91 00:04:20,779 --> 00:04:22,310 then-- 92 00:04:22,310 --> 00:04:23,510 boom! 93 00:04:23,510 --> 00:04:25,890 They're going to stop firing in the pattern that 94 00:04:25,890 --> 00:04:27,890 is the representation of whatever you're holding 95 00:04:27,890 --> 00:04:31,670 in mind, and this short term working piece of information 96 00:04:31,670 --> 00:04:33,320 is going to be gone. 97 00:04:39,200 --> 00:04:43,750 This is a theory in the scientific sense. 98 00:04:43,750 --> 00:04:45,390 In the, this is how we believe it works 99 00:04:45,390 --> 00:04:47,120 and there is good evidence for it sense, 100 00:04:47,120 --> 00:04:48,330 like the theory of gravity. 101 00:04:54,306 --> 00:04:55,302 All right. 102 00:05:20,240 --> 00:05:21,450 OK. 103 00:05:21,450 --> 00:05:22,620 Brief project moment. 104 00:05:22,620 --> 00:05:24,810 So we've got this hypothesis. 105 00:05:24,810 --> 00:05:29,090 We believe that working memory depends on sustained activity 106 00:05:29,090 --> 00:05:31,970 of a population of neurons. 107 00:05:31,970 --> 00:05:34,190 Time to think about what we've been thinking about, 108 00:05:34,190 --> 00:05:37,640 about how we find things out in this field. 109 00:05:37,640 --> 00:05:41,270 Your mission is to take five or 10 minutes 110 00:05:41,270 --> 00:05:45,860 and think about how you would support or refute 111 00:05:45,860 --> 00:05:51,680 this hypothesis: Working memory depends upon sustained activity 112 00:05:51,680 --> 00:05:54,120 of a population of neurons. 113 00:05:54,120 --> 00:05:57,950 So, write down answers to those four questions. 114 00:05:57,950 --> 00:05:59,930 You can talk to your classmates. 115 00:05:59,930 --> 00:06:01,430 This doesn't need to be a paragraph. 116 00:06:01,430 --> 00:06:03,290 You're not going to be handing it in, 117 00:06:03,290 --> 00:06:06,470 but we're going to talk about what we would do in order 118 00:06:06,470 --> 00:06:09,580 to make this work. 119 00:06:09,580 --> 00:06:11,930 Remember, an independent variable is the thing 120 00:06:11,930 --> 00:06:14,850 that you, the experimenter, are going to manipulate, 121 00:06:14,850 --> 00:06:15,947 are going to change. 122 00:06:15,947 --> 00:06:17,780 And then the dependent variable is the thing 123 00:06:17,780 --> 00:06:18,779 you're going to measure. 124 00:06:18,779 --> 00:06:20,627 You're going to see what happens. 125 00:06:20,627 --> 00:06:22,460 And then operational definitions are kind of 126 00:06:22,460 --> 00:06:24,590 like how you're going to change or how you're 127 00:06:24,590 --> 00:06:27,660 going to measure these things. 128 00:06:27,660 --> 00:06:29,514 Let's look at some actual studies 129 00:06:29,514 --> 00:06:31,430 and we'll see ones that set this up both ways, 130 00:06:31,430 --> 00:06:33,860 both with working memory as an independent 131 00:06:33,860 --> 00:06:36,650 or a dependent variable. 132 00:06:36,650 --> 00:06:40,990 These are some of the classic single cell recordings. 133 00:06:40,990 --> 00:06:44,480 Monkeys, animal models. 134 00:06:44,480 --> 00:06:47,290 So single cell recordings is that electrodes into the brain 135 00:06:47,290 --> 00:06:51,530 measures the pattern of firing of individual neurons. 136 00:06:51,530 --> 00:06:52,740 Can't do it in humans. 137 00:06:52,740 --> 00:06:55,198 So we've got monkeys that are trained in a delayed response 138 00:06:55,198 --> 00:06:56,000 task. 139 00:06:56,000 --> 00:06:59,900 And in this one, they were looking at a screen. 140 00:06:59,900 --> 00:07:03,395 And the screen had a fixation point in the center. 141 00:07:03,395 --> 00:07:05,270 And the monkeys looked at the fixation point. 142 00:07:05,270 --> 00:07:07,300 And they're trained that there will be-- 143 00:07:07,300 --> 00:07:10,190 a cue will appear and disappear. 144 00:07:10,190 --> 00:07:12,950 And they have to keep looking at the fixation point. 145 00:07:12,950 --> 00:07:15,860 And then shortly after a delay again, probably 146 00:07:15,860 --> 00:07:17,870 between five and 30 seconds is pretty 147 00:07:17,870 --> 00:07:20,360 typical for working memory tasks, 148 00:07:20,360 --> 00:07:23,780 then they get a go signal. 149 00:07:23,780 --> 00:07:26,330 And they move their eyes to where the cue was. 150 00:07:26,330 --> 00:07:29,814 So they have to hold the location of that cue in memory 151 00:07:29,814 --> 00:07:30,980 for the period of the delay. 152 00:07:34,150 --> 00:07:39,070 And what has been found is that for a task like this, 153 00:07:39,070 --> 00:07:41,040 the area that seems to be most involved 154 00:07:41,040 --> 00:07:44,220 is this dorsolateral prefrontal cortex. 155 00:07:44,220 --> 00:07:46,010 So, let's see, here's the central sulcus. 156 00:07:46,010 --> 00:07:50,330 So this is kind of, all of this is the frontal lobe, 157 00:07:50,330 --> 00:07:53,790 prefrontal can store everything from like here, forward. 158 00:07:53,790 --> 00:07:58,130 Dorsal on the top, lateral on the sides, 159 00:07:58,130 --> 00:08:00,710 dorsolateral prefrontal cortex. 160 00:08:00,710 --> 00:08:03,140 And they found in there, both cells that were active 161 00:08:03,140 --> 00:08:06,120 while the cue was being presented, 162 00:08:06,120 --> 00:08:10,382 which isn't when working memory is really being engaged. 163 00:08:10,382 --> 00:08:11,840 But they also found cells that were 164 00:08:11,840 --> 00:08:15,230 most active during that delay period, 165 00:08:15,230 --> 00:08:16,730 that once the cue went away and they 166 00:08:16,730 --> 00:08:19,677 were trying to maintain that-- 167 00:08:19,677 --> 00:08:21,260 and when the monkey was trying to hang 168 00:08:21,260 --> 00:08:23,750 onto that information about where the cue had been, 169 00:08:23,750 --> 00:08:28,100 this group of cells fired sustainedly. 170 00:08:28,100 --> 00:08:31,010 Just, I'm not sure that's a word. 171 00:08:31,010 --> 00:08:32,510 So in particular in this area, there 172 00:08:32,510 --> 00:08:36,110 are neurons that fire during the delay only when the cue is 173 00:08:36,110 --> 00:08:37,440 to a particular location. 174 00:08:37,440 --> 00:08:41,210 So we've got neurons that are preferring one cue 175 00:08:41,210 --> 00:08:44,990 location or another, that fire then when that particular cue 176 00:08:44,990 --> 00:08:48,110 location is the information that's being held in memory. 177 00:08:59,822 --> 00:09:02,907 All right. 178 00:09:02,907 --> 00:09:04,240 I'll distribute those among you. 179 00:09:04,240 --> 00:09:09,530 This is one set of the data from this particular study. 180 00:09:09,530 --> 00:09:11,650 Take those and pass them. 181 00:09:11,650 --> 00:09:14,980 So, you can see there's a diagram at the top of what 182 00:09:14,980 --> 00:09:15,950 the task was. 183 00:09:15,950 --> 00:09:19,090 But what I want to look at is along the bottom here, 184 00:09:19,090 --> 00:09:22,870 is a graph of the response of one particular neuron 185 00:09:22,870 --> 00:09:25,190 to cues in different locations. 186 00:09:25,190 --> 00:09:29,170 So, as arrayed around the fixation point. 187 00:09:29,170 --> 00:09:31,840 So, on the horizontal axis of each of these 188 00:09:31,840 --> 00:09:34,750 is time with the first section there 189 00:09:34,750 --> 00:09:37,780 being where the cue is given. 190 00:09:37,780 --> 00:09:41,320 And then, so this kind of area is where 191 00:09:41,320 --> 00:09:42,710 the cue is being presented. 192 00:09:42,710 --> 00:09:44,960 And then this is the delay period. 193 00:09:44,960 --> 00:09:47,290 And you can see that this particular neuron doesn't 194 00:09:47,290 --> 00:09:52,820 change its firing rate much for most of these cases. 195 00:09:52,820 --> 00:09:56,020 But it strongly prefers a cue in one direction. 196 00:09:56,020 --> 00:09:59,222 Which direction? 197 00:09:59,222 --> 00:10:00,680 Towards the upper left here, right? 198 00:10:00,680 --> 00:10:04,420 You can see that in this case, once this particular location 199 00:10:04,420 --> 00:10:08,480 is being held in working memory after the cue has been removed, 200 00:10:08,480 --> 00:10:09,490 this neuron goes nuts. 201 00:10:09,490 --> 00:10:11,650 And it just fires and fires and fires and fires. 202 00:10:11,650 --> 00:10:14,140 Its firing rate is much higher than these sort 203 00:10:14,140 --> 00:10:16,210 of baseline rates. 204 00:10:16,210 --> 00:10:18,910 You can see it gets activated a little bit for the straight up 205 00:10:18,910 --> 00:10:21,100 or straight to the left conditions. 206 00:10:21,100 --> 00:10:23,380 It's slightly higher than its baseline 207 00:10:23,380 --> 00:10:25,185 but it's definitely got a strong preference 208 00:10:25,185 --> 00:10:26,060 for one or the other. 209 00:10:26,060 --> 00:10:27,518 So this is a neuron that's involved 210 00:10:27,518 --> 00:10:31,990 in maintaining information about the location of the cue 211 00:10:31,990 --> 00:10:35,526 after the cue has disappeared. 212 00:10:35,526 --> 00:10:36,400 Does this make sense? 