WEBVTT 00:00:00.000 --> 00:00:01.456 align:middle line:90% [SQUEAKING] 00:00:01.456 --> 00:00:03.368 align:middle line:90% [RUSTLING] 00:00:03.368 --> 00:00:06.714 align:middle line:90% [CLICKING] 00:00:06.714 --> 00:00:09.900 align:middle line:90% 00:00:09.900 --> 00:00:13.040 align:middle line:84% JOSH MCDERMOTT: So we got three classes left. 00:00:13.040 --> 00:00:15.660 align:middle line:84% Today we're going to finish talking about attention. 00:00:15.660 --> 00:00:19.720 align:middle line:84% And then we're going to talk about touch, taste, and smell. 00:00:19.720 --> 00:00:21.926 align:middle line:90% OK? 00:00:21.926 --> 00:00:27.680 align:middle line:84% So last time we got started with attention, 00:00:27.680 --> 00:00:31.840 align:middle line:84% talking about the fact that it's kind of an ill-defined thing. 00:00:31.840 --> 00:00:35.960 align:middle line:84% It's a word that we're all probably comfortable using. 00:00:35.960 --> 00:00:39.400 align:middle line:84% And the idea that it's mostly been studied 00:00:39.400 --> 00:00:42.840 align:middle line:84% in the context of a pretty wide range of phenomena that 00:00:42.840 --> 00:00:46.280 align:middle line:84% don't necessarily always have a unifying framework, 00:00:46.280 --> 00:00:50.220 align:middle line:84% and that as of now, we're mostly lacking, 00:00:50.220 --> 00:00:52.620 align:middle line:84% I would say, a rigorous computational understanding. 00:00:52.620 --> 00:00:54.495 align:middle line:84% But there are a lot of interesting phenomena, 00:00:54.495 --> 00:00:59.080 align:middle line:84% and I think the time is ripe for actually making progress 00:00:59.080 --> 00:01:06.667 align:middle line:84% on a real, better theoretical understanding of attention. 00:01:06.667 --> 00:01:08.250 align:middle line:84% And these sorts of phenomena, I think, 00:01:08.250 --> 00:01:10.390 align:middle line:84% will provide the backbone for that. 00:01:10.390 --> 00:01:14.110 align:middle line:84% So we started off talking about cuing, and the idea 00:01:14.110 --> 00:01:18.790 align:middle line:84% that you can present a cue which provides information 00:01:18.790 --> 00:01:22.550 align:middle line:84% about the probable location of a subsequent stimulus 00:01:22.550 --> 00:01:25.670 align:middle line:84% that you have to either detect or recognize, 00:01:25.670 --> 00:01:29.770 align:middle line:84% and the fact that cues help you to detect something. 00:01:29.770 --> 00:01:32.310 align:middle line:84% So you get this dot here that indicates 00:01:32.310 --> 00:01:34.030 align:middle line:84% where something is likely to occur, 00:01:34.030 --> 00:01:36.288 align:middle line:84% and then you're faster at identifying 00:01:36.288 --> 00:01:37.330 align:middle line:90% what the letter would be. 00:01:37.330 --> 00:01:39.590 align:middle line:84% So the benefit of cuing is typically 00:01:39.590 --> 00:01:41.810 align:middle line:84% measured as a reduction in reaction time. 00:01:41.810 --> 00:01:43.830 align:middle line:84% So one of the findings that falls out 00:01:43.830 --> 00:01:46.230 align:middle line:84% of experiments like this is that the benefit 00:01:46.230 --> 00:01:49.390 align:middle line:84% that you get from a cue is localized in space, right? 00:01:49.390 --> 00:01:52.070 align:middle line:84% So there's some thing that happens 00:01:52.070 --> 00:01:55.030 align:middle line:84% to the subsequent processing of the stimulus that 00:01:55.030 --> 00:01:56.950 align:middle line:90% is spatially localized. 00:01:56.950 --> 00:02:00.440 align:middle line:84% And so it's often thought of with a spotlight metaphor. 00:02:00.440 --> 00:02:01.940 align:middle line:84% So there's this spotlight that gets 00:02:01.940 --> 00:02:05.600 align:middle line:84% kind of shown on some region of space that in some way, 00:02:05.600 --> 00:02:07.220 align:middle line:84% in some unspecified way, improves 00:02:07.220 --> 00:02:09.460 align:middle line:84% the processing of something that cashes out 00:02:09.460 --> 00:02:12.360 align:middle line:84% in the experiment as a faster reaction time. 00:02:12.360 --> 00:02:14.940 align:middle line:84% And then we talked about the difference between exogenous 00:02:14.940 --> 00:02:17.460 align:middle line:90% and endogenous cues. 00:02:17.460 --> 00:02:22.140 align:middle line:84% Where the exogenous cue, an example of which is shown here. 00:02:22.140 --> 00:02:23.640 align:middle line:90% So the dot pops up. 00:02:23.640 --> 00:02:25.940 align:middle line:90% It's like a physical transient. 00:02:25.940 --> 00:02:29.740 align:middle line:84% There's this involuntary reflex where attention 00:02:29.740 --> 00:02:32.540 align:middle line:84% gets drawn to that location and makes you faster 00:02:32.540 --> 00:02:33.280 align:middle line:90% at that location. 00:02:33.280 --> 00:02:34.905 align:middle line:84% But then there are also these cues that 00:02:34.905 --> 00:02:36.740 align:middle line:90% really require interpretation. 00:02:36.740 --> 00:02:38.977 align:middle line:84% So the arrow is indicating that something 00:02:38.977 --> 00:02:40.560 align:middle line:84% is likely to happen in that direction. 00:02:40.560 --> 00:02:45.540 align:middle line:84% And you have to know what the arrow means in order to do that. 00:02:45.540 --> 00:02:49.340 align:middle line:84% And so you can see differences between these two types of cues 00:02:49.340 --> 00:02:53.820 align:middle line:84% and plausibly two distinct types of attentional mechanisms 00:02:53.820 --> 00:02:56.540 align:middle line:84% in the latency at which they kick in. 00:02:56.540 --> 00:03:00.050 align:middle line:84% So we looked at how long it takes you 00:03:00.050 --> 00:03:01.610 align:middle line:90% to get a benefit from the cue. 00:03:01.610 --> 00:03:03.490 align:middle line:84% And this is an experimental paradigm 00:03:03.490 --> 00:03:07.210 align:middle line:84% where the time between the cue and the target is varied. 00:03:07.210 --> 00:03:10.530 align:middle line:84% And so what you find is that when the stimulus onset 00:03:10.530 --> 00:03:13.250 align:middle line:84% asynchrony, or SOA, is really short, 00:03:13.250 --> 00:03:15.448 align:middle line:84% you don't actually get a benefit from the cue. 00:03:15.448 --> 00:03:16.990 align:middle line:84% So that's what this graph is showing. 00:03:16.990 --> 00:03:18.560 align:middle line:90% So the y-axis is benefit. 00:03:18.560 --> 00:03:20.310 align:middle line:84% And this is the stimulus onset asynchrony. 00:03:20.310 --> 00:03:23.810 align:middle line:84% And so when that gets really, really short, obviously, you 00:03:23.810 --> 00:03:25.110 align:middle line:90% can't get a benefit of the cue. 00:03:25.110 --> 00:03:26.360 align:middle line:90% It takes some time to kick in. 00:03:26.360 --> 00:03:28.650 align:middle line:84% But when you have an exogenous cue, 00:03:28.650 --> 00:03:31.810 align:middle line:84% the benefit comes online more quickly 00:03:31.810 --> 00:03:35.210 align:middle line:84% than when it's a symbolic or endogenous cue, 00:03:35.210 --> 00:03:38.530 align:middle line:84% indicative of some distinct internal process. 00:03:38.530 --> 00:03:42.170 align:middle line:84% So we then transition to talk about evidence 00:03:42.170 --> 00:03:44.790 align:middle line:84% that the spotlight metaphor of attention is limited. 00:03:44.790 --> 00:03:47.290 align:middle line:84% So we talked about these multiple object tracking tasks 00:03:47.290 --> 00:03:49.650 align:middle line:84% where there's some subset of objects 00:03:49.650 --> 00:03:52.393 align:middle line:84% on the screen that are identified as things that you 00:03:52.393 --> 00:03:53.310 align:middle line:90% need to keep track of. 00:03:53.310 --> 00:03:55.690 align:middle line:90% They then start moving. 00:03:55.690 --> 00:04:00.000 align:middle line:84% And then at the end, you're queried as to whether or not 00:04:00.000 --> 00:04:01.767 align:middle line:84% a particular item was in that original set 00:04:01.767 --> 00:04:03.100 align:middle line:90% that you were supposed to track. 00:04:03.100 --> 00:04:06.120 align:middle line:84% So we tried this ourselves and then saw some data 00:04:06.120 --> 00:04:08.400 align:middle line:84% indicating that people have an ability 00:04:08.400 --> 00:04:11.223 align:middle line:84% to do this for four or five objects. 00:04:11.223 --> 00:04:13.640 align:middle line:84% It's a little bit harder than if you just have one or two, 00:04:13.640 --> 00:04:15.473 align:middle line:84% but you can definitely track multiple things 00:04:15.473 --> 00:04:18.513 align:middle line:90% at the same time. 00:04:18.513 --> 00:04:22.040 align:middle line:84% We saw an example of auditory attentive tracking, which 00:04:22.040 --> 00:04:26.000 align:middle line:84% might be necessary in cases where you're listening 00:04:26.000 --> 00:04:29.480 align:middle line:84% to multiple concurrent sources that are occupying 00:04:29.480 --> 00:04:31.177 align:middle line:90% the same approximate space. 00:04:31.177 --> 00:04:33.260 align:middle line:84% And you might have to track one of them over time. 00:04:33.260 --> 00:04:36.040 align:middle line:84% And we think that might be part of what happens in the cocktail 00:04:36.040 --> 00:04:37.040 align:middle line:90% party problem. 00:04:37.040 --> 00:04:41.000 align:middle line:84% And so we saw evidence that you can measure attention 00:04:41.000 --> 00:04:44.880 align:middle line:84% and attention to a thing by testing whether people 00:04:44.880 --> 00:04:48.640 align:middle line:84% are better at making a discrimination about the thing 00:04:48.640 --> 00:04:50.800 align:middle line:90% that they're attending. 00:04:50.800 --> 00:04:54.710 align:middle line:84% And so in this particular case, what you're discriminating 00:04:54.710 --> 00:04:57.390 align:middle line:84% is the presence or absence of the vibrato on one 00:04:57.390 --> 00:04:58.090 align:middle line:90% of the sources. 00:04:58.090 --> 00:05:01.710 align:middle line:84% And so what the results show is that for the source 00:05:01.710 --> 00:05:05.070 align:middle line:84% that you're tracking, your vibrato detection 00:05:05.070 --> 00:05:07.470 align:middle line:84% is better than the source that you're not tracking. 00:05:07.470 --> 00:05:10.190 align:middle line:84% So that gives you some way of actually testing the extent 00:05:10.190 --> 00:05:15.030 align:middle line:84% to which you've successfully attended to a particular thing. 00:05:15.030 --> 00:05:17.097 align:middle line:84% We noted that you can attend to features, right? 00:05:17.097 --> 00:05:19.430 align:middle line:84% So you can attend to the red things, or the blue things, 00:05:19.430 --> 00:05:21.910 align:middle line:84% or the horizontal things, and you become aware 00:05:21.910 --> 00:05:23.470 align:middle line:90% of their organization. 00:05:23.470 --> 00:05:25.990 align:middle line:84% We then talked about visual search 00:05:25.990 --> 00:05:28.390 align:middle line:84% where you have to detect some target that's 00:05:28.390 --> 00:05:31.070 align:middle line:84% maybe defined by features, and saw that there are 00:05:31.070 --> 00:05:33.990 align:middle line:84% some visual search displays where the target is 00:05:33.990 --> 00:05:35.170 align:middle line:90% very easy to detect. 00:05:35.170 --> 00:05:36.990 align:middle line:90% We say that it pops out. 00:05:36.990 --> 00:05:39.710 align:middle line:84% So we talked about how we typically 00:05:39.710 --> 00:05:42.230 align:middle line:84% measure performance in a visual search task 00:05:42.230 --> 00:05:46.150 align:middle line:84% by plotting the reaction times of how long it takes you 00:05:46.150 --> 00:05:48.490 align:middle line:84% to determine if the target is there or not, 00:05:48.490 --> 00:05:52.010 align:middle line:84% versus the number of items on the screen, the set size, OK? 00:05:52.010 --> 00:05:55.660 align:middle line:84% So these cases where the target so-called, "pops out" 00:05:55.660 --> 00:06:00.840 align:middle line:84% are indicated by results graphs that are flat. 00:06:00.840 --> 00:06:03.660 align:middle line:84% So the reaction time doesn't change with the number 00:06:03.660 --> 00:06:07.110 align:middle line:90% of items in the display. 00:06:07.110 --> 00:06:08.900 align:middle line:84% So these are typically situations 00:06:08.900 --> 00:06:12.980 align:middle line:84% where there's some unique value of an elementary feature. 00:06:12.980 --> 00:06:17.160 align:middle line:84% But then we also saw cases where the things don't pop out, right? 00:06:17.160 --> 00:06:18.820 align:middle line:84% And so a classic example of that is 00:06:18.820 --> 00:06:21.680 align:middle line:84% where the target is defined by a conjunction of features. 00:06:21.680 --> 00:06:27.260 align:middle line:84% So the target here is blue and vertical. 00:06:27.260 --> 00:06:30.300 align:middle line:84% And so in these cases, the results of the visual search 00:06:30.300 --> 00:06:32.900 align:middle line:84% experiment where you measure reaction time for displays 00:06:32.900 --> 00:06:35.680 align:middle line:84% of different sizes, look something like this. 00:06:35.680 --> 00:06:39.420 align:middle line:84% The reaction time grows with the set size. 00:06:39.420 --> 00:06:40.760 align:middle line:90% So the slope is not zero. 00:06:40.760 --> 00:06:43.900 align:middle line:84% It's got some number of milliseconds per item. 00:06:43.900 --> 00:06:47.580 align:middle line:84% So we saw a whole bunch of different examples of searches 00:06:47.580 --> 00:06:48.780 align:middle line:90% like that. 00:06:48.780 --> 00:06:55.050 align:middle line:84% And so the classical explanation for these slow searches 00:06:55.050 --> 00:06:59.070 align:middle line:84% is that the target is defined by a combination of properties, 00:06:59.070 --> 00:07:01.430 align:middle line:84% and so in order to detect the target, 00:07:01.430 --> 00:07:04.730 align:middle line:84% you have to combine those properties. 00:07:04.730 --> 00:07:09.290 align:middle line:84% And the combination is by hypothesis not automatic. 00:07:09.290 --> 00:07:13.170 align:middle line:84% And so one of the functions of attention 00:07:13.170 --> 00:07:16.810 align:middle line:84% is to facilitate the registration 00:07:16.810 --> 00:07:21.250 align:middle line:84% of the different properties that make up the target. 00:07:21.250 --> 00:07:24.390 align:middle line:84% And as a consequence, these searches are serial. 00:07:24.390 --> 00:07:26.850 align:middle line:84% So you have to move your attention around the display 00:07:26.850 --> 00:07:27.630 align:middle line:90% each time. 00:07:27.630 --> 00:07:30.470 align:middle line:84% Whatever is in the focus of attention gets combined. 00:07:30.470 --> 00:07:33.090 align:middle line:84% And then you check whether that's the target. 00:07:33.090 --> 00:07:33.930 align:middle line:90% And so it's slow. 00:07:33.930 --> 00:07:34.430 align:middle line:90% All right? 00:07:34.430 --> 00:07:36.073 align:middle line:84% And in particular, the time that it 00:07:36.073 --> 00:07:37.490 align:middle line:84% will take you to detect the target 00:07:37.490 --> 00:07:40.330 align:middle line:90% scales with the number of items. 00:07:40.330 --> 00:07:45.290 align:middle line:84% So you can measure kind of how fast the spotlight moves around. 00:07:45.290 --> 00:07:50.170 align:middle line:84% And so then we noted that visual search in the real world 00:07:50.170 --> 00:07:52.840 align:middle line:84% is often inefficient because a lot of the things 00:07:52.840 --> 00:07:56.600 align:middle line:84% that we in practice want to look for are defined by combinations 00:07:56.600 --> 00:07:58.545 align:middle line:90% of elementary features. 00:07:58.545 --> 00:07:59.920 align:middle line:84% And so the study of visual search 00:07:59.920 --> 00:08:02.960 align:middle line:84% has got a lot of real world relevance, 00:08:02.960 --> 00:08:06.040 align:middle line:84% both in terms of stuff that we do on a daily basis, but also 00:08:06.040 --> 00:08:08.760 align:middle line:84% some very important tasks, for instance, 00:08:08.760 --> 00:08:13.200 align:middle line:84% like finding abnormalities in like a radiology image 00:08:13.200 --> 00:08:16.760 align:middle line:84% if you're a doctor, or if you're working for TSA 00:08:16.760 --> 00:08:19.120 align:middle line:84% and you're looking at pictures from those scanners 00:08:19.120 --> 00:08:21.520 align:middle line:84% and you're looking for particular kinds of things 00:08:21.520 --> 00:08:23.440 align:middle line:84% that might look like a gun or whatever, 00:08:23.440 --> 00:08:25.680 align:middle line:84% that's essentially a visual search task, 00:08:25.680 --> 00:08:29.520 align:middle line:84% so lots of practical relevance for this kind of thing, 00:08:29.520 --> 00:08:33.799 align:middle line:84% and lots of interest in understanding 00:08:33.799 --> 00:08:36.440 align:middle line:84% what it is that causes problems for people 00:08:36.440 --> 00:08:39.860 align:middle line:84% and what you could potentially do to make it easier for people. 00:08:39.860 --> 00:08:43.080 align:middle line:90% 00:08:43.080 --> 00:08:45.640 align:middle line:84% And so the theoretical framework that was offered back 00:08:45.640 --> 00:08:48.342 align:middle line:84% in the early '80s to explain these kinds of phenomena was 00:08:48.342 --> 00:08:50.550 align:middle line:84% known as feature integration theory put forth by Anne 00:08:50.550 --> 00:08:51.590 align:middle line:90% Treisman. 00:08:51.590 --> 00:08:54.590 align:middle line:84% And the key idea is that there are distinct features that 00:08:54.590 --> 00:08:56.910 align:middle line:84% are represented separately, often 00:08:56.910 --> 00:08:59.670 align:middle line:84% posited to be in maps, that are registered 00:08:59.670 --> 00:09:02.630 align:middle line:90% to the retinotopic location. 00:09:02.630 --> 00:09:05.750 align:middle line:84% And then attention has this special function 00:09:05.750 --> 00:09:08.630 align:middle line:84% of allowing you to register the different features 00:09:08.630 --> 00:09:11.590 align:middle line:84% across locations, so combining things and binding them 00:09:11.590 --> 00:09:12.550 align:middle line:90% together. 00:09:12.550 --> 00:09:15.630 align:middle line:84% And one of the pieces of evidence for this notion 00:09:15.630 --> 00:09:18.030 align:middle line:84% came from the finding of illusory conjunctions 00:09:18.030 --> 00:09:21.168 align:middle line:84% where if you in part divert attention, in this case 00:09:21.168 --> 00:09:22.710 align:middle line:84% out to the periphery by asking people 00:09:22.710 --> 00:09:25.550 align:middle line:84% to pay attention to the numbers and then query them 00:09:25.550 --> 00:09:28.510 align:middle line:84% on the stuff in the middle, people will often 00:09:28.510 --> 00:09:29.970 align:middle line:90% make conjunction errors. 00:09:29.970 --> 00:09:33.733 align:middle line:84% So they'll miscombine the color in the letter, 00:09:33.733 --> 00:09:35.650 align:middle line:84% so they won't see some color that's not there. 00:09:35.650 --> 00:09:38.050 align:middle line:84% They won't tend to see some letter that's not there. 00:09:38.050 --> 00:09:41.830 align:middle line:84% But like they might say a green x or a red t. 00:09:41.830 --> 00:09:42.630 align:middle line:90% OK? 00:09:42.630 --> 00:09:45.990 align:middle line:84% And so this was taken as support for the idea 00:09:45.990 --> 00:09:48.660 align:middle line:84% that one of the things that happens when you're 00:09:48.660 --> 00:09:51.060 align:middle line:84% paying attention to something is that you're 00:09:51.060 --> 00:09:55.800 align:middle line:84% able to bind distinct properties of stimuli together, 00:09:55.800 --> 00:09:57.660 align:middle line:84% and that when attention is diverted, 00:09:57.660 --> 00:10:02.720 align:middle line:84% that binding is disrupted in some way. 00:10:02.720 --> 00:10:06.440 align:middle line:84% And so as I said last time, these days, 00:10:06.440 --> 00:10:10.460 align:middle line:84% it's hard to find people who are stalwart believers in feature 00:10:10.460 --> 00:10:13.520 align:middle line:84% integration theory for a bunch of different reasons. 00:10:13.520 --> 00:10:17.340 align:middle line:84% One of which is that the simple division 00:10:17.340 --> 00:10:19.660 align:middle line:84% between things that pop out and things that don't has 00:10:19.660 --> 00:10:21.260 align:middle line:90% proven to be more complicated. 00:10:21.260 --> 00:10:24.180 align:middle line:84% There's examples where you can get pop-out for properties that 00:10:24.180 --> 00:10:26.940 align:middle line:84% seem pretty high level and very distinct 00:10:26.940 --> 00:10:29.260 align:middle line:84% from the elementary features that were 00:10:29.260 --> 00:10:31.860 align:middle line:90% the cornerstone of the theory. 00:10:31.860 --> 00:10:33.540 align:middle line:84% But there's nonetheless, like the set 00:10:33.540 --> 00:10:36.900 align:middle line:84% of phenomena that kind of inspired that theory remained 00:10:36.900 --> 00:10:41.120 align:middle line:84% very real and things that I think demand explanation. 00:10:41.120 --> 00:10:42.960 align:middle line:84% And so that includes illusory conjunctions. 00:10:42.960 --> 00:10:45.260 align:middle line:84% It includes these differences between feature searches 00:10:45.260 --> 00:10:48.450 align:middle line:84% and conjunction searches, and so forth. 00:10:48.450 --> 00:10:48.950 align:middle line:90% All right. 00:10:48.950 --> 00:10:50.970 align:middle line:84% That's where we left off last time. 00:10:50.970 --> 00:10:54.710 align:middle line:84% Before I proceed, are there any questions about any of that? 00:10:54.710 --> 00:11:07.750 align:middle line:90% 00:11:07.750 --> 00:11:08.250 align:middle line:90% OK. 00:11:08.250 --> 00:11:11.610 align:middle line:90% 00:11:11.610 --> 00:11:12.110 align:middle line:90% All right. 00:11:12.110 --> 00:11:17.530 align:middle line:84% So to continue our tour of interesting phenomena related 00:11:17.530 --> 00:11:22.170 align:middle line:84% to attention, we're going to talk about the possibility 00:11:22.170 --> 00:11:24.650 align:middle line:84% that when you pay attention to something 00:11:24.650 --> 00:11:27.010 align:middle line:84% that improves the spatial resolution 00:11:27.010 --> 00:11:28.650 align:middle line:90% of the representation. 00:11:28.650 --> 00:11:35.090 align:middle line:84% And the idea behind this set of study 00:11:35.090 --> 00:11:37.690 align:middle line:84% and some others that are kind of like it, really 00:11:37.690 --> 00:11:40.250 align:middle line:84% goes way, way back right to the notion that 00:11:40.250 --> 00:11:44.010 align:middle line:84% rooted in introspection, that when you're really 00:11:44.010 --> 00:11:48.200 align:middle line:84% attending to something you can see it better. 00:11:48.200 --> 00:11:51.220 align:middle line:84% And that's kind of the whole reason why we pay attention, 00:11:51.220 --> 00:11:53.300 align:middle line:84% is somehow the processing is kind of improved. 00:11:53.300 --> 00:11:57.760 align:middle line:84% So the question is, in what respects do we see things better 00:11:57.760 --> 00:11:59.840 align:middle line:84% and can we actually measure that? 00:11:59.840 --> 00:12:03.460 align:middle line:84% And this is one example that tried to get at this. 00:12:03.460 --> 00:12:06.400 align:middle line:84% And I also like presenting this because it relates 00:12:06.400 --> 00:12:09.080 align:middle line:84% to some other things that we've talked about in the class, 00:12:09.080 --> 00:12:11.000 align:middle line:84% in particular texture segregation. 00:12:11.000 --> 00:12:13.880 align:middle line:84% So here we have a texture display. 00:12:13.880 --> 00:12:17.160 align:middle line:84% There's a little patch where the line segments are oriented 00:12:17.160 --> 00:12:19.680 align:middle line:84% differently from the rest of the texture. 00:12:19.680 --> 00:12:21.820 align:middle line:90% And you're able to detect that. 00:12:21.820 --> 00:12:26.040 align:middle line:84% And so this experiment makes use of this texture segregation 00:12:26.040 --> 00:12:26.540 align:middle line:90% task. 00:12:26.540 --> 00:12:32.320 align:middle line:84% Now, one of the peculiar things about this particular task 00:12:32.320 --> 00:12:35.440 align:middle line:84% is that unlike most other visual tasks, 00:12:35.440 --> 00:12:38.040 align:middle line:84% this task, texture segregation, is not 00:12:38.040 --> 00:12:41.400 align:middle line:84% easiest when the target is presented at the fovea. 00:12:41.400 --> 00:12:45.060 align:middle line:84% So for most kinds of things that I would ask you to do, 00:12:45.060 --> 00:12:47.510 align:middle line:84% you will do best at that task if you 00:12:47.510 --> 00:12:49.910 align:middle line:84% fixate the thing that you're being asked to detect, 00:12:49.910 --> 00:12:51.250 align:middle line:90% or discriminate, or whatever. 