213 00:10:41,180 --> 00:10:42,136 Cool. 214 00:10:42,136 --> 00:10:45,735 AUDIENCE: Why is it top left? 215 00:10:45,735 --> 00:10:46,610 ABBY NOYCE: That one? 216 00:10:46,610 --> 00:10:49,300 This neuron likes the top left location. 217 00:10:49,300 --> 00:10:51,920 There's probably other neurons that like all of the others. 218 00:10:51,920 --> 00:10:56,080 So this is just data from one individual cell, 219 00:10:56,080 --> 00:10:59,556 showing that the cells have directional preferences. 220 00:10:59,556 --> 00:11:01,930 It's like, think of it as being like the edge orientation 221 00:11:01,930 --> 00:11:03,304 neurons we talked about last week 222 00:11:03,304 --> 00:11:05,470 that like lines at particular angles. 223 00:11:05,470 --> 00:11:07,960 It's a matter of what's being, of how 224 00:11:07,960 --> 00:11:10,090 they're wired up to get input from cells 225 00:11:10,090 --> 00:11:10,990 earlier in the chain. 226 00:11:14,170 --> 00:11:19,690 All right, so that's one kind of classic experiment 227 00:11:19,690 --> 00:11:20,320 they looked at. 228 00:11:20,320 --> 00:11:22,361 They took the same experiment, and they said, OK, 229 00:11:22,361 --> 00:11:24,910 but that's just correlational evidence, you know? 230 00:11:24,910 --> 00:11:30,100 We know that when working memory is happening, then these cells 231 00:11:30,100 --> 00:11:31,420 fire. 232 00:11:31,420 --> 00:11:34,180 But that correlation is not causation. 233 00:11:34,180 --> 00:11:35,680 What's some more evidence that would 234 00:11:35,680 --> 00:11:38,200 help make this look causal that would help draw these two 235 00:11:38,200 --> 00:11:39,670 things together? 236 00:11:39,670 --> 00:11:43,180 And they looked at what happens when the monkey was wrong, 237 00:11:43,180 --> 00:11:45,490 when the monkey moved its eyes to the wrong location, 238 00:11:45,490 --> 00:11:47,920 not to the cued location. 239 00:11:47,920 --> 00:11:51,010 Was the activity of these cells that we believe to be 240 00:11:51,010 --> 00:11:52,960 working the same? 241 00:11:52,960 --> 00:11:53,980 Nope. 242 00:11:53,980 --> 00:11:58,870 So, in cases where the monkey did not do the task correctly, 243 00:11:58,870 --> 00:12:03,190 these working memory kind of maintenance activity cells 244 00:12:03,190 --> 00:12:05,380 were actually less active. 245 00:12:05,380 --> 00:12:09,070 Or they were active for a while and it dropped off. 246 00:12:09,070 --> 00:12:14,170 And this might just be that the memory is decaying. 247 00:12:14,170 --> 00:12:16,360 It might be that the monkey just stops 248 00:12:16,360 --> 00:12:19,570 paying attention and all of the neurons in its brain kind 249 00:12:19,570 --> 00:12:21,820 slow down a little bit and stop firing so fast. 250 00:12:25,000 --> 00:12:27,160 We're still looking at correlational and not causal 251 00:12:27,160 --> 00:12:28,143 evidence. 252 00:12:31,250 --> 00:12:33,680 So going one step further looking 253 00:12:33,680 --> 00:12:37,970 for a more direct connection, people 254 00:12:37,970 --> 00:12:40,070 looking at the same parts of the brain 255 00:12:40,070 --> 00:12:42,380 did some lesioning studies, so small lesions 256 00:12:42,380 --> 00:12:47,037 to particular chunks of that dorsolateral prefrontal cortex. 257 00:12:47,037 --> 00:12:48,620 That's the area where these cells that 258 00:12:48,620 --> 00:12:52,760 are active during the delay period seem to be located. 259 00:12:52,760 --> 00:12:56,975 And they found that if they did small lesions-- 260 00:12:56,975 --> 00:12:58,850 big lesions, working memory just dropped off. 261 00:12:58,850 --> 00:13:00,830 But if they did small lesions, you 262 00:13:00,830 --> 00:13:04,667 got monkeys who were OK at this task at most cue locations 263 00:13:04,667 --> 00:13:07,250 and then would have, like, one or two right next to each other 264 00:13:07,250 --> 00:13:09,170 where they were very poor. 265 00:13:09,170 --> 00:13:12,825 And the explanation that these researchers came up with 266 00:13:12,825 --> 00:13:15,200 was that we've managed to remove the neurons that respond 267 00:13:15,200 --> 00:13:18,140 to this particular location, while leaving intact 268 00:13:18,140 --> 00:13:21,290 neurons that respond to other ones. 269 00:13:21,290 --> 00:13:24,570 Somebody did the same thing with cooling neurons. 270 00:13:24,570 --> 00:13:26,660 So instead of actually lesioning them, 271 00:13:26,660 --> 00:13:30,500 you slide a little piece of metal into their brain 272 00:13:30,500 --> 00:13:32,000 and then you put the other end of it 273 00:13:32,000 --> 00:13:34,370 in, like, liquid nitrogen. And it, therefore, 274 00:13:34,370 --> 00:13:36,680 brings the temperature of the neurons right next to it 275 00:13:36,680 --> 00:13:39,650 down enough that they don't fire. 276 00:13:39,650 --> 00:13:42,360 But it doesn't do long term damage. 277 00:13:42,360 --> 00:13:47,690 So lesioning studies have a confound 278 00:13:47,690 --> 00:13:50,085 where animals that have le-- 279 00:13:50,085 --> 00:13:51,710 we know that when the brain is damaged, 280 00:13:51,710 --> 00:13:54,080 you'll see long term structural changes. 281 00:13:54,080 --> 00:13:59,490 As information and processes get moved to other areas, 282 00:13:59,490 --> 00:14:01,670 other areas take responsibility for what 283 00:14:01,670 --> 00:14:03,530 was originally lesioned. 284 00:14:03,530 --> 00:14:06,530 Whereas if you're just doing these kind of temporary 285 00:14:06,530 --> 00:14:08,090 knocking out one neural population, 286 00:14:08,090 --> 00:14:09,980 then you don't get that confound, because it's just 287 00:14:09,980 --> 00:14:10,813 a short term change. 288 00:14:14,370 --> 00:14:17,010 Because with lesioning, you do surgery on your animals 289 00:14:17,010 --> 00:14:19,509 and then you've got to let them recover for a couple of days 290 00:14:19,509 --> 00:14:22,530 before you can drag them out and make them work for their food 291 00:14:22,530 --> 00:14:25,124 again. 292 00:14:25,124 --> 00:14:27,540 And so that's long enough for this sort of rewiring around 293 00:14:27,540 --> 00:14:29,380 the damaged part of the brain to start happening. 294 00:14:29,380 --> 00:14:30,740 So they wanted to correct for that 295 00:14:30,740 --> 00:14:32,740 and make sure that wasn't what they were seeing. 296 00:14:35,180 --> 00:14:38,850 Again, lots of data all kind of converging on the same point. 297 00:14:48,670 --> 00:14:51,080 All right, so this is monkeys. 298 00:14:51,080 --> 00:14:52,670 That's cool and all, but really we 299 00:14:52,670 --> 00:14:54,689 want to know about working memory in humans. 300 00:14:54,689 --> 00:14:56,480 We can't sink electrodes into their brains. 301 00:14:56,480 --> 00:14:59,080 We certainly can't cut chunks out of their brains. 302 00:14:59,080 --> 00:15:02,590 But there's some evidence from humans nonetheless. 303 00:15:02,590 --> 00:15:04,190 FMRI has been used for this. 304 00:15:04,190 --> 00:15:07,130 FMRI shows activity, as you might expect, 305 00:15:07,130 --> 00:15:10,040 in the dorsolateral prefrontal cortex. 306 00:15:10,040 --> 00:15:11,870 Also, for these kinds of spatial tasks 307 00:15:11,870 --> 00:15:14,030 where you are cued to a location and have 308 00:15:14,030 --> 00:15:17,970 to hold that location in working memory for a delay period, 309 00:15:17,970 --> 00:15:21,440 then you'll see activity in the dorsal parietal cortex, 310 00:15:21,440 --> 00:15:24,770 kind of up here on the sides. 311 00:15:24,770 --> 00:15:28,070 And again, just like in the single cell recordings, 312 00:15:28,070 --> 00:15:30,020 you'll see this activity increase 313 00:15:30,020 --> 00:15:32,090 sustained throughout the entire delay period. 314 00:15:34,555 --> 00:15:35,930 So right now, it looks like we've 315 00:15:35,930 --> 00:15:38,360 got some pretty strong evidence saying 316 00:15:38,360 --> 00:15:41,930 that when you're working memory is working, when it's actively 317 00:15:41,930 --> 00:15:45,260 maintaining something, then you get an increase in firing 318 00:15:45,260 --> 00:15:49,090 among a set of neurons for as long as you're delaying it. 319 00:15:49,090 --> 00:15:51,590 So, as long as you're holding it throughout the delay period 320 00:15:51,590 --> 00:15:53,520 between you when you're given the item. 321 00:15:53,520 --> 00:15:58,211 And when you have to do something with it 322 00:15:58,211 --> 00:15:59,210 and can stop holding it. 323 00:16:14,600 --> 00:16:18,050 All right, so far we've been talking about paradigms where 324 00:16:18,050 --> 00:16:20,810 your people or animals are holding a single item 325 00:16:20,810 --> 00:16:23,320 in working memory at one time. 326 00:16:23,320 --> 00:16:26,030 And to get down into the nitty-gritty of how exactly 327 00:16:26,030 --> 00:16:29,391 an item is in working memory, what that means, 328 00:16:29,391 --> 00:16:30,890 we've got to understand what happens 329 00:16:30,890 --> 00:16:33,380 when you're trying to hold more than one item at a time. 330 00:16:33,380 --> 00:16:35,510 Clearly, certainly, from a subjective perspective 331 00:16:35,510 --> 00:16:37,759 holding two things in memory is different than holding 332 00:16:37,759 --> 00:16:38,865 one thing in memory. 