00:12:51.250 --> 00:12:55.190 align:middle line:84% But for texture segregation, that's not the case. 00:12:55.190 --> 00:12:57.790 align:middle line:84% You actually are best at texture segregation 00:12:57.790 --> 00:13:00.790 align:middle line:84% if you look a little bit to the side of the thing 00:13:00.790 --> 00:13:02.170 align:middle line:90% that you need to detect. 00:13:02.170 --> 00:13:05.455 align:middle line:90% 00:13:05.455 --> 00:13:10.190 align:middle line:84% And so the presumptive explanation for this phenomenon 00:13:10.190 --> 00:13:13.150 align:middle line:84% has to do with the dependence of the receptive 00:13:13.150 --> 00:13:16.310 align:middle line:84% fields in your visual system on eccentricity. 00:13:16.310 --> 00:13:19.230 align:middle line:84% So you remember how one of the core organizational principles 00:13:19.230 --> 00:13:22.790 align:middle line:84% of the visual system is that resolution 00:13:22.790 --> 00:13:25.670 align:middle line:90% changes with eccentricity? 00:13:25.670 --> 00:13:29.230 align:middle line:90% What's eccentricity? 00:13:29.230 --> 00:13:30.470 align:middle line:90% Yeah? 00:13:30.470 --> 00:13:32.030 align:middle line:84% AUDIENCE: Distance from the fovea. 00:13:32.030 --> 00:13:33.130 align:middle line:84% JOSH MCDERMOTT: Distance from the fovea. 00:13:33.130 --> 00:13:33.930 align:middle line:90% Yes, exactly. 00:13:33.930 --> 00:13:35.610 align:middle line:84% Eccentricity, distance from the fovea. 00:13:35.610 --> 00:13:38.230 align:middle line:84% So when you're at the fovea, the receptive fields 00:13:38.230 --> 00:13:39.330 align:middle line:90% tend to be small. 00:13:39.330 --> 00:13:42.050 align:middle line:84% And then as you move away from the fovea, they get larger. 00:13:42.050 --> 00:13:42.660 align:middle line:90% OK? 00:13:42.660 --> 00:13:44.800 align:middle line:90% And that's true in the retina. 00:13:44.800 --> 00:13:48.620 align:middle line:90% It's true in the visual cortex. 00:13:48.620 --> 00:13:49.120 align:middle line:90% OK. 00:13:49.120 --> 00:13:52.940 align:middle line:84% So receptive field sizes increase with eccentricity. 00:13:52.940 --> 00:13:56.460 align:middle line:84% If you think that the way that you solve this task 00:13:56.460 --> 00:14:01.880 align:middle line:84% is by using the outputs of filters, like oriented filters, 00:14:01.880 --> 00:14:07.380 align:middle line:84% Gabor-like filters that are processing this image, 00:14:07.380 --> 00:14:10.960 align:middle line:84% and if you suppose that those filters can vary in their scale, 00:14:10.960 --> 00:14:14.300 align:middle line:84% well there's going to be like a size of the filters that's 00:14:14.300 --> 00:14:17.640 align:middle line:84% optimal for this particular texture, where 00:14:17.640 --> 00:14:21.340 align:middle line:84% the size of the filter is about the same size as the elements 00:14:21.340 --> 00:14:22.620 align:middle line:90% in the texture. 00:14:22.620 --> 00:14:26.940 align:middle line:84% And so the idea is that there's an eccentricity at which 00:14:26.940 --> 00:14:30.180 align:middle line:84% the receptive fields are just right 00:14:30.180 --> 00:14:32.700 align:middle line:90% for that particular texture. 00:14:32.700 --> 00:14:35.680 align:middle line:84% And in many cases, if you're foveated it, 00:14:35.680 --> 00:14:37.940 align:middle line:84% the filters will be too small, and so you don't 00:14:37.940 --> 00:14:41.200 align:middle line:90% do quite as well at the task. 00:14:41.200 --> 00:14:44.410 align:middle line:84% So that's the premise of all of this. 00:14:44.410 --> 00:14:48.810 align:middle line:84% So that's just an interesting fact about texture segregation. 00:14:48.810 --> 00:14:51.930 align:middle line:84% So now in this experiment, they wanted 00:14:51.930 --> 00:14:56.410 align:middle line:84% to test what the effect would be of drawing attention 00:14:56.410 --> 00:14:57.850 align:middle line:90% to a location. 00:14:57.850 --> 00:15:01.050 align:middle line:84% And they're going to utilize the fact that performance 00:15:01.050 --> 00:15:04.530 align:middle line:84% on this task varies with eccentricity in a way that's 00:15:04.530 --> 00:15:07.830 align:middle line:84% plausibly dependent on the size of the receptive fields 00:15:07.830 --> 00:15:11.810 align:middle line:84% to then investigate how cuing attention to a location 00:15:11.810 --> 00:15:16.470 align:middle line:84% might change the size of the receptive fields. 00:15:16.470 --> 00:15:18.490 align:middle line:84% So here is a schematic of the task. 00:15:18.490 --> 00:15:19.770 align:middle line:90% So this is time. 00:15:19.770 --> 00:15:22.790 align:middle line:84% You got these different frames that people would be shown. 00:15:22.790 --> 00:15:25.610 align:middle line:90% So there's a fixation dot. 00:15:25.610 --> 00:15:28.350 align:middle line:84% And the organization of the trial, 00:15:28.350 --> 00:15:31.130 align:middle line:84% so this is one trial and a trial is 00:15:31.130 --> 00:15:33.690 align:middle line:84% defined by you giving one answer in response 00:15:33.690 --> 00:15:35.150 align:middle line:90% to the sequence of things. 00:15:35.150 --> 00:15:37.690 align:middle line:84% And so there's two intervals to the trial. 00:15:37.690 --> 00:15:38.190 align:middle line:90% All right? 00:15:38.190 --> 00:15:41.000 align:middle line:84% So you get a first cue and a first display followed 00:15:41.000 --> 00:15:43.900 align:middle line:84% by a brief mask, and then a second cue and a second display 00:15:43.900 --> 00:15:45.360 align:middle line:90% followed by a brief mask. 00:15:45.360 --> 00:15:47.923 align:middle line:90% 00:15:47.923 --> 00:15:50.340 align:middle line:84% You have to have the first display and the second display, 00:15:50.340 --> 00:15:52.160 align:middle line:84% and only one of those is going to have 00:15:52.160 --> 00:15:53.260 align:middle line:90% what's called the target. 00:15:53.260 --> 00:15:55.320 align:middle line:84% So that's a little segment of the texture 00:15:55.320 --> 00:15:58.080 align:middle line:84% that's oppositely oriented to the rest of the texture. 00:15:58.080 --> 00:15:59.840 align:middle line:84% That's the thing you have to detect. 00:15:59.840 --> 00:16:03.120 align:middle line:84% So your job when you do this experiment, is to look at this 00:16:03.120 --> 00:16:06.092 align:middle line:84% and say, the target was in the second display in this case. 00:16:06.092 --> 00:16:07.800 align:middle line:84% Or maybe it would be in the first display 00:16:07.800 --> 00:16:08.960 align:middle line:90% in some other case. 00:16:08.960 --> 00:16:10.420 align:middle line:90% So it's first or second. 00:16:10.420 --> 00:16:15.280 align:middle line:84% Now for both intervals there is a cue, which 00:16:15.280 --> 00:16:17.320 align:middle line:90% is a little thing that pops up. 00:16:17.320 --> 00:16:20.740 align:middle line:90% It's an exogenous cue. 00:16:20.740 --> 00:16:23.020 align:middle line:84% It draws your attention to a particular location. 00:16:23.020 --> 00:16:26.760 align:middle line:84% And so the cue is always valid in the sense 00:16:26.760 --> 00:16:30.760 align:middle line:84% that when the target appears it's at the cued location. 00:16:30.760 --> 00:16:32.840 align:middle line:84% But in one of the intervals there 00:16:32.840 --> 00:16:36.280 align:middle line:90% isn't going to be a target. 00:16:36.280 --> 00:16:36.820 align:middle line:90% All right. 00:16:36.820 --> 00:16:39.298 align:middle line:84% So you get the cue and then a display, another cue 00:16:39.298 --> 00:16:40.090 align:middle line:90% and then a display. 00:16:40.090 --> 00:16:41.757 align:middle line:84% And so you can see in this case, the cue 00:16:41.757 --> 00:16:45.490 align:middle line:84% is at the same spatial position as that target. 00:16:45.490 --> 00:16:45.990 align:middle line:90% OK? 00:16:45.990 --> 00:16:53.270 align:middle line:90% 00:16:53.270 --> 00:17:00.950 align:middle line:84% So the idea here is to compare what 00:17:00.950 --> 00:17:05.430 align:middle line:84% happens when you are cued to the target's location, when you're 00:17:05.430 --> 00:17:06.150 align:middle line:90% not. 00:17:06.150 --> 00:17:08.609 align:middle line:84% So I already told you that when you're not cued, 00:17:08.609 --> 00:17:10.730 align:middle line:84% when it's just a vanilla texture segregation task, 00:17:10.730 --> 00:17:13.910 align:middle line:84% this task has this funny property that performance 00:17:13.910 --> 00:17:19.349 align:middle line:84% is best not at the fovea, but at some intermediate eccentricity. 00:17:19.349 --> 00:17:22.550 align:middle line:84% And so the question is whether that is altered 00:17:22.550 --> 00:17:23.930 align:middle line:90% by the presence of the cue. 00:17:23.930 --> 00:17:28.790 align:middle line:84% And so if you think that maybe somehow the attentional cue is 00:17:28.790 --> 00:17:31.230 align:middle line:84% serving to improve spatial resolution by shrinking 00:17:31.230 --> 00:17:33.810 align:middle line:84% the receptive fields, then you might expect things to change. 00:17:33.810 --> 00:17:37.730 align:middle line:90% 00:17:37.730 --> 00:17:39.980 align:middle line:84% All right, before I go on, any questions about the way 00:17:39.980 --> 00:17:41.760 align:middle line:90% that this experiment works? 00:17:41.760 --> 00:17:52.460 align:middle line:90% 00:17:52.460 --> 00:17:53.700 align:middle line:90% So here are the results. 00:17:53.700 --> 00:17:54.200 align:middle line:90% All right. 00:17:54.200 --> 00:17:55.160 align:middle line:90% It's a little bit complicated. 00:17:55.160 --> 00:17:56.400 align:middle line:90% So let's walk through it. 00:17:56.400 --> 00:17:59.620 align:middle line:84% So for the time being, just look at the one 00:17:59.620 --> 00:18:03.860 align:middle line:84% on the left, in this particular panel, panel a. 00:18:03.860 --> 00:18:07.360 align:middle line:84% So the graph is plotting percent correct, 00:18:07.360 --> 00:18:09.140 align:middle line:84% so it's how well you do on the task, 00:18:09.140 --> 00:18:11.420 align:middle line:90% as a function of eccentricity. 00:18:11.420 --> 00:18:12.940 align:middle line:90% So this is 0. 00:18:12.940 --> 00:18:14.280 align:middle line:90% That's the fovea. 00:18:14.280 --> 00:18:16.620 align:middle line:84% And then this is moving out to greater eccentricities. 00:18:16.620 --> 00:18:17.260 align:middle line:90% OK? 00:18:17.260 --> 00:18:20.040 align:middle line:84% So for the moment, just look at the open symbols. 00:18:20.040 --> 00:18:23.760 align:middle line:84% So that's the case where the cue is neutral. 00:18:23.760 --> 00:18:28.100 align:middle line:84% So the cue is not telling you the location of the target. 00:18:28.100 --> 00:18:30.500 align:middle line:84% And so in this particular case, what happens 00:18:30.500 --> 00:18:34.380 align:middle line:84% is that performance is best at these intermediate 00:18:34.380 --> 00:18:36.320 align:middle line:84% eccentricities like four or five degrees. 00:18:36.320 --> 00:18:38.490 align:middle line:84% So this is the phenomena that we just talked about. 00:18:38.490 --> 00:18:41.530 align:middle line:84% So when the target appears right at the fovea, 00:18:41.530 --> 00:18:43.448 align:middle line:84% you're worse than when it appears 00:18:43.448 --> 00:18:44.990 align:middle line:84% at these intermediate eccentricities. 00:18:44.990 --> 00:18:47.073 align:middle line:84% And then when it gets too far out in the periphery 00:18:47.073 --> 00:18:48.330 align:middle line:90% you're again worse. 00:18:48.330 --> 00:18:55.150 align:middle line:84% So this kind of unusual trait of the texture segregation task. 00:18:55.150 --> 00:18:58.250 align:middle line:84% So now the question is, how is that curve 00:18:58.250 --> 00:19:01.990 align:middle line:84% altered by being cued to the location of the target? 00:19:01.990 --> 00:19:04.030 align:middle line:84% And so that's what the red curve shows you. 00:19:04.030 --> 00:19:07.370 align:middle line:84% So the red squares here are showing you 00:19:07.370 --> 00:19:09.370 align:middle line:84% performance on this texture segregation task 00:19:09.370 --> 00:19:11.370 align:middle line:84% is a function of eccentricity when you're 00:19:11.370 --> 00:19:14.460 align:middle line:90% cued to the target's location. 