333 00:16:38,865 --> 00:16:41,000 You've got two things in there. 334 00:16:41,000 --> 00:16:44,850 So what sorts of neurological changes would we expect to see? 335 00:16:44,850 --> 00:16:46,820 So there's two kind of possibilities 336 00:16:46,820 --> 00:16:49,610 when you're holding more items in working memory. 337 00:16:49,610 --> 00:16:52,850 You might see nothing change at all, but that seems unlikely. 338 00:16:52,850 --> 00:16:54,710 You might see that more parts of your brain 339 00:16:54,710 --> 00:16:58,190 get involved, get pulled in to help with this working memory 340 00:16:58,190 --> 00:16:59,041 task. 341 00:16:59,041 --> 00:17:01,040 Or you might see that pretty much the same parts 342 00:17:01,040 --> 00:17:03,830 of the brain are involved, but that the amount by which 343 00:17:03,830 --> 00:17:08,359 activity increases becomes more for each extra activity, 344 00:17:08,359 --> 00:17:12,290 each extra item that's being held in memory. 345 00:17:12,290 --> 00:17:14,930 And the FMRI data seems to show that both of these things 346 00:17:14,930 --> 00:17:17,900 happen, both as you increase the set of items 347 00:17:17,900 --> 00:17:21,410 that you have to remember then more brain regions are 348 00:17:21,410 --> 00:17:26,400 recruited and pulled in to help keep that maintenance, 349 00:17:26,400 --> 00:17:28,700 keep that representation. 350 00:17:28,700 --> 00:17:32,600 And you also see activity in the regions 351 00:17:32,600 --> 00:17:33,784 that are involved increase. 352 00:17:33,784 --> 00:17:35,450 They actually become more active as they 353 00:17:35,450 --> 00:17:36,658 are representing more things. 354 00:17:43,981 --> 00:17:45,480 One of the things that's interesting 355 00:17:45,480 --> 00:17:50,460 is we have this idea that different parts of the brain-- 356 00:17:50,460 --> 00:17:53,820 we have this idea that working memory has different storage 357 00:17:53,820 --> 00:17:55,620 spaces for different kinds of information 358 00:17:55,620 --> 00:17:57,360 that's kind of key to this Baddeley model 359 00:17:57,360 --> 00:18:01,060 that we're working with. 360 00:18:01,060 --> 00:18:02,710 And if the Baddeley model is correct, 361 00:18:02,710 --> 00:18:03,840 one of the things you'd expect to see 362 00:18:03,840 --> 00:18:05,890 is that for different kinds of information, 363 00:18:05,890 --> 00:18:08,056 you'd see different brain regions activated when you 364 00:18:08,056 --> 00:18:09,540 hold them in working memory. 365 00:18:09,540 --> 00:18:11,640 And that seems to be more, that seems 366 00:18:11,640 --> 00:18:13,997 to be at least somewhat true. 367 00:18:13,997 --> 00:18:16,080 For example, like we said, for these spatial tasks 368 00:18:16,080 --> 00:18:18,720 you'll see the parietal areas in the parietal lobe getting 369 00:18:18,720 --> 00:18:21,500 recruited to help maintain that information. 370 00:18:21,500 --> 00:18:23,190 For, like, object recognition tasks, 371 00:18:23,190 --> 00:18:26,130 you'll see areas in the temporal lobe getting dragged in. 372 00:18:26,130 --> 00:18:28,170 So there's definitely at least some amount 373 00:18:28,170 --> 00:18:30,780 of activity in different brain areas 374 00:18:30,780 --> 00:18:36,150 for remembering different kinds of material in working memory. 375 00:18:36,150 --> 00:18:39,430 All right, so what happens if you're 376 00:18:39,430 --> 00:18:43,130 holding onto more things? 377 00:18:43,130 --> 00:18:45,010 One of the kind of classic challenging 378 00:18:45,010 --> 00:18:47,830 working memory tasks that we've talked about before 379 00:18:47,830 --> 00:18:50,070 is this N-back task. 380 00:18:50,070 --> 00:18:51,670 There's a stimulus currently being 381 00:18:51,670 --> 00:18:54,760 presented to match the one that's one item back or two 382 00:18:54,760 --> 00:18:58,314 items back or five items back. 383 00:18:58,314 --> 00:19:00,730 This is, remember, the working memory task with that paper 384 00:19:00,730 --> 00:19:04,600 we read for the very first week of class used. 385 00:19:04,600 --> 00:19:07,450 So, if you're doing a two-back task, 386 00:19:07,450 --> 00:19:09,430 then the first n gets presented. 387 00:19:09,430 --> 00:19:11,680 The answer is no, there's nothing two back. 388 00:19:11,680 --> 00:19:14,740 Get secondly for the first l. 389 00:19:14,740 --> 00:19:17,800 Actually, I think for two backs, are these all, no's? 390 00:19:17,800 --> 00:19:19,120 Yeah, it's all no for two-back. 391 00:19:19,120 --> 00:19:21,190 But for three-backs, for example, 392 00:19:21,190 --> 00:19:23,650 that second k would be a yes, it matches the k 393 00:19:23,650 --> 00:19:25,640 that is three back from it. 394 00:19:25,640 --> 00:19:27,602 So, when you're just doing a one-back task, 395 00:19:27,602 --> 00:19:29,560 you've only got to ever remember the thing that 396 00:19:29,560 --> 00:19:32,460 was shown to you just before the thing you're now showing. 397 00:19:32,460 --> 00:19:36,322 And as n goes up as you do a two-back or a three-back task, 398 00:19:36,322 --> 00:19:37,030 this gets harder. 399 00:19:37,030 --> 00:19:38,363 It requires more working memory. 400 00:19:38,363 --> 00:19:43,294 You've got to hold more of the previous stimuli in mind. 401 00:19:43,294 --> 00:19:44,710 People like this task because they 402 00:19:44,710 --> 00:19:46,510 can change the memory load. 403 00:19:46,510 --> 00:19:50,479 They can show two subjects the exact same set of stimuli 404 00:19:50,479 --> 00:19:53,020 and have one of them do it as a one-back task and one of them 405 00:19:53,020 --> 00:19:55,000 do it as a three-back task. 406 00:19:55,000 --> 00:19:56,500 And the only thing that's changing 407 00:19:56,500 --> 00:19:58,030 is the working memory load. 408 00:19:58,030 --> 00:20:01,240 They're getting the same stimuli in the same order, 409 00:20:01,240 --> 00:20:02,740 so it takes out some confounds that 410 00:20:02,740 --> 00:20:05,831 can be based on just basic perceptual differences. 411 00:20:09,520 --> 00:20:12,710 So, FMRI says that you get increased activity 412 00:20:12,710 --> 00:20:16,530 in the lateral prefrontal cortex and in the parietal cortex. 413 00:20:16,530 --> 00:20:19,500 And in both of these cases, it goes up linearly with n, 414 00:20:19,500 --> 00:20:23,880 so each time that you add another item to n. 415 00:20:23,880 --> 00:20:25,860 So if you go, the difference between a one-back 416 00:20:25,860 --> 00:20:30,051 and a two-back task is a certain amount of activity increase. 417 00:20:30,051 --> 00:20:31,800 And then the difference between a two-back 418 00:20:31,800 --> 00:20:33,900 and a three-back task is, again, the same amount 419 00:20:33,900 --> 00:20:35,070 of activity increase. 420 00:20:35,070 --> 00:20:40,200 It's a linear increase in the lateral prefrontal cortex 421 00:20:40,200 --> 00:20:43,230 and in the parietal cortex. 422 00:20:43,230 --> 00:20:47,280 But one of the problems with this particular task model 423 00:20:47,280 --> 00:20:51,450 is that it's not just requiring you to hold items in memory. 424 00:20:51,450 --> 00:20:53,340 It's requiring you to actually hold 425 00:20:53,340 --> 00:20:54,790 a lot of information about them. 426 00:20:54,790 --> 00:20:57,100 And then keep changing that information. 427 00:20:57,100 --> 00:21:01,230 So, if you're getting this, you know, first of all, 428 00:21:01,230 --> 00:21:02,970 this n is the item you're looking at. 429 00:21:02,970 --> 00:21:05,070 And then the n is one back. 430 00:21:05,070 --> 00:21:08,040 And then the l is one back and the n is now two back. 431 00:21:08,040 --> 00:21:10,670 And then the l is one back and the l is two back 432 00:21:10,670 --> 00:21:12,320 and the n is now three back. 433 00:21:12,320 --> 00:21:14,070 And you've got to kind of keep relabelling 434 00:21:14,070 --> 00:21:15,611 each of these thing that's in memory. 435 00:21:15,611 --> 00:21:16,480 They're not static. 436 00:21:16,480 --> 00:21:18,040 They're changing. 437 00:21:18,040 --> 00:21:20,440 And that's really something that's an executive task. 438 00:21:20,440 --> 00:21:21,523 It's not just an item set. 439 00:21:21,523 --> 00:21:25,570 It's the actual working with the information. 440 00:21:25,570 --> 00:21:30,840 So there's a confound here between that executive task, 441 00:21:30,840 --> 00:21:34,350 where you're actively changing the information in working 442 00:21:34,350 --> 00:21:36,900 memory, and the maintenance process itself, 443 00:21:36,900 --> 00:21:39,493 where you're merely maintaining these representations. 444 00:21:44,330 --> 00:21:47,230 So if you wanted to try to separate those, 445 00:21:47,230 --> 00:21:50,050 see what comes from just maintaining information 446 00:21:50,050 --> 00:21:54,060 in the brain and what comes from actively working on it, 447 00:21:54,060 --> 00:21:56,680 you'd probably want to find a task that doesn't have all that 448 00:21:56,680 --> 00:21:59,620 increased executive processes. 449 00:21:59,620 --> 00:22:03,100 Straight up ordinary item recognition tasks, 450 00:22:03,100 --> 00:22:07,172 so I show you one item or two items or three items. 451 00:22:07,172 --> 00:22:08,130 Look, it's a butterfly! 452 00:22:08,130 --> 00:22:09,670 Take it away again. 453 00:22:09,670 --> 00:22:11,230 Wait 10 seconds. 454 00:22:11,230 --> 00:22:14,002 Show you the item again, look, it's a cat. 455 00:22:14,002 --> 00:22:15,460 Is this the item I just showed you? 456 00:22:18,060 --> 00:22:19,820 Good. 457 00:22:19,820 --> 00:22:22,190 So this is a probe item. 458 00:22:22,190 --> 00:22:23,780 It's like a test item here. 459 00:22:23,780 --> 00:22:26,334 AUDIENCE: The butterfly is inside the cat. 460 00:22:26,334 --> 00:22:28,250 ABBY NOYCE: Possibly, but it is not the object 461 00:22:28,250 --> 00:22:30,270 I just showed you, nonetheless. 462 00:22:30,270 --> 00:22:32,260 So this is a simpler task than the n-back task. 463 00:22:32,260 --> 00:22:35,420 It's less demanding. 464 00:22:35,420 --> 00:22:37,810 The executive requirements don't change with set size. 465 00:22:37,810 --> 00:22:39,950 You're merely maintaining a set of objects here. 466 00:22:39,950 --> 00:22:43,610 You're not working with them the way you are in the n-back task. 467 00:22:43,610 --> 00:22:46,310 And you find pretty much the same FMRI data, 468 00:22:46,310 --> 00:22:49,700 which is that you get these linear increases in activity 469 00:22:49,700 --> 00:22:52,232 in prefrontal and parietal areas. 470 00:22:52,232 --> 00:22:53,690 So it looks like the change that we 471 00:22:53,690 --> 00:22:55,700 see in the n-back task is probably coming 472 00:22:55,700 --> 00:23:01,670 from the increased maintenance requirements of the larger n, 473 00:23:01,670 --> 00:23:03,806 looking at a three back versus the two back, 474 00:23:03,806 --> 00:23:05,180 and not just because you're doing 475 00:23:05,180 --> 00:23:07,680 all that extra work on these things to re-categorized them. 476 00:23:29,000 --> 00:23:36,780 All right, so we know that holding items in working memory 477 00:23:36,780 --> 00:23:40,350 requires that this set of neurons 478 00:23:40,350 --> 00:23:43,480 keep firing for a sustained period of time. 479 00:23:43,480 --> 00:23:45,990 So how does that happen? 480 00:23:45,990 --> 00:23:48,240 Is there, like, some kind of command center that 481 00:23:48,240 --> 00:23:51,030 says, OK, I'm going to tell this set of neurons 482 00:23:51,030 --> 00:23:57,040 to fire in order to keep this piece of information in mind? 483 00:23:57,040 --> 00:23:58,946 Every time, as far as anyone can tell, 484 00:23:58,946 --> 00:24:01,570 as far as I can tell, every time in neuroscience you're tempted 485 00:24:01,570 --> 00:24:03,736 to believe that there is some kind of little command 486 00:24:03,736 --> 00:24:06,970 center that's making a conscious decision of some sort, 487 00:24:06,970 --> 00:24:09,760 this is a dangerous idea that you want to stay away from. 488 00:24:09,760 --> 00:24:12,070 A lot more often what's happening 489 00:24:12,070 --> 00:24:15,040 is that a network of neurons works together in the way 490 00:24:15,040 --> 00:24:16,060 that they are wired up. 491 00:24:16,060 --> 00:24:19,630 They kind of all encourage each other or discourage each other 492 00:24:19,630 --> 00:24:23,020 and settle on a particular pattern. 493 00:24:23,020 --> 00:24:26,750 And that seems to be true of how this works. 494 00:24:26,750 --> 00:24:29,470 So, Donald Hebb, who was one of the big guys 495 00:24:29,470 --> 00:24:32,529 in the '40s and '50s in neuroscience-- 496 00:24:32,529 --> 00:24:34,570 we'll talk about him a little bit more next week, 497 00:24:34,570 --> 00:24:36,160 another idea of his. 498 00:24:36,160 --> 00:24:41,200 He said, OK, so we know that the idea of having 499 00:24:41,200 --> 00:24:43,480 this neat and tidy command center 500 00:24:43,480 --> 00:24:46,000 that's directing everything the brain does is a bad idea. 501 00:24:46,000 --> 00:24:48,730 It doesn't seem to be how it works. 502 00:24:48,730 --> 00:24:51,796 Can we just do this within a group of neurons? 503 00:24:51,796 --> 00:24:53,170 Can the neurons that are involved 504 00:24:53,170 --> 00:24:56,770 in maintaining a pattern be connected to each other, such 505 00:24:56,770 --> 00:24:58,600 that they maintain the pattern? 506 00:24:58,600 --> 00:25:01,450 That each one by firing excites the other neurons 507 00:25:01,450 --> 00:25:03,850 in the pattern and so, as they all fire together, 508 00:25:03,850 --> 00:25:05,740 it just keeps the firing rate up? 509 00:25:05,740 --> 00:25:07,930 Can this sustained activity be something that's 510 00:25:07,930 --> 00:25:14,050 just kind of self-reinforcing? 511 00:25:14,050 --> 00:25:15,327 So each neuron fires. 512 00:25:15,327 --> 00:25:17,410 It stimulates other neurons and they stimulate it. 513 00:25:17,410 --> 00:25:20,650 And the whole thing keeps itself going simply 514 00:25:20,650 --> 00:25:22,750 by the connections between the neurons. 515 00:25:22,750 --> 00:25:27,350 So this was his theory in the late '40s. 516 00:25:27,350 --> 00:25:30,082 And both our models of how neurons work 517 00:25:30,082 --> 00:25:31,540 and our knowledge about how neurons 518 00:25:31,540 --> 00:25:33,615 are connected to each other weren't as good 519 00:25:33,615 --> 00:25:34,490 then as they are now. 520 00:25:34,490 --> 00:25:37,490 So when he first came out with this, there was a lot of, 521 00:25:37,490 --> 00:25:38,740 we don't think this will work. 522 00:25:38,740 --> 00:25:40,690 Will neurons really do this? 523 00:25:40,690 --> 00:25:42,850 Since then, people have started looking at it a bit 524 00:25:42,850 --> 00:25:45,010 more carefully. 525 00:25:45,010 --> 00:25:48,040 And it turns out that if you put together 526 00:25:48,040 --> 00:25:50,500 a model of a neural network, so a computer program 527 00:25:50,500 --> 00:25:53,170 where you've got simulated neurons that are connected 528 00:25:53,170 --> 00:25:56,500 to each other using information from how we know 529 00:25:56,500 --> 00:25:59,200 neurons interact with each other, how they work 530 00:25:59,200 --> 00:26:02,736 physiologically, what sorts of inputs they are likely to have, 531 00:26:02,736 --> 00:26:05,110 can you build a network that will self-sustain like this? 532 00:26:05,110 --> 00:26:09,110 Where once you get it going, it will just hold itself up there? 533 00:26:09,110 --> 00:26:10,600 And the answer seems to be, yes. 534 00:26:10,600 --> 00:26:12,100 We can do that. 535 00:26:12,100 --> 00:26:14,890 You get these patterns of reverberating activity 536 00:26:14,890 --> 00:26:17,680 where each neuron excites the other neurons 537 00:26:17,680 --> 00:26:20,410 in this cluster that are maintaining whatever 538 00:26:20,410 --> 00:26:21,700 information. 539 00:26:21,700 --> 00:26:24,977 And so if you look at individual neurons in one of these models, 540 00:26:24,977 --> 00:26:27,310 you get firing rates that are very similar to what you'd 541 00:26:27,310 --> 00:26:31,090 see in a real life electrophysiology 542 00:26:31,090 --> 00:26:33,760 measure of the behavior of a particular neuron that's 543 00:26:33,760 --> 00:26:34,960 involved in working memory. 544 00:26:41,250 --> 00:26:43,440 All right, want to see one? 545 00:26:43,440 --> 00:26:49,630 So this is a model made by a guy named Wang at Yale. 546 00:26:49,630 --> 00:26:51,880 So you've got, that's time moving along the top here. 547 00:26:51,880 --> 00:26:55,030 This is a set of neurons, simulated neurons, 548 00:26:55,030 --> 00:26:56,043 that are firing. 549 00:26:56,043 --> 00:26:56,543 Ready? 550 00:26:56,543 --> 00:26:57,751 So there's about to be a cue. 551 00:26:57,751 --> 00:26:58,930 So the cue is to the left. 552 00:26:58,930 --> 00:27:00,680 And you can see this population of neurons 553 00:27:00,680 --> 00:27:02,530 that responds to that area. 554 00:27:02,530 --> 00:27:04,570 And the cue has gone, but the firing 555 00:27:04,570 --> 00:27:08,290 among the population of neurons there is holding itself up. 556 00:27:08,290 --> 00:27:13,640 You can see that the spike just maintains over time. 557 00:27:13,640 --> 00:27:25,425 It just keeps going, and going, and going. 558 00:27:25,425 --> 00:27:26,800 AUDIENCE: This guy's name's Wang? 559 00:27:26,800 --> 00:27:28,300 ABBY NOYCE: So there's. 560 00:27:28,300 --> 00:27:31,750 Yep, wanglab.yale.edu, I believe. 561 00:27:36,340 --> 00:27:39,982 So he's working with one of these models of how 562 00:27:39,982 --> 00:27:41,440 you can build a neural network that 563 00:27:41,440 --> 00:27:44,420 will self-sustain like that. 564 00:27:44,420 --> 00:27:47,340 You want to see it again? 565 00:27:47,340 --> 00:27:49,169 You want to see it again. 566 00:27:49,169 --> 00:27:51,460 So, again you'll notice there's a baseline firing rate. 567 00:27:51,460 --> 00:27:53,710 None of the neurons are really firing a whole lot more 568 00:27:53,710 --> 00:27:56,360 than the other. 