00:19:14.460 --> 00:19:18.810 align:middle line:84% Now the first thing to note here is this. 00:19:18.810 --> 00:19:21.130 align:middle line:84% So in this particular case, you're 00:19:21.130 --> 00:19:22.470 align:middle line:90% being cued to the location. 00:19:22.470 --> 00:19:24.730 align:middle line:90% And the cue makes you worse. 00:19:24.730 --> 00:19:26.450 align:middle line:90% So that's pretty weird. 00:19:26.450 --> 00:19:27.850 align:middle line:84% So normally we think of attention 00:19:27.850 --> 00:19:29.810 align:middle line:90% as actually making you better. 00:19:29.810 --> 00:19:32.850 align:middle line:84% But this is a case where the cue pops up, 00:19:32.850 --> 00:19:37.280 align:middle line:84% and that actually impairs your performance. 00:19:37.280 --> 00:19:39.880 align:middle line:84% But the other thing that you can see from this 00:19:39.880 --> 00:19:43.760 align:middle line:84% is that if you look at the overall shape of the red curve, 00:19:43.760 --> 00:19:47.840 align:middle line:84% like the eccentricity at which your performance is the best 00:19:47.840 --> 00:19:52.720 align:middle line:84% has actually shifted a little bit towards the periphery. 00:19:52.720 --> 00:19:54.980 align:middle line:84% This is real data, so it's kind of jaggedy. 00:19:54.980 --> 00:19:57.980 align:middle line:84% But if you think about smoothing out those two curves, 00:19:57.980 --> 00:20:01.840 align:middle line:84% the peak of the red curve is more eccentric than the peak 00:20:01.840 --> 00:20:05.660 align:middle line:90% of the white curve. 00:20:05.660 --> 00:20:11.600 align:middle line:84% And that's consistent with the idea that when that cue pops up, 00:20:11.600 --> 00:20:14.040 align:middle line:84% the receptive fields get a little bit smaller. 00:20:14.040 --> 00:20:18.120 align:middle line:84% And as a consequence of that, the eccentricity 00:20:18.120 --> 00:20:21.760 align:middle line:84% at which the receptive fields are best matched to the texture 00:20:21.760 --> 00:20:29.720 align:middle line:84% moves out to the periphery a little bit, so pretty wild. 00:20:29.720 --> 00:20:32.300 align:middle line:84% So what's up with the other two columns? 00:20:32.300 --> 00:20:34.763 align:middle line:90% 00:20:34.763 --> 00:20:36.430 align:middle line:84% The other two columns are a manipulation 00:20:36.430 --> 00:20:38.070 align:middle line:90% of the viewing distance. 00:20:38.070 --> 00:20:41.590 align:middle line:84% And the idea behind that is that when 00:20:41.590 --> 00:20:45.110 align:middle line:84% you move closer to the display or further to the display, 00:20:45.110 --> 00:20:49.270 align:middle line:84% the size of the texture elements on the retina is changing. 00:20:49.270 --> 00:20:51.710 align:middle line:84% And the consequence of that is that one 00:20:51.710 --> 00:20:54.910 align:middle line:84% would predict that the eccentricity at which 00:20:54.910 --> 00:20:58.750 align:middle line:84% the receptive fields will be optimally matched to the texture 00:20:58.750 --> 00:21:00.510 align:middle line:90% is also going to change. 00:21:00.510 --> 00:21:03.770 align:middle line:84% And that's more or less what happens. 00:21:03.770 --> 00:21:08.430 align:middle line:84% So you can see that when you're close and you're far, 00:21:08.430 --> 00:21:13.350 align:middle line:84% those curves have different peaks But in all of the cases, 00:21:13.350 --> 00:21:15.550 align:middle line:84% in all three of those viewing distances, 00:21:15.550 --> 00:21:18.310 align:middle line:84% you see the same kind of characteristic, which 00:21:18.310 --> 00:21:22.710 align:middle line:84% is that close to the fovea being cued to the location 00:21:22.710 --> 00:21:24.630 align:middle line:90% makes your performance worse. 00:21:24.630 --> 00:21:29.330 align:middle line:84% And the overall peak of the curve 00:21:29.330 --> 00:21:31.550 align:middle line:84% for the red condition where you're cued 00:21:31.550 --> 00:21:33.400 align:middle line:90% shifts out to the periphery. 00:21:33.400 --> 00:21:36.380 align:middle line:90% 00:21:36.380 --> 00:21:36.880 align:middle line:90% All right. 00:21:36.880 --> 00:21:39.700 align:middle line:84% So this is consistent with the idea 00:21:39.700 --> 00:21:45.220 align:middle line:84% that attention is altering the receptive fields, effectively 00:21:45.220 --> 00:21:49.580 align:middle line:84% shrinking them, which maybe in general, like for lots 00:21:49.580 --> 00:21:51.540 align:middle line:90% of things, might be helpful. 00:21:51.540 --> 00:21:54.180 align:middle line:84% In some sense, it improves spatial resolution, maybe. 00:21:54.180 --> 00:21:57.060 align:middle line:84% But in this particular texture segregation task, 00:21:57.060 --> 00:21:59.140 align:middle line:84% this leads to performance decrements 00:21:59.140 --> 00:22:01.760 align:middle line:84% when the filters are already too small for the texture, 00:22:01.760 --> 00:22:02.920 align:middle line:90% namely at the fovea. 00:22:02.920 --> 00:22:04.060 align:middle line:90% OK? 00:22:04.060 --> 00:22:06.900 align:middle line:84% So the texture segregation task is being used here 00:22:06.900 --> 00:22:10.460 align:middle line:84% as a diagnostic to understand what's 00:22:10.460 --> 00:22:13.920 align:middle line:84% changing in the visual system when you get cued, 00:22:13.920 --> 00:22:16.420 align:middle line:84% when your attention is drawn to a particular location. 00:22:16.420 --> 00:22:21.140 align:middle line:84% And so this is a really unusual result, 00:22:21.140 --> 00:22:24.390 align:middle line:90% both because it's one of a-- 00:22:24.390 --> 00:22:26.200 align:middle line:84% I can't think of any other cases. 00:22:26.200 --> 00:22:28.820 align:middle line:84% There's probably some, but I can't think of other cases 00:22:28.820 --> 00:22:31.020 align:middle line:84% where attention actually makes performance 00:22:31.020 --> 00:22:32.630 align:middle line:90% worse under some conditions. 00:22:32.630 --> 00:22:36.250 align:middle line:84% So that's really unusual and interesting. 00:22:36.250 --> 00:22:40.090 align:middle line:84% And then also it's a notable result in the sense 00:22:40.090 --> 00:22:47.570 align:middle line:84% that it provides some suggestion of how attention is actually 00:22:47.570 --> 00:22:49.150 align:middle line:90% altering visual processing. 00:22:49.150 --> 00:22:51.970 align:middle line:90% 00:22:51.970 --> 00:22:52.630 align:middle line:90% All right. 00:22:52.630 --> 00:22:54.728 align:middle line:90% So this is another experiment. 00:22:54.728 --> 00:22:56.270 align:middle line:84% This is actually from the same group. 00:22:56.270 --> 00:22:58.930 align:middle line:84% This is the lab of Marisa Carrasco, 00:22:58.930 --> 00:23:03.170 align:middle line:84% who is an expert on visual attention at NYU. 00:23:03.170 --> 00:23:06.970 align:middle line:84% And this was an attempt to ask, again, 00:23:06.970 --> 00:23:09.370 align:middle line:84% something about the extent to which attention 00:23:09.370 --> 00:23:11.290 align:middle line:84% changes the representation of things 00:23:11.290 --> 00:23:14.810 align:middle line:84% visually, but in this case, in a way that might really 00:23:14.810 --> 00:23:16.630 align:middle line:90% affect subjective appearance. 00:23:16.630 --> 00:23:21.210 align:middle line:84% And again, it has its roots in speculation 00:23:21.210 --> 00:23:23.330 align:middle line:84% from over 100 years ago where people 00:23:23.330 --> 00:23:26.190 align:middle line:84% would observe that when paying attention to something, 00:23:26.190 --> 00:23:28.280 align:middle line:84% there was a sense in which things look clearer. 00:23:28.280 --> 00:23:30.280 align:middle line:84% The question is, what does it mean for something 00:23:30.280 --> 00:23:31.180 align:middle line:90% to look clearer? 00:23:31.180 --> 00:23:34.320 align:middle line:84% And so here they're asking whether attention 00:23:34.320 --> 00:23:37.360 align:middle line:84% has the effect of altering perceived 00:23:37.360 --> 00:23:39.720 align:middle line:84% contrast, in other words, making things 00:23:39.720 --> 00:23:43.080 align:middle line:84% appear to be higher contrast than if you weren't 00:23:43.080 --> 00:23:44.460 align:middle line:90% paying attention to them. 00:23:44.460 --> 00:23:48.900 align:middle line:84% So again, they have a clever design to get at this. 00:23:48.900 --> 00:23:53.920 align:middle line:84% So here's the sequence of things that happens on a single trial. 00:23:53.920 --> 00:23:56.360 align:middle line:84% So you start out, you're fixating. 00:23:56.360 --> 00:23:57.760 align:middle line:90% OK. 00:23:57.760 --> 00:24:00.775 align:middle line:90% And then there's a cue. 00:24:00.775 --> 00:24:02.400 align:middle line:84% And the cue can either be neutral which 00:24:02.400 --> 00:24:05.800 align:middle line:84% means that it's at fixation, or peripheral, 00:24:05.800 --> 00:24:07.640 align:middle line:84% meaning that it will be at a location 00:24:07.640 --> 00:24:09.160 align:middle line:90% of a subsequent stimulus. 00:24:09.160 --> 00:24:12.620 align:middle line:84% So then you get the stimulus display. 00:24:12.620 --> 00:24:16.900 align:middle line:84% And the stimulus display consists of two gratings here. 00:24:16.900 --> 00:24:21.260 align:middle line:84% And so the task here is to judge the orientation of the grating 00:24:21.260 --> 00:24:23.360 align:middle line:90% that is higher contrast. 00:24:23.360 --> 00:24:26.480 align:middle line:84% And then you press a key to indicate 00:24:26.480 --> 00:24:28.480 align:middle line:90% what the orientation is. 00:24:28.480 --> 00:24:31.670 align:middle line:84% So here they're not actually interested in the judgments 00:24:31.670 --> 00:24:32.450 align:middle line:90% of orientation. 00:24:32.450 --> 00:24:34.590 align:middle line:84% The judgments of orientation is just 00:24:34.590 --> 00:24:37.790 align:middle line:84% a way to get people to report on the thing that's 00:24:37.790 --> 00:24:40.710 align:middle line:84% higher contrast without explicitly getting them 00:24:40.710 --> 00:24:42.570 align:middle line:90% to make a contrast judgment. 00:24:42.570 --> 00:24:47.110 align:middle line:84% So in this case, one grating is fixed in contrast. 00:24:47.110 --> 00:24:50.310 align:middle line:84% That's going to be called the standard And the other is 00:24:50.310 --> 00:24:53.950 align:middle line:90% varied across trials 00:24:53.950 --> 00:24:54.450 align:middle line:90% All right. 00:24:54.450 --> 00:24:56.190 align:middle line:84% And so what they're interested in 00:24:56.190 --> 00:25:00.030 align:middle line:84% is whether being cued to the location 00:25:00.030 --> 00:25:05.030 align:middle line:84% here will make people more likely to choose 00:25:05.030 --> 00:25:08.230 align:middle line:84% the thing that's not actually higher in contrast. 00:25:08.230 --> 00:25:12.870 align:middle line:84% So the idea is that when you're paying attention to something, 00:25:12.870 --> 00:25:15.990 align:middle line:84% the perceived contrast gets boosted, 00:25:15.990 --> 00:25:19.150 align:middle line:84% then you're going to erroneously potentially choose 00:25:19.150 --> 00:25:22.810 align:middle line:84% the one that is potentially actually lower contrast. 00:25:22.810 --> 00:25:24.070 align:middle line:90% So here's the results, guys. 00:25:24.070 --> 00:25:25.070 align:middle line:90% Let's walk through this. 00:25:25.070 --> 00:25:25.570 align:middle line:90% OK. 00:25:25.570 --> 00:25:28.260 align:middle line:84% So the y-axis here, so just focus 00:25:28.260 --> 00:25:29.600 align:middle line:90% like on this particular panel. 00:25:29.600 --> 00:25:32.340 align:middle line:84% So the y-axis here is plotting the proportion 00:25:32.340 --> 00:25:37.860 align:middle line:84% of trials on which people report that the contrast that they 00:25:37.860 --> 00:25:41.660 align:middle line:84% perceive of the test is greater than the standard. 00:25:41.660 --> 00:25:43.820 align:middle line:84% So the standard is the one that's always fixed 00:25:43.820 --> 00:25:45.580 align:middle line:90% and the test is varied. 00:25:45.580 --> 00:25:48.660 align:middle line:84% And the x-axis here is the contrast of the test. 00:25:48.660 --> 00:25:50.240 align:middle line:90% So what do we expect? 00:25:50.240 --> 00:25:53.700 align:middle line:84% Well, we expect that when the contrast of the test 00:25:53.700 --> 00:25:57.680 align:middle line:84% is really low, people will always choose the standard. 00:25:57.680 --> 00:26:00.900 align:middle line:84% And so this should be close to 0. 