569 00:27:56,360 --> 00:27:58,452 And then when that cue is presented, 570 00:27:58,452 --> 00:28:00,160 these are populations of neurons that are 571 00:28:00,160 --> 00:28:02,240 tuned to different locations. 572 00:28:02,240 --> 00:28:04,660 So all the ones that are tuned to that location 573 00:28:04,660 --> 00:28:08,440 there, 180 degrees from kind of the default point, 574 00:28:08,440 --> 00:28:10,600 are activated when the cue appears 575 00:28:10,600 --> 00:28:13,510 and then they continue to be activated after the cue 576 00:28:13,510 --> 00:28:14,010 is gone. 577 00:28:14,010 --> 00:28:18,520 So they're not only excited by the actual perceptual input 578 00:28:18,520 --> 00:28:20,500 from that cue. 579 00:28:20,500 --> 00:28:23,999 They're also responding to each other, 580 00:28:23,999 --> 00:28:26,290 they're exciting each other and keeping the firing rate 581 00:28:26,290 --> 00:28:27,640 up among the circuit. 582 00:28:27,640 --> 00:28:28,826 AUDIENCE: What does d mean? 583 00:28:28,826 --> 00:28:29,950 ABBY NOYCE: D is for delay. 584 00:28:29,950 --> 00:28:32,624 So this is a delay period. 585 00:28:32,624 --> 00:28:34,540 It's like in your classic working memory task, 586 00:28:34,540 --> 00:28:36,289 you'd have a cue and then a delay and then 587 00:28:36,289 --> 00:28:39,292 a test of some kind. 588 00:28:39,292 --> 00:28:40,750 And this one, we don't care so much 589 00:28:40,750 --> 00:28:41,950 about what the test is per se. 590 00:28:41,950 --> 00:28:43,630 We're interested in how the neurons behave 591 00:28:43,630 --> 00:28:45,130 during these different time periods. 592 00:28:49,050 --> 00:28:51,920 And it's a cool video. 593 00:28:51,920 --> 00:28:54,880 Questions? 594 00:28:54,880 --> 00:28:55,500 Cool. 595 00:28:55,500 --> 00:28:57,877 So that's an example of the kind of model, 596 00:28:57,877 --> 00:28:59,460 of this kind of model showing that you 597 00:28:59,460 --> 00:29:02,160 can get a group of neurons that just keep 598 00:29:02,160 --> 00:29:05,428 a circuit like that maintained. 599 00:29:05,428 --> 00:29:07,428 AUDIENCE: So, does this mean that we hold things 600 00:29:07,428 --> 00:29:10,094 in our memory because our cells do it for us? 601 00:29:10,094 --> 00:29:10,920 Or not because-- 602 00:29:10,920 --> 00:29:12,000 ABBY NOYCE: Everything you do in your brain, 603 00:29:12,000 --> 00:29:13,770 you do because your cells do it for you. 604 00:29:13,770 --> 00:29:16,717 You want to do things because your cells do it for you. 605 00:29:16,717 --> 00:29:18,300 This is one of those really dangerous, 606 00:29:18,300 --> 00:29:22,022 this is one those things that's really hard to think about. 607 00:29:22,022 --> 00:29:23,970 AUDIENCE: OK, in this one study, like, 608 00:29:23,970 --> 00:29:25,877 my dad showed me in the Wall Street Journal, 609 00:29:25,877 --> 00:29:29,820 where they tested people and went about decision-- 610 00:29:29,820 --> 00:29:31,634 it was about decision making. 611 00:29:31,634 --> 00:29:36,485 And they had to, like press a button with their left hand 612 00:29:36,485 --> 00:29:37,068 or right hand. 613 00:29:37,068 --> 00:29:40,530 And that was, like, basically, the basics [INAUDIBLE].. 614 00:29:40,530 --> 00:29:43,295 And, by the end of the study, the testers could actually, 615 00:29:43,295 --> 00:29:44,670 by looking at the neural patterns 616 00:29:44,670 --> 00:29:48,960 could predict which hand they would use to press 617 00:29:48,960 --> 00:29:52,766 the button, like, two seconds before it actually happened. 618 00:29:52,766 --> 00:29:54,890 ABBY NOYCE: Yeah, there's some classic work on this 619 00:29:54,890 --> 00:29:57,710 and I don't remember what the guy's name is who kind of did 620 00:29:57,710 --> 00:29:58,999 the pioneering stuff on this. 621 00:29:58,999 --> 00:30:01,040 But he set up a paradigm where he had people just 622 00:30:01,040 --> 00:30:04,590 sit there with a button. 623 00:30:04,590 --> 00:30:08,930 And I don't remember what measuring, what 624 00:30:08,930 --> 00:30:10,550 imaging they were using, but they were 625 00:30:10,550 --> 00:30:12,650 using an imaging technique. 626 00:30:12,650 --> 00:30:15,280 And he, and they were looking at a big clock, 627 00:30:15,280 --> 00:30:18,340 and he said, OK, at some point decide 628 00:30:18,340 --> 00:30:22,040 to press the button and notice what time the clock says, 629 00:30:22,040 --> 00:30:24,110 like with just the second hand going around it, 630 00:30:24,110 --> 00:30:27,400 when you decide, when you make that decision. 631 00:30:27,400 --> 00:30:30,500 And what he found out is that consistently, you'd 632 00:30:30,500 --> 00:30:34,335 see changes like in people's premotor cortex, which, 633 00:30:34,335 --> 00:30:35,960 like, coordinates your motor activities 634 00:30:35,960 --> 00:30:39,590 and stuff, before they consciously experienced 635 00:30:39,590 --> 00:30:41,780 having decided to move. 636 00:30:41,780 --> 00:30:44,450 So your brain starts doing, preparing 637 00:30:44,450 --> 00:30:46,640 for the action you're going to take before you feel 638 00:30:46,640 --> 00:30:49,700 like you've made that decision, which is weird 639 00:30:49,700 --> 00:30:52,160 and a little bit scary. 640 00:30:52,160 --> 00:30:54,530 But once you start getting down into the nitty gritty 641 00:30:54,530 --> 00:30:57,110 of neuroscience talking about free will 642 00:30:57,110 --> 00:30:58,820 becomes a little bit depressing. 643 00:30:58,820 --> 00:31:01,040 So, I mean-- 644 00:31:01,040 --> 00:31:03,041 AUDIENCE: What about, like, when I 645 00:31:03,041 --> 00:31:05,590 think there's a brand new study, did a study 646 00:31:05,590 --> 00:31:07,066 with epileptic patients. 647 00:31:07,066 --> 00:31:10,870 He'd stimulate one part of the brain and make one, 648 00:31:10,870 --> 00:31:13,169 he'd say OK, don't move your right arm, 649 00:31:13,169 --> 00:31:16,070 then he'd study what part of the brain that [INAUDIBLE].. 650 00:31:16,070 --> 00:31:18,500 ABBY NOYCE: I believe it. 651 00:31:18,500 --> 00:31:22,010 That's not as, I don't find, personally 652 00:31:22,010 --> 00:31:24,230 I don't find that as disconcerting as this idea 653 00:31:24,230 --> 00:31:27,680 that when I make a decision, my conscious experience of making 654 00:31:27,680 --> 00:31:30,950 a decision is preceded by a bunch of unconscious 655 00:31:30,950 --> 00:31:33,920 already made decision stuff. 656 00:31:33,920 --> 00:31:37,280 That's, that's a little bit weird. 657 00:31:37,280 --> 00:31:40,850 I mean I know that my actions are caused by firing of neurons 658 00:31:40,850 --> 00:31:46,260 in my brain, that doesn't throw me as much personally. 659 00:31:46,260 --> 00:31:48,260 AUDIENCE: I think that [INAUDIBLE] that, 660 00:31:48,260 --> 00:31:50,760 like [INAUDIBLE],, other times, patients would say, 661 00:31:50,760 --> 00:31:53,900 I didn't didn't move my arm, you did. 662 00:31:53,900 --> 00:31:55,710 ABBY NOYCE: Yeah, so your perception of, 663 00:31:55,710 --> 00:31:57,330 because you don't have that decision 664 00:31:57,330 --> 00:32:03,630 to move coupled with the actual motion. 665 00:32:03,630 --> 00:32:07,980 Anyway, back on topic. 666 00:32:07,980 --> 00:32:09,960 One of the things that was consistent about all 667 00:32:09,960 --> 00:32:13,050 of those different studies that we were talking about, 668 00:32:13,050 --> 00:32:17,040 about different parts of the brain involved 669 00:32:17,040 --> 00:32:18,810 in working memory and maintaining 670 00:32:18,810 --> 00:32:21,540 a representation, seemed to have, hey look, 671 00:32:21,540 --> 00:32:25,110 this region of the prefrontal cortex is heavily involved. 672 00:32:25,110 --> 00:32:28,430 Prefrontal cortex does all the cool stuff. 673 00:32:28,430 --> 00:32:30,930 Now prefrontal cortex seems to be heavily involved in what's 674 00:32:30,930 --> 00:32:33,550 called executive functions. 675 00:32:33,550 --> 00:32:37,110 Think about Baddeley's central executive, here. 676 00:32:37,110 --> 00:32:41,190 And it certainly seems to be really important in maintaining 677 00:32:41,190 --> 00:32:42,000 representations. 678 00:32:42,000 --> 00:32:44,280 It's not the only area of the brain that does it. 679 00:32:44,280 --> 00:32:46,260 You'll see this happening in temporal lobes 680 00:32:46,260 --> 00:32:48,090 and in parietal lobes. 681 00:32:48,090 --> 00:32:51,900 But the prefrontal cortex seems to be behaving differently 682 00:32:51,900 --> 00:32:53,620 than those other areas. 683 00:32:53,620 --> 00:32:55,230 So, this is a study. 684 00:32:55,230 --> 00:32:58,330 They had that same image recognition task. 685 00:32:58,330 --> 00:32:59,310 Look, a butterfly. 686 00:32:59,310 --> 00:33:00,570 Take it away. 687 00:33:00,570 --> 00:33:01,610 Look, another object. 688 00:33:01,610 --> 00:33:04,260 Is it the same one as I just showed you or different? 689 00:33:04,260 --> 00:33:07,980 But during the delay period, they actually 690 00:33:07,980 --> 00:33:09,840 showed other images. 691 00:33:09,840 --> 00:33:12,750 So I'd be like, look, a butterfly. 692 00:33:12,750 --> 00:33:13,740 Remember the butterfly. 693 00:33:13,740 --> 00:33:15,030 Look, a beach ball. 694 00:33:15,030 --> 00:33:16,920 Look, a crescent wrench. 695 00:33:16,920 --> 00:33:19,190 Look, a chair. 696 00:33:19,190 --> 00:33:20,190 OK, test time. 697 00:33:20,190 --> 00:33:21,060 Look, a butterfly. 698 00:33:21,060 --> 00:33:23,570 Is this what you just saw? 699 00:33:23,570 --> 00:33:27,310 And so you'd have all of these images presented. 700 00:33:27,310 --> 00:33:29,080 Butterfly is the one you just saw. 701 00:33:29,080 --> 00:33:32,389 But what's holding onto that image? 702 00:33:32,389 --> 00:33:33,930 So if you don't have the distractors, 703 00:33:33,930 --> 00:33:37,734 if it's just a butterfly, delay, butterfly again, 704 00:33:37,734 --> 00:33:39,150 then you'll see sustained activity 705 00:33:39,150 --> 00:33:43,677 in both the temporal areas and in the prefrontal cortex. 706 00:33:43,677 --> 00:33:45,510 But if you get these distractors in between, 707 00:33:45,510 --> 00:33:49,260 when the first distractors are shown, 708 00:33:49,260 --> 00:33:52,737 that representation in the temporal cortex goes away. 709 00:33:52,737 --> 00:33:53,820 It just stops being there. 710 00:33:53,820 --> 00:33:56,250 Whereas prefrontal cortex's activity 711 00:33:56,250 --> 00:33:59,160 remains high throughout the entire delay period. 712 00:34:02,220 --> 00:34:05,850 Prefrontal cortex seems to be special, not 713 00:34:05,850 --> 00:34:08,699 so much in terms of what kind of material it's storing. 714 00:34:08,699 --> 00:34:11,880 It's not, you know, just visual spatial or just phonological 715 00:34:11,880 --> 00:34:13,600 or anything like that. 716 00:34:13,600 --> 00:34:17,820 But it's better, almost, at storing it. 717 00:34:17,820 --> 00:34:20,220 It seems to be specialized to hold on to this stuff 718 00:34:20,220 --> 00:34:22,469 when you want to hold onto it, when you're consciously 719 00:34:22,469 --> 00:34:25,860 trying to hold onto a particular piece of information and not 720 00:34:25,860 --> 00:34:27,300 something else. 721 00:34:27,300 --> 00:34:29,790 It seems to be better at resisting distractions, 722 00:34:29,790 --> 00:34:34,530 better at maintaining that representation. 723 00:34:34,530 --> 00:34:36,854 And it's, of course, also running 724 00:34:36,854 --> 00:34:38,020 all of that executive stuff. 725 00:34:56,489 --> 00:34:59,520 So prefrontal cortex definitely seems 726 00:34:59,520 --> 00:35:04,200 to be involved in the storage stages of working memory. 727 00:35:04,200 --> 00:35:06,827 And it seems to be better at it than some other parts 728 00:35:06,827 --> 00:35:07,410 of your brain. 729 00:35:34,750 --> 00:35:38,930 But Baddeley's model says that, hey look, executive 730 00:35:38,930 --> 00:35:41,870 and executive control and storage are 731 00:35:41,870 --> 00:35:43,700 different functions, but we're seeing 732 00:35:43,700 --> 00:35:46,970 them both happen in more or less the same portions 733 00:35:46,970 --> 00:35:48,410 of your brain. 734 00:35:48,410 --> 00:35:50,040 Is this a problem? 735 00:35:50,040 --> 00:35:53,540 So there's two kind of possibilities that 736 00:35:53,540 --> 00:35:54,974 don't conflict with the model. 737 00:35:54,974 --> 00:35:56,390 There's a third possibility, which 738 00:35:56,390 --> 00:35:59,090 is that the model is wrong, but the model, thus far, 739 00:35:59,090 --> 00:36:01,340 has been really, really useful in trying to break down 740 00:36:01,340 --> 00:36:02,990 what working memory is. 741 00:36:02,990 --> 00:36:05,234 Nobody is quite willing to throw it out. 742 00:36:05,234 --> 00:36:07,400 It could just be that there are different subregions 743 00:36:07,400 --> 00:36:10,010 or different subpopulations of neurons 744 00:36:10,010 --> 00:36:16,430 that are involved in storage and maintenance of working memory 745 00:36:16,430 --> 00:36:19,190 items versus in this kind of executive control 746 00:36:19,190 --> 00:36:21,477 of working items. 747 00:36:21,477 --> 00:36:23,060 And there's another possibility, which 748 00:36:23,060 --> 00:36:28,299 is this kind of tweaking of our model. 749 00:36:28,299 --> 00:36:29,840 Which says that the prefrontal cortex 750 00:36:29,840 --> 00:36:35,780 is specialized to maintain information that is 751 00:36:35,780 --> 00:36:37,980 related to a particular goal. 752 00:36:37,980 --> 00:36:40,790 So if you have a goal, you want to do something. 753 00:36:40,790 --> 00:36:43,730 Prefrontal cortex hangs on to that goal, 754 00:36:43,730 --> 00:36:48,710 modifies, works with information relative to that goal 755 00:36:48,710 --> 00:36:55,680 versus merely just holding items in memory in the sense 756 00:36:55,680 --> 00:36:57,680 where-- like, walking through the world requires 757 00:36:57,680 --> 00:36:59,450 that you have some amount of visual information 758 00:36:59,450 --> 00:37:00,320 and working memory. 759 00:37:00,320 --> 00:37:02,690 You've got to remember where things were when you just 760 00:37:02,690 --> 00:37:06,180 looked at them and all of that. 761 00:37:06,180 --> 00:37:08,490 So this is called the Goal Maintenance Model. 762 00:37:08,490 --> 00:37:10,280 This is the prefrontal cortex maintains 763 00:37:10,280 --> 00:37:13,370 the information about what your goal is 764 00:37:13,370 --> 00:37:18,110 and it directs your attention and behavior 765 00:37:18,110 --> 00:37:19,180 to attain that goal. 766 00:37:19,180 --> 00:37:21,500 So this is classic executive function there. 767 00:37:21,500 --> 00:37:28,540 Directing attention, coordinating behaviors, 768 00:37:28,540 --> 00:37:30,060 prefrontal cortex is involved. 769 00:37:30,060 --> 00:37:36,810 For example, if you are, so for example, 770 00:37:36,810 --> 00:37:40,790 I used to work about half an hour from my house. 771 00:37:40,790 --> 00:37:42,924 And that was my route every morning. 772 00:37:42,924 --> 00:37:44,840 I went down the end of the road, I turn right. 773 00:37:44,840 --> 00:37:47,030 I went a couple miles, I turned left. 774 00:37:47,030 --> 00:37:48,590 This was my driving route. 775 00:37:48,590 --> 00:37:51,050 If I tried to go somewhere that was going to be about 2/3 776 00:37:51,050 --> 00:37:53,600 of the way to work and then make a different turn, 777 00:37:53,600 --> 00:37:56,060 if I'm not paying attention when I get to that turn, 778 00:37:56,060 --> 00:37:59,240 I'm going to go straight past it and stay on my usual route 779 00:37:59,240 --> 00:38:00,970 to work. 780 00:38:00,970 --> 00:38:04,760 Prefrontal cortex not doing its job. 781 00:38:04,760 --> 00:38:08,240 If I, in order to remember to make that left hand turn, 782 00:38:08,240 --> 00:38:12,740 I've got to remember that, hey, I need to do this. 783 00:38:12,740 --> 00:38:14,560 Keeping a goal in mind. 784 00:38:14,560 --> 00:38:16,280 Prefrontal cortex is maintaining that I 785 00:38:16,280 --> 00:38:19,580 need to make this left hand turn that's not on my usual route. 786 00:38:19,580 --> 00:38:23,470 And it's got to direct my behavior. 787 00:38:23,470 --> 00:38:25,720 It's got to say, OK, the prefrontal cortex says, 788 00:38:25,720 --> 00:38:27,500 OK, get in the left hand lane. 789 00:38:27,500 --> 00:38:28,790 Move over. 790 00:38:28,790 --> 00:38:33,749 All of that is being modulated by this goal-related ability 791 00:38:33,749 --> 00:38:34,790 of the prefrontal cortex. 792 00:38:39,082 --> 00:38:41,078 AUDIENCE: If it doesn't do its job, 793 00:38:41,078 --> 00:38:44,289 and you forget a lot of things that you want to do, like, 794 00:38:44,289 --> 00:38:45,080 five minutes after? 795 00:38:45,080 --> 00:38:47,046 ABBY NOYCE: That would be bad. 796 00:38:47,046 --> 00:38:49,170 AUDIENCE: Is that Alzheimer's or is that different? 797 00:38:49,170 --> 00:38:50,711 ABBY NOYCE: Alzheimer's is different. 798 00:38:50,711 --> 00:38:55,290 Alzheimer's, prefrontal cortex is one of the areas that 799 00:38:55,290 --> 00:38:56,570 degrades in Alzheimer's. 800 00:38:56,570 --> 00:38:58,050 It's certainly not the only one. 801 00:39:02,180 --> 00:39:07,980 And there's a particular pattern of behavioral deficits that you 802 00:39:07,980 --> 00:39:09,690 see in Alzheimer's. 803 00:39:09,690 --> 00:39:12,460 And the short term memory loss is really only one of them. 804 00:39:16,920 --> 00:39:20,780 But, yeah, definitely prefrontal cortex not doing all of this 805 00:39:20,780 --> 00:39:25,520 executive control that, like, inhibits irrational behavior 806 00:39:25,520 --> 00:39:28,310 and keeps you, and does attention, 807 00:39:28,310 --> 00:39:32,780 and helps you maintain goals, all of those are abilities that 808 00:39:32,780 --> 00:39:36,840 you see decreased in Alzheimer's. 809 00:39:36,840 --> 00:39:39,710 All right, so, think about this in context 810 00:39:39,710 --> 00:39:43,500 of that paper on motivated seeing that we did last week. 811 00:39:43,500 --> 00:39:47,870 Where subjects were given an ambiguous stimulus. 