00:26:00.900 --> 00:26:03.420 align:middle line:84% As the contrast of the test increases, 00:26:03.420 --> 00:26:06.175 align:middle line:84% well, once it becomes higher contrast than the standard, 00:26:06.175 --> 00:26:08.300 align:middle line:84% you would expect that people would choose the test. 00:26:08.300 --> 00:26:10.100 align:middle line:90% And so you end up at 100%. 00:26:10.100 --> 00:26:12.540 align:middle line:84% So you get this psychometric function that's 00:26:12.540 --> 00:26:16.300 align:middle line:90% going to go from 0 up to 100. 00:26:16.300 --> 00:26:20.540 align:middle line:84% Now the point at which the curves 00:26:20.540 --> 00:26:23.460 align:middle line:84% have a value of 50% on the y-axis 00:26:23.460 --> 00:26:27.730 align:middle line:84% is what we're going to call the point of subjective equality. 00:26:27.730 --> 00:26:32.730 align:middle line:84% That is the contrast of the test stimulus 00:26:32.730 --> 00:26:37.210 align:middle line:84% that effectively looks the same as the contrast of the standard. 00:26:37.210 --> 00:26:42.330 align:middle line:84% And so we expect that when the cue is neutral, 00:26:42.330 --> 00:26:45.730 align:middle line:84% the point of subjective equality should be actual equality. 00:26:45.730 --> 00:26:48.370 align:middle line:84% It's just when the test is going to be the same contrast 00:26:48.370 --> 00:26:49.150 align:middle line:90% as the standard. 00:26:49.150 --> 00:26:53.210 align:middle line:84% So the black curve is what happens under the neutral cuing 00:26:53.210 --> 00:26:53.910 align:middle line:90% condition. 00:26:53.910 --> 00:26:56.970 align:middle line:84% And so you get some point of subjective equality 00:26:56.970 --> 00:27:01.770 align:middle line:84% that is very close to the actual contrast of the standard. 00:27:01.770 --> 00:27:04.210 align:middle line:84% Now the interesting thing is what 00:27:04.210 --> 00:27:06.230 align:middle line:84% will happen for the blue and the red curves. 00:27:06.230 --> 00:27:09.790 align:middle line:84% So the blue curve is where the test location is cued. 00:27:09.790 --> 00:27:12.950 align:middle line:84% So that little dot pops up at the location of the test. 00:27:12.950 --> 00:27:16.010 align:middle line:84% And the red curve is where the standard is cued. 00:27:16.010 --> 00:27:20.090 align:middle line:84% The dot pops up at the location of the standard. 00:27:20.090 --> 00:27:21.150 align:middle line:90% That's my advisee. 00:27:21.150 --> 00:27:23.770 align:middle line:90% 00:27:23.770 --> 00:27:24.910 align:middle line:90% So what happens here? 00:27:24.910 --> 00:27:27.220 align:middle line:84% So let's look at the point of subjective equality. 00:27:27.220 --> 00:27:28.780 align:middle line:84% So that's where these curves hit 50%. 00:27:28.780 --> 00:27:31.920 align:middle line:84% So for the blue curve the point of subjective equality 00:27:31.920 --> 00:27:34.580 align:middle line:90% is a lower contrast, right? 00:27:34.580 --> 00:27:37.540 align:middle line:90% So that's when the test is cued. 00:27:37.540 --> 00:27:40.080 align:middle line:84% And so the point of this is that when you cue somebody 00:27:40.080 --> 00:27:42.840 align:middle line:84% to the location of the test, it can 00:27:42.840 --> 00:27:47.220 align:middle line:84% be a little bit lower contrast and look equal to the standard. 00:27:47.220 --> 00:27:52.200 align:middle line:84% And when in fact, they're equal in contrast. 00:27:52.200 --> 00:27:56.080 align:middle line:84% When it's cued, people are more likely to actually say 00:27:56.080 --> 00:27:59.960 align:middle line:84% that the test is higher in contrast than the standard. 00:27:59.960 --> 00:28:02.220 align:middle line:84% Whereas when the standard is cued, 00:28:02.220 --> 00:28:04.980 align:middle line:84% the point of subjective equality is shifted to something greater. 00:28:04.980 --> 00:28:06.480 align:middle line:84% So that means you've got to crank up 00:28:06.480 --> 00:28:08.080 align:middle line:84% the contrast of the test a little bit 00:28:08.080 --> 00:28:10.580 align:middle line:84% to get it to look the same as the standard. 00:28:10.580 --> 00:28:13.200 align:middle line:90% 00:28:13.200 --> 00:28:13.700 align:middle line:90% All right. 00:28:13.700 --> 00:28:17.040 align:middle line:84% So the inference here is that when your attention is 00:28:17.040 --> 00:28:20.360 align:middle line:84% involuntarily drawn to some location, 00:28:20.360 --> 00:28:23.280 align:middle line:84% the perceived contrast at the location 00:28:23.280 --> 00:28:26.350 align:middle line:84% is a little bit higher, which might 00:28:26.350 --> 00:28:29.190 align:middle line:84% have the consequence of making something look clearer, 00:28:29.190 --> 00:28:33.107 align:middle line:90% or crisper, easier to see. 00:28:33.107 --> 00:28:35.190 align:middle line:84% This other graph here is just the exact same thing 00:28:35.190 --> 00:28:37.010 align:middle line:84% at a different spatial frequency. 00:28:37.010 --> 00:28:40.630 align:middle line:84% So CPD is cycles per degree, so this is essentially 00:28:40.630 --> 00:28:42.470 align:middle line:90% just a replication. 00:28:42.470 --> 00:28:46.190 align:middle line:84% And then the bottom curves here are a control experiment 00:28:46.190 --> 00:28:50.870 align:middle line:84% where this interstimulus interval, or ISI, so 00:28:50.870 --> 00:28:55.750 align:middle line:84% that's the time interval between the cue and the stimulus. 00:28:55.750 --> 00:28:58.690 align:middle line:84% So in the graph that we were just looking at, 00:28:58.690 --> 00:29:01.250 align:middle line:84% that was pretty short, like 50 milliseconds. 00:29:01.250 --> 00:29:02.750 align:middle line:84% And in this control experiment, it's 00:29:02.750 --> 00:29:04.870 align:middle line:90% extended to 500 milliseconds. 00:29:04.870 --> 00:29:07.670 align:middle line:84% And there you can see the effect goes away. 00:29:07.670 --> 00:29:09.830 align:middle line:84% And so what that means is that whatever 00:29:09.830 --> 00:29:14.110 align:middle line:84% happens when that attentional cue pops up there, is transient. 00:29:14.110 --> 00:29:18.710 align:middle line:84% So you get a temporary change to visual processing. 00:29:18.710 --> 00:29:20.290 align:middle line:90% And then it goes away. 00:29:20.290 --> 00:29:24.220 align:middle line:90% 00:29:24.220 --> 00:29:26.680 align:middle line:84% OK, so this is a pretty cool result. 00:29:26.680 --> 00:29:29.860 align:middle line:84% It's confirming this intuition that when you're 00:29:29.860 --> 00:29:31.500 align:middle line:84% paying attention to something, things 00:29:31.500 --> 00:29:36.260 align:middle line:84% are kind of easier to see, and suggesting 00:29:36.260 --> 00:29:39.700 align:middle line:84% that one correlate of that or one consequence of that 00:29:39.700 --> 00:29:44.220 align:middle line:84% is that things look higher contrast. 00:29:44.220 --> 00:29:45.880 align:middle line:84% What questions you got about that? 00:29:45.880 --> 00:29:53.540 align:middle line:90% 00:29:53.540 --> 00:29:54.340 align:middle line:90% Yeah? 00:29:54.340 --> 00:29:56.940 align:middle line:84% AUDIENCE: I'm still unsure what the cycles per degree part. 00:29:56.940 --> 00:29:57.960 align:middle line:84% JOSH MCDERMOTT: Just say it louder? 00:29:57.960 --> 00:29:58.580 align:middle line:90% Sorry. 00:29:58.580 --> 00:30:01.038 align:middle line:84% AUDIENCE: I'm just unsure about the cycles per degree part. 00:30:01.038 --> 00:30:03.900 align:middle line:84% Is that how many times they show a certain orientation? 00:30:03.900 --> 00:30:05.900 align:middle line:84% JOSH MCDERMOTT: It's just the spatial frequency 00:30:05.900 --> 00:30:07.380 align:middle line:90% of the grating. 00:30:07.380 --> 00:30:11.000 align:middle line:84% So remember, you measure spatial frequency in cycles per degree. 00:30:11.000 --> 00:30:15.500 align:middle line:84% So it's just how many cycles of the grating 00:30:15.500 --> 00:30:16.640 align:middle line:90% happen in one degree. 00:30:16.640 --> 00:30:19.080 align:middle line:84% And so you just have to set that to some value. 00:30:19.080 --> 00:30:21.270 align:middle line:84% And they just tried two different values. 00:30:21.270 --> 00:30:23.850 align:middle line:84% It's just a replication to show that the effect seems 00:30:23.850 --> 00:30:26.090 align:middle line:90% to be somewhat general. 00:30:26.090 --> 00:30:28.190 align:middle line:84% So it's just a property of the stimulus. 00:30:28.190 --> 00:30:34.490 align:middle line:90% 00:30:34.490 --> 00:30:34.990 align:middle line:90% Yeah? 00:30:34.990 --> 00:30:37.032 align:middle line:84% AUDIENCE: I'm sorry, you might have covered this, 00:30:37.032 --> 00:30:39.490 align:middle line:84% but is there an explanation for why [INAUDIBLE] any reason 00:30:39.490 --> 00:30:41.330 align:middle line:90% to have more variation? 00:30:41.330 --> 00:30:44.290 align:middle line:84% JOSH MCDERMOTT: Oh, no, I don't think so. 00:30:44.290 --> 00:30:47.588 align:middle line:84% Yeah, I look at those graphs and I say, well, 00:30:47.588 --> 00:30:49.630 align:middle line:84% there's qualitatively the same effect at the two. 00:30:49.630 --> 00:30:52.570 align:middle line:84% But yeah, it's true that it's bigger in one than the other. 00:30:52.570 --> 00:30:55.490 align:middle line:84% No, I think there's an explanation for that. 00:30:55.490 --> 00:30:57.250 align:middle line:90% Yeah. 00:30:57.250 --> 00:31:01.570 align:middle line:84% So can anybody tell me why did they design 00:31:01.570 --> 00:31:03.330 align:middle line:90% this the way that they did? 00:31:03.330 --> 00:31:08.570 align:middle line:84% So what's the point of having people do this orientation task? 00:31:08.570 --> 00:31:10.890 align:middle line:84% So if really what you're interested in 00:31:10.890 --> 00:31:14.050 align:middle line:84% is whether the cue is going to alter perceived contrast, 00:31:14.050 --> 00:31:17.690 align:middle line:84% why not just directly ask people about the contrast? 00:31:17.690 --> 00:31:20.160 align:middle line:90% Can anybody think of why? 00:31:20.160 --> 00:31:21.000 align:middle line:90% Yeah? 00:31:21.000 --> 00:31:24.240 align:middle line:84% AUDIENCE: Is it because it tests the more natural and unconscious 00:31:24.240 --> 00:31:27.520 align:middle line:84% way we view the world as opposed to focusing on it? 00:31:27.520 --> 00:31:30.847 align:middle line:84% Is the idea that, if we tell them to focus on contrast, 00:31:30.847 --> 00:31:32.680 align:middle line:84% it will skew the results to something that's 00:31:32.680 --> 00:31:36.320 align:middle line:90% more synthetic than natural? 00:31:36.320 --> 00:31:38.460 align:middle line:84% JOSH MCDERMOTT: Well, I mean, maybe in a sense. 00:31:38.460 --> 00:31:42.040 align:middle line:90% 00:31:42.040 --> 00:31:45.620 align:middle line:84% What about anybody else, got any other ideas? 00:31:45.620 --> 00:31:46.200 align:middle line:90% Yeah? 00:31:46.200 --> 00:31:50.640 align:middle line:84% AUDIENCE: Is it because contrast can be like an ambiguous term, 00:31:50.640 --> 00:31:53.548 align:middle line:84% people will slightly interpret in different Ways 00:31:53.548 --> 00:31:55.840 align:middle line:84% JOSH MCDERMOTT: Well, but they are asking them about it 00:31:55.840 --> 00:31:58.600 align:middle line:84% in the sense that the task is to report 00:31:58.600 --> 00:32:01.340 align:middle line:84% the orientation of the grating that's higher contrast. 00:32:01.340 --> 00:32:04.800 align:middle line:84% So I presume they're showing people 00:32:04.800 --> 00:32:06.540 align:middle line:84% some examples of high and low contrast, 00:32:06.540 --> 00:32:08.340 align:middle line:84% just so they get the hang of that. 00:32:08.340 --> 00:32:10.360 align:middle line:84% So hopefully that's not an issue. 00:32:10.360 --> 00:32:12.720 align:middle line:84% But I think there's another thing that could 00:32:12.720 --> 00:32:14.140 align:middle line:90% be a reason why they did this. 00:32:14.140 --> 00:32:15.500 align:middle line:90% Anybody else have another idea? 00:32:15.500 --> 00:32:19.510 align:middle line:90% 00:32:19.510 --> 00:32:21.390 align:middle line:84% So I think one reason to do this is 00:32:21.390 --> 00:32:23.430 align:middle line:84% that you if the main point of the task 00:32:23.430 --> 00:32:26.690 align:middle line:84% is to actually discriminate the contrast of these two things, 00:32:26.690 --> 00:32:29.670 align:middle line:84% people are going to pay attention to both of them. 00:32:29.670 --> 00:32:31.715 align:middle line:84% So you'll get this cue, but then people 00:32:31.715 --> 00:32:34.090 align:middle line:84% will be dividing their attention to look at both of them. 00:32:34.090 --> 00:32:36.530 align:middle line:84% And presumably that might weaken the effect. 