812 00:39:47,870 --> 00:39:51,290 They had, consciously or not, a particular preference 813 00:39:51,290 --> 00:39:53,740 to perceive that stimulus in one way or another, right? 814 00:39:53,740 --> 00:39:54,850 Remember the horse-seal? 815 00:39:57,450 --> 00:39:59,160 What do you think? 816 00:39:59,160 --> 00:40:04,530 Is prefrontal cortex involved in how you perceive 817 00:40:04,530 --> 00:40:07,695 that ambiguous stimulus? 818 00:40:10,761 --> 00:40:13,260 The question here, the question that you would ask, I think, 819 00:40:13,260 --> 00:40:17,280 is does that preference to not drink 820 00:40:17,280 --> 00:40:23,220 the viscous chunky foul smelling vegetarian, vegan, smoothie, 821 00:40:23,220 --> 00:40:24,230 is a goal? 822 00:40:24,230 --> 00:40:26,779 And it's probably not, well, it's probably actually 823 00:40:26,779 --> 00:40:29,070 something that's pretty close to the front of your mind 824 00:40:29,070 --> 00:40:34,596 as you watch your score get lower and lower, but-- 825 00:40:34,596 --> 00:40:36,906 AUDIENCE: Was that a pun, or? 826 00:40:36,906 --> 00:40:38,760 ABBY NOYCE: No. 827 00:40:38,760 --> 00:40:40,980 I tend to be the one who makes puns by accident 828 00:40:40,980 --> 00:40:44,525 and then other people laugh at them. 829 00:40:44,525 --> 00:40:46,551 AUDIENCE: Front of your mind, prefrontal-- 830 00:40:46,551 --> 00:40:48,300 ABBY NOYCE: Prefrontal cortex is the stuff 831 00:40:48,300 --> 00:40:50,672 in front of the frontal cortex. 832 00:40:50,672 --> 00:40:52,338 AUDIENCE: So it could actually be a pun. 833 00:40:52,338 --> 00:40:53,337 ABBY NOYCE: It would be. 834 00:40:53,337 --> 00:40:54,987 AUDIENCE: [INAUDIBLE] 835 00:40:54,987 --> 00:40:56,570 ABBY NOYCE: So one of the things that, 836 00:40:56,570 --> 00:40:59,111 one of the questions here, and it's one to which I don't have 837 00:40:59,111 --> 00:41:02,290 an answer to, is whether kind of, 838 00:41:02,290 --> 00:41:05,120 for something to be considered a goal in this goal maintenance 839 00:41:05,120 --> 00:41:09,050 model, depends on it being a goal that you are consciously 840 00:41:09,050 --> 00:41:09,770 aware of or not. 841 00:41:13,570 --> 00:41:15,970 Just something to think about. 842 00:41:15,970 --> 00:41:18,730 Anyway, all of this stuff is interrelated. 843 00:41:18,730 --> 00:41:22,130 Cognition requires lots of different capabilities. 844 00:41:22,130 --> 00:41:23,380 I like looking at connections. 845 00:41:23,380 --> 00:41:29,650 OK, so we've been talking about, moving right along, 846 00:41:29,650 --> 00:41:32,950 we've been talking about how different parts of your brain 847 00:41:32,950 --> 00:41:36,610 are involved in maintaining representations of information. 848 00:41:36,610 --> 00:41:39,260 We haven't talked much about what exactly 849 00:41:39,260 --> 00:41:40,690 these representations are. 850 00:41:40,690 --> 00:41:44,000 We know that there are patterns of firing in particular groups 851 00:41:44,000 --> 00:41:44,500 of neurons. 852 00:41:47,977 --> 00:41:49,310 So, we're going to switch gears. 853 00:41:49,310 --> 00:41:51,518 We're going to go back to talking kind of information 854 00:41:51,518 --> 00:41:55,360 theory a little bit about what kinds of representations 855 00:41:55,360 --> 00:41:56,200 the brain uses. 856 00:41:56,200 --> 00:41:59,860 Remember, a representation is a physical state 857 00:41:59,860 --> 00:42:05,480 that stands for something, for very broad events of something. 858 00:42:05,480 --> 00:42:08,750 And it carries information-- 859 00:42:08,750 --> 00:42:09,750 wow, I can't type today. 860 00:42:09,750 --> 00:42:12,850 I'm sorry guys-- an object, event, or concept, 861 00:42:12,850 --> 00:42:14,440 and carries information. 862 00:42:17,440 --> 00:42:20,751 Representations must be intentional and information 863 00:42:20,751 --> 00:42:21,250 carrying. 864 00:42:21,250 --> 00:42:24,370 And this is intentionality in the philosophy sense, 865 00:42:24,370 --> 00:42:26,200 not in the, like, you must intend 866 00:42:26,200 --> 00:42:29,310 to make a representation. 867 00:42:29,310 --> 00:42:32,650 That use of intentional is from its original Latin root. 868 00:42:32,650 --> 00:42:34,639 And it means that it has to refer to something. 869 00:42:34,639 --> 00:42:36,430 It's from a Latin root that means to point. 870 00:42:39,050 --> 00:42:43,300 So the representation must be, must refer to something, it 871 00:42:43,300 --> 00:42:44,380 can't just-- 872 00:42:44,380 --> 00:42:46,420 and a representation has to carry 873 00:42:46,420 --> 00:42:52,840 information about whatever the something is that it's about. 874 00:42:52,840 --> 00:42:56,800 So, if, for example, your representation of this chair 875 00:42:56,800 --> 00:42:59,409 might hold the fact that it's usually in this classroom, 876 00:42:59,409 --> 00:43:00,700 you've seen it for three weeks. 877 00:43:00,700 --> 00:43:01,840 It's red. 878 00:43:01,840 --> 00:43:03,700 It can be sat in. 879 00:43:03,700 --> 00:43:05,680 It's kind of funky looking. 880 00:43:05,680 --> 00:43:08,800 All of these are pieces of information 881 00:43:08,800 --> 00:43:12,220 about this chair that would be contained 882 00:43:12,220 --> 00:43:16,061 within your representation. 883 00:43:16,061 --> 00:43:17,854 Your representation of this chair 884 00:43:17,854 --> 00:43:20,270 might include the fact that it is a member of this broader 885 00:43:20,270 --> 00:43:21,380 category, "chairs." 886 00:43:21,380 --> 00:43:25,040 You'd have connections like that. 887 00:43:25,040 --> 00:43:28,250 So the information the guys, who are like, hey look, cognition 888 00:43:28,250 --> 00:43:29,480 is about information. 889 00:43:29,480 --> 00:43:32,240 What kinds of information might we be dealing with? 890 00:43:32,240 --> 00:43:34,940 Have four main types of representations 891 00:43:34,940 --> 00:43:36,770 that people talk about. 892 00:43:36,770 --> 00:43:41,720 And the most basic is this kind of modality specific image. 893 00:43:41,720 --> 00:43:45,320 So you take a photograph of something. 894 00:43:45,320 --> 00:43:47,060 That's a representation. 895 00:43:47,060 --> 00:43:51,860 It causes changes in the CCD on your digital camera. 896 00:43:51,860 --> 00:43:55,160 It contains information about the amount of light 897 00:43:55,160 --> 00:43:57,110 out there in the world. 898 00:43:57,110 --> 00:44:00,680 It points to a picture of whatever you took a picture of. 899 00:44:03,800 --> 00:44:08,690 And so, an image type representation in your brain 900 00:44:08,690 --> 00:44:13,160 is, again, very simple, information about the light 901 00:44:13,160 --> 00:44:16,190 that is reflected off of stuff and onto your retina 902 00:44:16,190 --> 00:44:20,051 at each point out there in the world. 903 00:44:20,051 --> 00:44:21,800 We talked about, last week, that the brain 904 00:44:21,800 --> 00:44:26,840 stores this basic image information in early stages 905 00:44:26,840 --> 00:44:28,340 of visual processing. 906 00:44:28,340 --> 00:44:32,720 Remember the monkey brain with the bullseye pattern overlaid 907 00:44:32,720 --> 00:44:36,120 right on the brain, there? 908 00:44:36,120 --> 00:44:39,980 That retinotopic organization where information about each 909 00:44:39,980 --> 00:44:42,514 point just lines right up on the back of your brain. 910 00:44:42,514 --> 00:44:44,930 So the brain doesn't do a whole lot with this information, 911 00:44:44,930 --> 00:44:47,960 but it definitely has this particular type 912 00:44:47,960 --> 00:44:49,275 of representation. 913 00:44:52,460 --> 00:44:56,800 There's an early visual processing format. 914 00:45:21,000 --> 00:45:25,530 At later stages of processing, then you're 915 00:45:25,530 --> 00:45:29,400 using a more sophisticated representation, again, 916 00:45:29,400 --> 00:45:31,810 of a visual stimulus of some sort. 917 00:45:31,810 --> 00:45:34,160 And this is a feature record. 918 00:45:34,160 --> 00:45:36,660 We talked about this when we talked about object recognition 919 00:45:36,660 --> 00:45:38,880 and vision last week. 920 00:45:38,880 --> 00:45:41,340 So a feature is a meaningful sensory aspect 921 00:45:41,340 --> 00:45:42,350 of a perceived stimulus. 922 00:45:42,350 --> 00:45:44,520 So, edges, for example, are definitely 923 00:45:44,520 --> 00:45:45,870 a feature of the visual world. 924 00:45:45,870 --> 00:45:48,990 And we know the brain is very good at finding edges. 925 00:45:48,990 --> 00:45:52,590 It's got cells that are specialized to detect edges. 926 00:45:52,590 --> 00:45:55,920 And so there's probably at least, and so a neuron 927 00:45:55,920 --> 00:45:59,070 that detects edges is a representation. 928 00:45:59,070 --> 00:46:03,000 It's a representation of a particular sensory aspect, 929 00:46:03,000 --> 00:46:06,660 of a meaningful sensory aspect of a stimulus. 930 00:46:06,660 --> 00:46:11,190 Frogs have cells in their, cells in their retinas 931 00:46:11,190 --> 00:46:14,700 that respond to dark dots on light backgrounds that 932 00:46:14,700 --> 00:46:16,440 move around. 933 00:46:16,440 --> 00:46:17,470 Their bug detectors. 934 00:46:17,470 --> 00:46:19,592 They're responding to a meaningful thing 935 00:46:19,592 --> 00:46:20,550 out there in the world. 936 00:46:20,550 --> 00:46:23,440 To particular aspects of it. 937 00:46:23,440 --> 00:46:28,900 So that, those cells are a feature record representation. 938 00:46:28,900 --> 00:46:33,450 A representation of a bug, of something I want to eat. 939 00:46:33,450 --> 00:46:35,400 If you are a frog. 940 00:46:35,400 --> 00:46:37,185 Probably not if you are a human. 941 00:46:37,185 --> 00:46:41,150 AUDIENCE: My brother ate a fish food once. 942 00:46:41,150 --> 00:46:44,085 ABBY NOYCE: I think I have sampled fish food once. 943 00:46:44,085 --> 00:46:44,960 I think I will admit. 944 00:46:48,110 --> 00:46:49,897 I was young, I knew no better. 945 00:46:49,897 --> 00:46:50,855 AUDIENCE: Was it yummy? 946 00:46:50,855 --> 00:46:52,197 ABBY NOYCE: No. 947 00:46:52,197 --> 00:46:53,530 It was, like, the little flakes. 948 00:46:53,530 --> 00:46:54,700 And I tried one flake. 949 00:46:54,700 --> 00:46:56,591 And I said, I'm not doing that again. 950 00:46:56,591 --> 00:47:00,650 AUDIENCE: He said at the time, it tasted good. 951 00:47:00,650 --> 00:47:01,940 ABBY NOYCE: That's intense. 952 00:47:01,940 --> 00:47:02,524 Older brother? 953 00:47:02,524 --> 00:47:03,189 Younger brother? 954 00:47:03,189 --> 00:47:04,570 AUDIENCE: He's my older brother. 955 00:47:04,570 --> 00:47:06,861 ABBY NOYCE: He was probably trying to get you to do it. 956 00:47:08,690 --> 00:47:11,440 Speaking as an oldest child. 957 00:47:11,440 --> 00:47:13,910 OK, so now let's talk-- there's, the other two 958 00:47:13,910 --> 00:47:19,740 are these kind of more abstract types of representation. 959 00:47:19,740 --> 00:47:22,860 So for a long time people had this idea of amodal symbols. 960 00:47:22,860 --> 00:47:24,620 So the feature records and the images 961 00:47:24,620 --> 00:47:26,330 that we were talking about before 962 00:47:26,330 --> 00:47:29,090 are tied to a particular sensory modality. 963 00:47:29,090 --> 00:47:31,010 They allow you to represent things 964 00:47:31,010 --> 00:47:32,820 in their particular sensory terms, 965 00:47:32,820 --> 00:47:35,640 but they won't help you to, for example, 966 00:47:35,640 --> 00:47:40,730 connect your image of this chair with the word "chair" 967 00:47:40,730 --> 00:47:42,890 with, like, your broader concept about chairs-- 968 00:47:42,890 --> 00:47:44,806 that they exist, that they are for sitting in, 969 00:47:44,806 --> 00:47:47,580 that they generally have a seat and back and four legs. 970 00:47:47,580 --> 00:47:49,700 All the things that are about chairs, 971 00:47:49,700 --> 00:47:53,300 which aren't necessarily visual sorts of things. 972 00:47:53,300 --> 00:47:57,020 So most people use amodal representations 973 00:47:57,020 --> 00:47:59,690 to talk about the relationships of different objects 974 00:47:59,690 --> 00:48:02,190 in the world to one another. 975 00:48:02,190 --> 00:48:04,460 So you'll see people talking about frames 976 00:48:04,460 --> 00:48:09,175 and semantic networks, which are both ways of specifying. 977 00:48:09,175 --> 00:48:10,550 And just in case it wasn't clear, 978 00:48:10,550 --> 00:48:13,370 we're back into information processing theory 979 00:48:13,370 --> 00:48:16,220 and not so much into what the brain does here. 980 00:48:16,220 --> 00:48:18,560 But they specify relationships between objects. 981 00:48:18,560 --> 00:48:22,100 So there's, you know, water bottle on the table, 982 00:48:22,100 --> 00:48:24,500 and that kind of "on the" relationship would be 983 00:48:24,500 --> 00:48:26,717 an amodal symbol in your brain. 984 00:48:26,717 --> 00:48:29,300 It doesn't have anything to do with the visual representation, 985 00:48:29,300 --> 00:48:31,670 per se, but it's information about how objects 986 00:48:31,670 --> 00:48:34,310 are arranged in the world. 987 00:48:34,310 --> 00:48:36,470 Basically, frames and semantic networks 988 00:48:36,470 --> 00:48:39,990 are storing roughly the same pieces of information. 989 00:48:39,990 --> 00:48:42,030 And these are, there are subtle differences 990 00:48:42,030 --> 00:48:45,740 that I don't entirely understand exactly how they're wired up. 991 00:48:45,740 --> 00:48:47,449 And scientists write nasty letters 992 00:48:47,449 --> 00:48:49,740 to journals about each other, about which one is right. 993 00:48:52,610 --> 00:48:58,460 And then there's also what's called a property list amodal 994 00:48:58,460 --> 00:48:59,180 symbol. 995 00:48:59,180 --> 00:49:05,960 So for example, for this water bottle, 996 00:49:05,960 --> 00:49:09,500 you might have the property that it is blue, 997 00:49:09,500 --> 00:49:12,710 which is a little bit too concrete, really for this. 998 00:49:12,710 --> 00:49:15,170 But you might have "contains water" as a property. 999 00:49:15,170 --> 00:49:17,810 You might have "made of plastic" as a property. 1000 00:49:17,810 --> 00:49:21,590 You might have "unbreakable" as a property. 1001 00:49:21,590 --> 00:49:24,230 These aren't strict perceptual ideas, 1002 00:49:24,230 --> 00:49:26,210 but they are things that are connected 1003 00:49:26,210 --> 00:49:27,305 to one particular object. 1004 00:49:29,920 --> 00:49:33,140 So amodal symbols are kind of necessary 1005 00:49:33,140 --> 00:49:35,060 for any good theoretical model of how 1006 00:49:35,060 --> 00:49:37,372 the brain stores information. 1007 00:49:37,372 --> 00:49:39,830 There's no real evidence that there are cells in your brain 1008 00:49:39,830 --> 00:49:41,660 that do this, per se. 1009 00:49:41,660 --> 00:49:43,430 There's no good neurological evidence 1010 00:49:43,430 --> 00:49:47,310 for these types of representations. 1011 00:49:47,310 --> 00:49:49,317 So they're kind of up in the air. 1012 00:49:49,317 --> 00:49:50,900 More recently, people have been trying 1013 00:49:50,900 --> 00:49:55,160 to wrap their heads around these not directly perceptual 1014 00:49:55,160 --> 00:49:58,310 representations with a slightly different model, which 1015 00:49:58,310 --> 00:50:02,270 is as a pattern of activation in a neural network. 1016 00:50:02,270 --> 00:50:05,510 So you have a group of neurons, or group 1017 00:50:05,510 --> 00:50:07,490 of populations of neurons. 1018 00:50:07,490 --> 00:50:11,330 And for any given concept or idea 1019 00:50:11,330 --> 00:50:14,870 that you're trying to represent, then, some of them will fire 1020 00:50:14,870 --> 00:50:16,167 and some of them won't fire. 1021 00:50:16,167 --> 00:50:17,750 So you might have a particular pattern 1022 00:50:17,750 --> 00:50:21,590 of neurons that fires for this "on top of" concept. 1023 00:50:21,590 --> 00:50:26,150 Whether for the laptop being on top of the table or the binder 1024 00:50:26,150 --> 00:50:30,470 being on top of the chair, you'd have a pattern of neurons 1025 00:50:30,470 --> 00:50:31,740 that represents this. 1026 00:50:31,740 --> 00:50:35,360 And one of the things that's nice about this model 1027 00:50:35,360 --> 00:50:40,100 is that you can have, for example, I have two chairs 1028 00:50:40,100 --> 00:50:41,020 here. 1029 00:50:41,020 --> 00:50:41,581 Chair. 1030 00:50:41,581 --> 00:50:42,080 Chair. 1031 00:50:42,080 --> 00:50:44,210 And they're different. 1032 00:50:44,210 --> 00:50:45,680 You probably have representations 1033 00:50:45,680 --> 00:50:50,390 in your brains of these chairs that are not entirely the same. 1034 00:50:50,390 --> 00:50:53,180 But also, those chairs are more like each other 1035 00:50:53,180 --> 00:50:55,730 than they are like the water bottle. 1036 00:50:55,730 --> 00:50:58,860 So from a purely intuitive point of view, 1037 00:50:58,860 --> 00:51:01,610 it would make sense that your representations of each 1038 00:51:01,610 --> 00:51:04,760 of these chairs are more alike than either one of them 1039 00:51:04,760 --> 00:51:07,550 is like your representation of the water bottle. 1040 00:51:07,550 --> 00:51:11,480 And a model like this, where you're representing concepts 1041 00:51:11,480 --> 00:51:15,740 as a pattern in a group of neurons firing 1042 00:51:15,740 --> 00:51:17,630 rather than in a particular neuron responding 1043 00:51:17,630 --> 00:51:19,970 to one thing or the other, lets you 1044 00:51:19,970 --> 00:51:23,240 take advantage of these differences among entities 1045 00:51:23,240 --> 00:51:26,060 within a category while still giving you a way to have 1046 00:51:26,060 --> 00:51:27,840 category representations. 1047 00:51:30,600 --> 00:51:34,530 There's not good empirical evidence for this yet. 1048 00:51:34,530 --> 00:51:37,920 But it seems to fit what we understand about how neurons 1049 00:51:37,920 --> 00:51:44,250 work better than the straight up amodal symbols older 1050 00:51:44,250 --> 00:51:47,840 school model. 1051 00:51:47,840 --> 00:51:52,650 OK, questions.