00:32:36.530 --> 00:32:39.190 align:middle line:84% So I think the reason to actually just have them 00:32:39.190 --> 00:32:43.810 align:middle line:84% do the orientation task is to mitigate that to some extent. 00:32:43.810 --> 00:32:46.372 align:middle line:90% Right? 00:32:46.372 --> 00:32:47.830 align:middle line:84% So this is one of the things that's 00:32:47.830 --> 00:32:52.630 align:middle line:84% always tricky about studying attention is that it's 00:32:52.630 --> 00:32:56.750 align:middle line:84% very difficult to query the nature of what you're not 00:32:56.750 --> 00:32:59.830 align:middle line:84% attending to behaviorally, because if you ask somebody 00:32:59.830 --> 00:33:02.510 align:middle line:84% about something, then they pay attention to it, right? 00:33:02.510 --> 00:33:07.950 align:middle line:84% So yeah, I think that's the logic behind the design. 00:33:07.950 --> 00:33:09.110 align:middle line:90% OK. 00:33:09.110 --> 00:33:12.670 align:middle line:84% This is a pretty interesting effect 00:33:12.670 --> 00:33:16.150 align:middle line:84% where attention appears to alter the perception of time. 00:33:16.150 --> 00:33:18.120 align:middle line:84% So I'm going to play you this movie. 00:33:18.120 --> 00:33:21.680 align:middle line:84% And it's going to show a bunch of stuff on the screen. 00:33:21.680 --> 00:33:23.460 align:middle line:90% Things are going to flash up. 00:33:23.460 --> 00:33:25.360 align:middle line:84% Everything's going to be the same duration. 00:33:25.360 --> 00:33:28.980 align:middle line:84% But you're going to notice that there'll 00:33:28.980 --> 00:33:31.458 align:middle line:84% be one thing that appears to not be 00:33:31.458 --> 00:33:33.000 align:middle line:84% the same duration as everything else. 00:33:33.000 --> 00:33:35.352 align:middle line:90% And that's an illusion. 00:33:35.352 --> 00:33:38.304 align:middle line:90% [VIDEO PLAYBACK] 00:33:38.304 --> 00:33:51.096 align:middle line:90% 00:33:51.096 --> 00:33:53.580 align:middle line:90% [END PLAYBACK] 00:33:53.580 --> 00:33:58.840 align:middle line:84% So each of the disks is up for the same amount of time. 00:33:58.840 --> 00:34:01.755 align:middle line:90% 00:34:01.755 --> 00:34:06.180 align:middle line:84% And the one that expands is also up for the same amount of time. 00:34:06.180 --> 00:34:09.842 align:middle line:84% But people normally say that it looks like maybe 50% longer 00:34:09.842 --> 00:34:10.800 align:middle line:90% or something like that. 00:34:10.800 --> 00:34:13.380 align:middle line:90% Right? 00:34:13.380 --> 00:34:15.610 align:middle line:84% So you can actually measure this. 00:34:15.610 --> 00:34:19.958 align:middle line:84% So you can do experiments with this, 00:34:19.958 --> 00:34:22.250 align:middle line:84% where you vary the relative durations of the thing that 00:34:22.250 --> 00:34:26.465 align:middle line:84% expands and the static ones, and find the duration at which 00:34:26.465 --> 00:34:27.590 align:middle line:90% things look to be the same. 00:34:27.590 --> 00:34:31.170 align:middle line:84% And then you can calculate what they call a temporal expansion 00:34:31.170 --> 00:34:31.909 align:middle line:90% factor. 00:34:31.909 --> 00:34:37.690 align:middle line:84% So 1.5 means that the expanding thing looks like it's about 50% 00:34:37.690 --> 00:34:40.090 align:middle line:90% longer than everything else. 00:34:40.090 --> 00:34:42.870 align:middle line:84% And so that was true across a range of durations here. 00:34:42.870 --> 00:34:44.929 align:middle line:90% That's the thing circled in red. 00:34:44.929 --> 00:34:48.690 align:middle line:84% So the oddball seems about 50% longer. 00:34:48.690 --> 00:34:52.929 align:middle line:84% And there is this weird thing here 00:34:52.929 --> 00:34:55.489 align:middle line:84% that is unexplained that at very short durations, 00:34:55.489 --> 00:34:57.897 align:middle line:84% you get what looks like an opposite effect. 00:34:57.897 --> 00:34:59.730 align:middle line:84% I don't know what the explanation is of that 00:34:59.730 --> 00:35:02.130 align:middle line:90% and neither do the authors. 00:35:02.130 --> 00:35:04.370 align:middle line:84% But this is not about things that 00:35:04.370 --> 00:35:05.910 align:middle line:90% are expanding seeming longer. 00:35:05.910 --> 00:35:08.390 align:middle line:84% So if you do the kind of opposite design, 00:35:08.390 --> 00:35:11.590 align:middle line:84% so you have a series of things that expand, 00:35:11.590 --> 00:35:14.370 align:middle line:84% and then the oddball is something that doesn't expand, 00:35:14.370 --> 00:35:18.600 align:middle line:84% you get about the same effect, maybe a little bit smaller. 00:35:18.600 --> 00:35:22.240 align:middle line:84% But again, it's the thing that is unusual or distinct 00:35:22.240 --> 00:35:25.200 align:middle line:84% in the sequence that ends up looking longer. 00:35:25.200 --> 00:35:30.080 align:middle line:84% And so the presumptive explanation of this 00:35:30.080 --> 00:35:34.920 align:middle line:84% is that the thing that is unusual or deviant 00:35:34.920 --> 00:35:36.840 align:middle line:90% attracts attention. 00:35:36.840 --> 00:35:40.900 align:middle line:84% And that causes the thing to appear longer. 00:35:40.900 --> 00:35:45.823 align:middle line:84% Now is the attention that gets attracted to the eyeballs, 00:35:45.823 --> 00:35:47.240 align:middle line:84% is that the same kind of attention 00:35:47.240 --> 00:35:48.980 align:middle line:84% that we've been studying all these other experiments? 00:35:48.980 --> 00:35:50.120 align:middle line:90% That's totally unclear. 00:35:50.120 --> 00:35:52.007 align:middle line:84% So we always use the word attention 00:35:52.007 --> 00:35:53.340 align:middle line:90% to apply to all of these things. 00:35:53.340 --> 00:35:54.715 align:middle line:84% And again, it's very much unclear 00:35:54.715 --> 00:35:57.300 align:middle line:84% whether we're talking about the same thing across experiments. 00:35:57.300 --> 00:36:00.600 align:middle line:84% But it's a phenomena that is often described 00:36:00.600 --> 00:36:03.160 align:middle line:90% in terms of attention. 00:36:03.160 --> 00:36:05.120 align:middle line:84% Similar things also happen with sound. 00:36:05.120 --> 00:36:09.240 align:middle line:84% Maybe it's a little bit weaker, but the lower line here 00:36:09.240 --> 00:36:12.580 align:middle line:84% is the same kind of effect with sound, where in this case, 00:36:12.580 --> 00:36:16.190 align:middle line:84% I'm pretty sure I think these were beeps that occasionally 00:36:16.190 --> 00:36:17.450 align:middle line:90% changed in frequency. 00:36:17.450 --> 00:36:20.630 align:middle line:90% 00:36:20.630 --> 00:36:27.030 align:middle line:84% So some of you may have been in a car accident at some point, 00:36:27.030 --> 00:36:31.550 align:middle line:84% or encountered some other thing that caused things 00:36:31.550 --> 00:36:34.250 align:middle line:84% to seem like they were slower for some brief period of time. 00:36:34.250 --> 00:36:36.150 align:middle line:90% It's very commonly reported. 00:36:36.150 --> 00:36:38.230 align:middle line:84% People will say that in these situations 00:36:38.230 --> 00:36:40.590 align:middle line:84% where they're like highly aroused, 00:36:40.590 --> 00:36:43.750 align:middle line:84% that the experience of time seems to slow down. 00:36:43.750 --> 00:36:48.270 align:middle line:84% And so this phenomenon is believed to potentially 00:36:48.270 --> 00:36:50.230 align:middle line:84% be related to that kind of everyday experience 00:36:50.230 --> 00:36:51.890 align:middle line:90% that people sometimes report. 00:36:51.890 --> 00:36:54.170 align:middle line:84% So when your attention is drawn to something, 00:36:54.170 --> 00:36:55.490 align:middle line:90% it seems to last longer. 00:36:55.490 --> 00:36:58.670 align:middle line:84% So it's an effect that's been documented. 00:36:58.670 --> 00:37:02.230 align:middle line:84% We don't really have a great understanding of what happens 00:37:02.230 --> 00:37:02.810 align:middle line:90% or why. 00:37:02.810 --> 00:37:04.370 align:middle line:90% There's lots of open questions. 00:37:04.370 --> 00:37:07.510 align:middle line:90% 00:37:07.510 --> 00:37:09.470 align:middle line:84% Does the fact that things seem longer mean 00:37:09.470 --> 00:37:12.990 align:middle line:84% that you're doing more computation during that time? 00:37:12.990 --> 00:37:16.210 align:middle line:84% Is the sense of time actually maybe related to some kind 00:37:16.210 --> 00:37:17.030 align:middle line:90% of mental work? 00:37:17.030 --> 00:37:19.527 align:middle line:90% Is time supermodel? 00:37:19.527 --> 00:37:20.110 align:middle line:90% We don't know. 00:37:20.110 --> 00:37:21.930 align:middle line:84% These are all kind of interesting things 00:37:21.930 --> 00:37:26.570 align:middle line:84% that would be fun to explore in the future. 00:37:26.570 --> 00:37:31.810 align:middle line:84% You can also select things with your attention in time. 00:37:31.810 --> 00:37:34.610 align:middle line:84% So you can be paying attention to something 00:37:34.610 --> 00:37:40.050 align:middle line:84% and have a sequence of stimuli occur 00:37:40.050 --> 00:37:43.870 align:middle line:84% and to have to pay attention to one of them. 00:37:43.870 --> 00:37:46.090 align:middle line:84% And you can either have some temporal expectation 00:37:46.090 --> 00:37:47.810 align:middle line:84% of where that's going to be, or you 00:37:47.810 --> 00:37:49.755 align:middle line:84% can be looking for a particular kind of thing. 00:37:49.755 --> 00:37:51.130 align:middle line:84% And there's some very interesting 00:37:51.130 --> 00:37:53.970 align:middle line:84% and well-known and robust effects 00:37:53.970 --> 00:37:57.065 align:middle line:84% that characterize our attentional selection in time. 00:37:57.065 --> 00:37:59.190 align:middle line:84% And one of them is what I'm about to show you here. 00:37:59.190 --> 00:38:03.170 align:middle line:84% And this arises in a setting that's 00:38:03.170 --> 00:38:07.370 align:middle line:84% shown here, where there is a sequence of stimuli 00:38:07.370 --> 00:38:08.710 align:middle line:90% that pop up on the screen. 00:38:08.710 --> 00:38:14.280 align:middle line:84% In this case, they're letters And your job in this experiment 00:38:14.280 --> 00:38:18.880 align:middle line:84% is to look for white letters or X's. 00:38:18.880 --> 00:38:21.660 align:middle line:90% So here's a white letter. 00:38:21.660 --> 00:38:23.240 align:middle line:90% And you would press 1 key. 00:38:23.240 --> 00:38:26.000 align:middle line:84% And then here's an X And you would press a different key. 00:38:26.000 --> 00:38:32.720 align:middle line:84% And the effect that I'm going to show you 00:38:32.720 --> 00:38:38.120 align:middle line:84% relates to the effect of your having detected 00:38:38.120 --> 00:38:42.720 align:middle line:84% the white letter, which is known as T1, on your ability 00:38:42.720 --> 00:38:46.200 align:middle line:84% to detect the secondary target known as T2. 00:38:46.200 --> 00:38:50.280 align:middle line:84% And that turns out to be a function of how separated 00:38:50.280 --> 00:38:55.800 align:middle line:90% these things are in time 00:38:55.800 --> 00:38:56.300 align:middle line:90% All right. 00:38:56.300 --> 00:39:01.560 align:middle line:84% And so what this graph shows is the percentage 00:39:01.560 --> 00:39:04.240 align:middle line:84% of the second target that's detected, 00:39:04.240 --> 00:39:07.060 align:middle line:84% given that you correctly detected T1, 00:39:07.060 --> 00:39:09.180 align:middle line:90% which is that white letter OK? 00:39:09.180 --> 00:39:12.790 align:middle line:84% And it's plotted as a position of T2 00:39:12.790 --> 00:39:14.710 align:middle line:90% in the list relative to T1. 00:39:14.710 --> 00:39:19.030 align:middle line:84% So 1 here means that the X comes right after the white letter, 00:39:19.030 --> 00:39:22.450 align:middle line:84% 2 means that it comes two letters after, and so forth. 00:39:22.450 --> 00:39:24.210 align:middle line:90% So look at the red curve here. 00:39:24.210 --> 00:39:26.830 align:middle line:84% This is like what somebody like me would do. 00:39:26.830 --> 00:39:32.710 align:middle line:84% And what this is showing is that in the few positions following 00:39:32.710 --> 00:39:35.670 align:middle line:84% the white letter, people are impaired 00:39:35.670 --> 00:39:38.710 align:middle line:84% at detecting the black X. And this is called 00:39:38.710 --> 00:39:40.770 align:middle line:90% the attentional blink right. 00:39:40.770 --> 00:39:44.270 align:middle line:84% It's like there's a refractory period on your ability 00:39:44.270 --> 00:39:49.870 align:middle line:84% to select or detect stuff that you're looking for. 00:39:49.870 --> 00:39:52.590 align:middle line:84% Now this particular graph, it's got a second line 00:39:52.590 --> 00:39:55.590 align:middle line:84% on it, the blue curve, which is people that spend a lot of time 00:39:55.590 --> 00:39:56.490 align:middle line:90% playing video games. 00:39:56.490 --> 00:39:58.250 align:middle line:84% And I think maybe a lecture or two 00:39:58.250 --> 00:39:59.750 align:middle line:84% ago, somebody asked me some question 00:39:59.750 --> 00:40:05.310 align:middle line:84% and I mentioned how there's a body of work indicating 00:40:05.310 --> 00:40:07.940 align:middle line:84% that people that play lots of video games 00:40:07.940 --> 00:40:09.640 align:middle line:84% have slightly different visual systems, 00:40:09.640 --> 00:40:11.380 align:middle line:84% and in particular, their performance 00:40:11.380 --> 00:40:14.700 align:middle line:84% on a lot of visual attention tasks 00:40:14.700 --> 00:40:17.020 align:middle line:90% is different, usually better. 00:40:17.020 --> 00:40:19.580 align:middle line:84% So in this particular case, the attentional blink 00:40:19.580 --> 00:40:22.640 align:middle line:84% is reduced in people that play lots of video games, 00:40:22.640 --> 00:40:24.140 align:middle line:84% I guess because you have to do lots 00:40:24.140 --> 00:40:28.980 align:middle line:84% of detecting of things in rapid sequence 00:40:28.980 --> 00:40:31.900 align:middle line:90% when you play a video game. 00:40:31.900 --> 00:40:34.060 align:middle line:84% This is like the metaphor, again, 00:40:34.060 --> 00:40:38.020 align:middle line:84% we don't really have rigorous mechanistic accounts of this, 00:40:38.020 --> 00:40:40.480 align:middle line:84% but it has been very well documented. 00:40:40.480 --> 00:40:46.340 align:middle line:84% But this is the metaphor is that there's this first fish that's 00:40:46.340 --> 00:40:47.280 align:middle line:90% swimming by. 00:40:47.280 --> 00:40:48.800 align:middle line:90% And you've got to detect that. 00:40:48.800 --> 00:40:51.220 align:middle line:84% And that involves doing something to it. 00:40:51.220 --> 00:40:53.560 align:middle line:84% You grab it with your net and then you empty it. 00:40:53.560 --> 00:40:56.480 align:middle line:84% And it takes some time to empty the net. 00:40:56.480 --> 00:40:59.040 align:middle line:84% And during that time when you're emptying the net, 00:40:59.040 --> 00:41:00.440 align:middle line:90% the second fish comes along. 00:41:00.440 --> 00:41:03.380 align:middle line:84% And if the net hasn't been emptied yet, you don't grab it. 00:41:03.380 --> 00:41:04.940 align:middle line:84% So there's something that happens 00:41:04.940 --> 00:41:08.050 align:middle line:84% when you're detecting stuff, and it takes it a little while 00:41:08.050 --> 00:41:10.690 align:middle line:90% to reset. 00:41:10.690 --> 00:41:13.250 align:middle line:84% So it's called the attentional blink. 00:41:13.250 --> 00:41:17.570 align:middle line:84% So this also happens for sound, behaves very similarly. 00:41:17.570 --> 00:41:19.730 align:middle line:84% So it's the same kind of task where in this case 00:41:19.730 --> 00:41:22.690 align:middle line:90% there's a sequence of sounds. 00:41:22.690 --> 00:41:26.347 align:middle line:84% And I can't remember what distinguished 00:41:26.347 --> 00:41:27.430 align:middle line:90% the different sounds here. 00:41:27.430 --> 00:41:30.168 align:middle line:84% But there's two different types of targets. 00:41:30.168 --> 00:41:31.710 align:middle line:84% And again, it's the exact same graph. 00:41:31.710 --> 00:41:34.690 align:middle line:84% So this is the proportion of the second target that you correctly 00:41:34.690 --> 00:41:38.530 align:middle line:84% detect, conditioned on having detected the first target. 00:41:38.530 --> 00:41:41.430 align:middle line:84% And when it follows that in short succession, 00:41:41.430 --> 00:41:46.170 align:middle line:84% you're worse than when it comes some period of time afterwards. 00:41:46.170 --> 00:41:49.190 align:middle line:90% 00:41:49.190 --> 00:41:50.990 align:middle line:90% So that's the attentional blink. 00:41:50.990 --> 00:41:54.230 align:middle line:84% And it refers to the dynamics of attention over time. 00:41:54.230 --> 00:41:57.610 align:middle line:84% And again, is what we're talking about here when 00:41:57.610 --> 00:42:00.690 align:middle line:84% we use the word attention, is this the same thing as when 00:42:00.690 --> 00:42:05.370 align:middle line:84% we're talking about exogenous and endogenous cuing? 00:42:05.370 --> 00:42:09.200 align:middle line:90% I think it's not clear. 00:42:09.200 --> 00:42:10.220 align:middle line:90% Why you might ask? 00:42:10.220 --> 00:42:14.160 align:middle line:84% So well, why do we use the same word to talk about this stuff? 00:42:14.160 --> 00:42:16.060 align:middle line:90% And that's a good question. 00:42:16.060 --> 00:42:20.040 align:middle line:84% It probably has its roots in the colloquial usage of the word 00:42:20.040 --> 00:42:21.960 align:middle line:84% attention, where we often talk about the fact 00:42:21.960 --> 00:42:25.000 align:middle line:84% that we're paying attention to something. 00:42:25.000 --> 00:42:27.920 align:middle line:84% A lot of these tasks have the common character 00:42:27.920 --> 00:42:30.640 align:middle line:90% that they feel effortful. 00:42:30.640 --> 00:42:34.680 align:middle line:84% So volitional attention, often you 00:42:34.680 --> 00:42:37.120 align:middle line:84% feel like you're trying to do something. 00:42:37.120 --> 00:42:39.320 align:middle line:84% But whether or not it's really the same thing 00:42:39.320 --> 00:42:42.060 align:middle line:84% happening in your head, not totally clear. 00:42:42.060 --> 00:42:42.760 align:middle line:90% OK? 00:42:42.760 --> 00:42:43.260 align:middle line:90% All right. 00:42:43.260 --> 00:42:44.843 align:middle line:84% Questions about the attentional blink? 00:42:44.843 --> 00:42:48.240 align:middle line:90% 00:42:48.240 --> 00:42:49.140 align:middle line:90% Yeah? 00:42:49.140 --> 00:42:55.440 align:middle line:90% AUDIENCE: What does [INAUDIBLE] 00:42:55.440 --> 00:42:57.880 align:middle line:84% JOSH MCDERMOTT: So the black line 00:42:57.880 --> 00:43:06.470 align:middle line:84% is like a control where the T1 didn't actually occur. 00:43:06.470 --> 00:43:08.710 align:middle line:84% So this is showing how good you would 00:43:08.710 --> 00:43:14.310 align:middle line:84% be if you hadn't detected the first target because it wasn't 00:43:14.310 --> 00:43:14.910 align:middle line:90% there. 00:43:14.910 --> 00:43:19.210 align:middle line:84% And this is just showing that you're worse relative to that. 00:43:19.210 --> 00:43:22.350 align:middle line:84% But then that difference diminishes and maybe 00:43:22.350 --> 00:43:26.250 align:middle line:84% is eliminated by the time you're at position 8 or whatever. 00:43:26.250 --> 00:43:26.750 align:middle line:90% Yeah. 00:43:26.750 --> 00:43:31.950 align:middle line:90% 00:43:31.950 --> 00:43:36.990 align:middle line:84% So another of the many fascinating phenomena 00:43:36.990 --> 00:43:40.690 align:middle line:84% related to attention is what's known as change blindness. 00:43:40.690 --> 00:43:43.230 align:middle line:84% So how many of you have encountered change blindness 00:43:43.230 --> 00:43:44.535 align:middle line:90% in other settings? 00:43:44.535 --> 00:43:45.410 align:middle line:90% OK, a handful of you. 00:43:45.410 --> 00:43:48.830 align:middle line:84% Yeah so this is a pretty well-known effect. 00:43:48.830 --> 00:43:52.150 align:middle line:84% And it again, has a ton of real-world relevance. 00:43:52.150 --> 00:43:56.315 align:middle line:84% So we think that magicians are utilizing our susceptibility 00:43:56.315 --> 00:43:57.690 align:middle line:84% to change blindness all the time. 00:43:57.690 --> 00:43:59.510 align:middle line:84% And change blindness refers to the fact 00:43:59.510 --> 00:44:02.340 align:middle line:84% that people are in many contexts, 00:44:02.340 --> 00:44:08.180 align:middle line:84% surprisingly poor at noticing certain large-scale events that 00:44:08.180 --> 00:44:10.540 align:middle line:90% go on around us. 00:44:10.540 --> 00:44:14.620 align:middle line:84% So this was initially discovered in experiments using saccades. 00:44:14.620 --> 00:44:17.300 align:middle line:84% So people noticed just by accident 00:44:17.300 --> 00:44:19.980 align:middle line:84% that if you make a change to a display 00:44:19.980 --> 00:44:23.080 align:middle line:84% while people are in the midst of making an eye movement, 00:44:23.080 --> 00:44:26.860 align:middle line:84% so moving your eyes from here to here, 00:44:26.860 --> 00:44:28.780 align:middle line:84% if during that brief period of time 00:44:28.780 --> 00:44:30.360 align:middle line:84% you make a big change to the display, 00:44:30.360 --> 00:44:34.390 align:middle line:84% people just often wouldn't notice. 00:44:34.390 --> 00:44:38.740 align:middle line:84% And that got this large research program started. 00:44:38.740 --> 00:44:41.245 align:middle line:84% And much simpler demonstrations have since been developed. 00:44:41.245 --> 00:44:43.120 align:middle line:84% So I'll just show you a few examples of this. 00:44:43.120 --> 00:44:48.660 align:middle line:84% So we're going to alternate between two images OK. 00:44:48.660 --> 00:44:53.160 align:middle line:84% And there is something that is different about the two images. 00:44:53.160 --> 00:44:56.340 align:middle line:84% And I want you to raise your hand 00:44:56.340 --> 00:45:01.570 align:middle line:90% when you see what's different. 00:45:01.570 --> 00:45:09.420 align:middle line:84% Here we have two hands, three hands, four, five, six. 00:45:09.420 --> 00:45:11.170 align:middle line:84% If we keep doing this, eventually everyone 00:45:11.170 --> 00:45:12.890 align:middle line:90% will see it, right. 00:45:12.890 --> 00:45:14.910 align:middle line:90% But it takes a very long time. 00:45:14.910 --> 00:45:18.250 align:middle line:84% OK, so just to speed things up, here I'm 00:45:18.250 --> 00:45:21.890 align:middle line:84% going to put the two images at the same location. 00:45:21.890 --> 00:45:25.290 align:middle line:84% And everybody can see that there's that notebook thing that 00:45:25.290 --> 00:45:27.470 align:middle line:90% has been taken off of the desk. 00:45:27.470 --> 00:45:27.970 align:middle line:90% OK? 00:45:27.970 --> 00:45:33.130 align:middle line:90% 00:45:33.130 --> 00:45:34.470 align:middle line:90% Here's another case. 00:45:34.470 --> 00:45:37.410 align:middle line:84% So again raise your hand when you 00:45:37.410 --> 00:45:40.750 align:middle line:84% see the change that's happening between the two images. 00:45:40.750 --> 00:45:50.890 align:middle line:90% 00:45:50.890 --> 00:45:57.270 align:middle line:84% So we still have about half the class that hasn't seen it. 00:45:57.270 --> 00:45:59.880 align:middle line:90% So I'll just point to this. 00:45:59.880 --> 00:46:04.630 align:middle line:84% And then everyone says, oh, because once you attend to it, 00:46:04.630 --> 00:46:05.380 align:middle line:90% it's very obvious. 00:46:05.380 --> 00:46:09.720 align:middle line:90% Here's another one. 00:46:09.720 --> 00:46:11.180 align:middle line:90% Raise your hand when you see it. 00:46:11.180 --> 00:46:19.800 align:middle line:90% 00:46:19.800 --> 00:46:21.500 align:middle line:84% Some of you are still missing it. 00:46:21.500 --> 00:46:22.180 align:middle line:90% Look at this. 00:46:22.180 --> 00:46:25.200 align:middle line:90% 00:46:25.200 --> 00:46:26.120 align:middle line:90% Yeah. 00:46:26.120 --> 00:46:26.700 align:middle line:90% All right. 00:46:26.700 --> 00:46:27.200 align:middle line:90% OK. 00:46:27.200 --> 00:46:29.800 align:middle line:90% 00:46:29.800 --> 00:46:31.735 align:middle line:84% How many of you have seen this before? 00:46:31.735 --> 00:46:32.860 align:middle line:90% Yeah, just a couple of you. 00:46:32.860 --> 00:46:35.480 align:middle line:84% So there's some of cheesy music that 00:46:35.480 --> 00:46:37.420 align:middle line:84% accompanies this, which is beside the point, 00:46:37.420 --> 00:46:41.880 align:middle line:84% but it's a different kind of demo 00:46:41.880 --> 00:46:44.360 align:middle line:84% where there's a very, very big change that 00:46:44.360 --> 00:46:47.120 align:middle line:84% is going to be made to this that happens kind of gradually 00:46:47.120 --> 00:46:49.840 align:middle line:90% over the course of the video. 00:46:49.840 --> 00:46:51.740 align:middle line:90% And let's see if you see it. 00:46:51.740 --> 00:46:53.740 align:middle line:84% If you've already seen this and know what it is, 00:46:53.740 --> 00:46:54.532 align:middle line:90% don't say anything. 00:46:54.532 --> 00:46:55.280 align:middle line:90% All right? 00:46:55.280 --> 00:46:58.208 align:middle line:90% [VIDEO PLAYBACK] 00:46:58.208 --> 00:47:14.312 align:middle line:90% 00:47:14.312 --> 00:47:15.790 align:middle line:90% [END PLAYBACK] 00:47:15.790 --> 00:47:20.470 align:middle line:84% So did anybody see anything change? 00:47:20.470 --> 00:47:21.730 align:middle line:90% OK, a couple people. 00:47:21.730 --> 00:47:25.790 align:middle line:84% So note that what's the color of the sweatshirt 00:47:25.790 --> 00:47:29.483 align:middle line:90% of this guy at the start. 00:47:29.483 --> 00:47:30.150 align:middle line:90% AUDIENCE: Green. 00:47:30.150 --> 00:47:31.317 align:middle line:90% JOSH MCDERMOTT: Green, good. 00:47:31.317 --> 00:47:37.750 align:middle line:90% And what's the color at the end? 00:47:37.750 --> 00:47:38.710 align:middle line:90% AUDIENCE: Blue. 00:47:38.710 --> 00:47:40.930 align:middle line:84% JOSH MCDERMOTT: No, it's turquoise. 00:47:40.930 --> 00:47:44.930 align:middle line:84% And then now they're back in the starting position. 00:47:44.930 --> 00:47:49.910 align:middle line:84% So what happened here is the color of the sweatshirts morph. 00:47:49.910 --> 00:47:52.550 align:middle line:90% 00:47:52.550 --> 00:47:55.870 align:middle line:84% All right, and then by the end, they've 00:47:55.870 --> 00:47:57.690 align:middle line:84% rotated around the circle, right? 00:47:57.690 --> 00:48:01.800 align:middle line:84% So at the start, this guy is that turquoise blue. 00:48:01.800 --> 00:48:03.140 align:middle line:90% And this guy is green. 00:48:03.140 --> 00:48:10.500 align:middle line:84% And then they slowly change the colors. 00:48:10.500 --> 00:48:16.300 align:middle line:84% Until the guy in front is blue and the person to the right 00:48:16.300 --> 00:48:18.780 align:middle line:90% is purple. 00:48:18.780 --> 00:48:20.710 align:middle line:84% And then they switch around places 00:48:20.710 --> 00:48:22.460 align:middle line:84% So this is a case where, again, the change 00:48:22.460 --> 00:48:24.400 align:middle line:90% is made very continuously. 00:48:24.400 --> 00:48:29.013 align:middle line:84% And if I just go like this, well they're 00:48:29.013 --> 00:48:31.680 align:middle line:84% dancing so it's kind of hard to get them exactly the right spot. 00:48:31.680 --> 00:48:33.840 align:middle line:84% But it's a pretty salient change, right? 00:48:33.840 --> 00:48:38.440 align:middle line:84% And you can see that the brown person, 00:48:38.440 --> 00:48:41.080 align:middle line:84% the guy with the brown shirt is turned into green. 00:48:41.080 --> 00:48:42.840 align:middle line:90% These are big changes. 00:48:42.840 --> 00:48:44.420 align:middle line:84% But then they're done continuously 00:48:44.420 --> 00:48:46.380 align:middle line:90% and you don't see them. 00:48:46.380 --> 00:48:47.337 align:middle line:90% OK. 00:48:47.337 --> 00:48:49.420 align:middle line:84% And then there's also these really famous examples 00:48:49.420 --> 00:48:51.212 align:middle line:84% where stuff like this happens in real life. 00:48:51.212 --> 00:48:53.940 align:middle line:84% So there's this legendary study that 00:48:53.940 --> 00:48:57.450 align:middle line:84% was done where on a college campus, 00:48:57.450 --> 00:49:01.570 align:middle line:84% there were experimenters who would 00:49:01.570 --> 00:49:05.650 align:middle line:84% stop random people on campus and ask them for directions 00:49:05.650 --> 00:49:06.770 align:middle line:90% to some building. 00:49:06.770 --> 00:49:09.870 align:middle line:84% And so they'd be talking to them for a brief period of time. 00:49:09.870 --> 00:49:13.410 align:middle line:84% And during that conversation, there 00:49:13.410 --> 00:49:15.050 align:middle line:84% would be a couple people who would 00:49:15.050 --> 00:49:17.730 align:middle line:84% walk by with a door, a big door, and they 00:49:17.730 --> 00:49:20.610 align:middle line:84% would walk in between the person who 00:49:20.610 --> 00:49:23.930 align:middle line:84% was being asked for directions and then the experimenter. 00:49:23.930 --> 00:49:27.130 align:middle line:84% And while the door was between them, 00:49:27.130 --> 00:49:29.043 align:middle line:84% this person would hop onto carrying the door, 00:49:29.043 --> 00:49:31.710 align:middle line:84% and the person carrying the door would take this person's place. 00:49:31.710 --> 00:49:36.370 align:middle line:84% So the person that this unsuspecting participant 00:49:36.370 --> 00:49:38.410 align:middle line:90% was talking to would change. 00:49:38.410 --> 00:49:40.890 align:middle line:84% And the finding is just that a lot of the time, they 00:49:40.890 --> 00:49:43.330 align:middle line:84% just don't notice that the person that they're talking to 00:49:43.330 --> 00:49:44.070 align:middle line:90% has changed. 00:49:44.070 --> 00:49:45.690 align:middle line:84% So there's lots and lots and lots 00:49:45.690 --> 00:49:47.030 align:middle line:90% of examples of stuff like this. 00:49:47.030 --> 00:49:49.370 align:middle line:90% OK. 00:49:49.370 --> 00:49:54.250 align:middle line:84% And it's just quite surprising because you look out 00:49:54.250 --> 00:49:57.320 align:middle line:84% at the world, and you have the sense 00:49:57.320 --> 00:50:00.600 align:middle line:84% that your visual perception is very rich. 00:50:00.600 --> 00:50:04.120 align:middle line:84% I feel like I can see all of you, right? 00:50:04.120 --> 00:50:07.720 align:middle line:84% And these experiments suggest that actually there's 00:50:07.720 --> 00:50:12.880 align:middle line:84% lots of stuff that we're either not seeing, or not remembering, 00:50:12.880 --> 00:50:14.620 align:middle line:90% or not representing in some way. 00:50:14.620 --> 00:50:19.760 align:middle line:84% So the standard explanation of change blindness 00:50:19.760 --> 00:50:22.660 align:middle line:84% is that you need attention to see change. 00:50:22.660 --> 00:50:26.960 align:middle line:84% So without it, observers are change blind. 00:50:26.960 --> 00:50:29.920 align:middle line:84% And so the intriguing implication 00:50:29.920 --> 00:50:31.720 align:middle line:84% is that this impression that we have 00:50:31.720 --> 00:50:37.000 align:middle line:84% of rich sensory representations is largely illusory. 00:50:37.000 --> 00:50:39.280 align:middle line:84% So we think that change blindness phenomena suggests 00:50:39.280 --> 00:50:42.440 align:middle line:84% that the actual sensory representations that we're 00:50:42.440 --> 00:50:45.080 align:middle line:84% representing in our heads are fairly sparse, 00:50:45.080 --> 00:50:48.460 align:middle line:84% and that relatively little is retained over time. 00:50:48.460 --> 00:50:51.080 align:middle line:90% 00:50:51.080 --> 00:50:51.580 align:middle line:90% All right. 00:50:51.580 --> 00:50:54.110 align:middle line:84% So in everyday life, we usually see change. 00:50:54.110 --> 00:50:55.850 align:middle line:84% If we didn't, it would be a problem. 00:50:55.850 --> 00:50:56.830 align:middle line:90% OK? 00:50:56.830 --> 00:50:59.950 align:middle line:84% But what enables this typically is usually 00:50:59.950 --> 00:51:02.270 align:middle line:84% when things change in the world, there's 00:51:02.270 --> 00:51:04.530 align:middle line:84% some low-level transient in the image. 00:51:04.530 --> 00:51:05.970 align:middle line:90% There's a motion signal. 00:51:05.970 --> 00:51:08.230 align:middle line:84% So something moves or something flickers. 00:51:08.230 --> 00:51:11.670 align:middle line:84% So if we go back to this example, 00:51:11.670 --> 00:51:14.190 align:middle line:84% you see the book disappearing because there's 00:51:14.190 --> 00:51:18.730 align:middle line:84% this big thing that causes flicker in the image. 00:51:18.730 --> 00:51:20.350 align:middle line:84% And so what prevented you from seeing 00:51:20.350 --> 00:51:23.630 align:middle line:84% that was having the image displaced 00:51:23.630 --> 00:51:27.030 align:middle line:84% so that you didn't have that local signal 00:51:27.030 --> 00:51:29.370 align:middle line:90% to take advantage of. 00:51:29.370 --> 00:51:32.070 align:middle line:90% 00:51:32.070 --> 00:51:34.110 align:middle line:84% And so all of these examples that I've shown you 00:51:34.110 --> 00:51:38.990 align:middle line:84% of change blindness, something is interfering with attention 00:51:38.990 --> 00:51:40.670 align:middle line:90% being drawn to the change. 00:51:40.670 --> 00:51:42.698 align:middle line:90% So there's flicker. 00:51:42.698 --> 00:51:44.990 align:middle line:84% So in those examples where we alternate between the two 00:51:44.990 --> 00:51:47.470 align:middle line:84% images, there's a blank screen in between. 00:51:47.470 --> 00:51:52.110 align:middle line:84% And so that prevents these local motion signals 00:51:52.110 --> 00:51:54.060 align:middle line:90% that would normally be created. 00:51:54.060 --> 00:51:57.320 align:middle line:84% Sometimes there'll be occlusions that get put into place. 00:51:57.320 --> 00:51:59.140 align:middle line:84% And in the case of the color, it's 00:51:59.140 --> 00:52:01.660 align:middle line:84% that the change is very, very slow. 00:52:01.660 --> 00:52:04.900 align:middle line:84% So you don't get this rapid change 00:52:04.900 --> 00:52:07.360 align:middle line:84% that would normally be detected by your visual system. 00:52:07.360 --> 00:52:11.578 align:middle line:84% So if something interferes with the drawing of attention, 00:52:11.578 --> 00:52:14.120 align:middle line:84% then attention doesn't go to the change and you won't see it. 00:52:14.120 --> 00:52:17.362 align:middle line:84% So in those examples where the images were alternating 00:52:17.362 --> 00:52:19.820 align:middle line:84% back and forth, like the one with the plane engine missing, 00:52:19.820 --> 00:52:22.415 align:middle line:84% so we had waited a little while and a bunch of people 00:52:22.415 --> 00:52:23.040 align:middle line:90% hadn't seen it. 00:52:23.040 --> 00:52:24.880 align:middle line:84% And so then I just point to the engine and then of course, 00:52:24.880 --> 00:52:25.640 align:middle line:90% everybody sees it. 00:52:25.640 --> 00:52:27.600 align:middle line:84% Because that draws your attention to the thing. 00:52:27.600 --> 00:52:28.100 align:middle line:90% OK? 00:52:28.100 --> 00:52:33.000 align:middle line:90% 00:52:33.000 --> 00:52:36.620 align:middle line:84% So another important thing to note 00:52:36.620 --> 00:52:40.980 align:middle line:84% is there doesn't seem to be much accumulation 00:52:40.980 --> 00:52:42.840 align:middle line:90% of built-up representation. 00:52:42.840 --> 00:52:47.780 align:middle line:84% So we could repeat those demos of the pictures alternating back 00:52:47.780 --> 00:52:51.690 align:middle line:84% and forth and let you look at the first picture for a while 00:52:51.690 --> 00:52:52.490 align:middle line:90% first. 00:52:52.490 --> 00:52:54.890 align:middle line:84% So you could look all over the image 00:52:54.890 --> 00:52:57.410 align:middle line:84% and then start the same demo, and it wouldn't really 00:52:57.410 --> 00:52:58.270 align:middle line:90% make a difference. 00:52:58.270 --> 00:53:01.770 align:middle line:90% 00:53:01.770 --> 00:53:04.490 align:middle line:84% It seems that there's this representation that's 00:53:04.490 --> 00:53:07.910 align:middle line:84% built up by attention, which typically moves with your eyes. 00:53:07.910 --> 00:53:09.350 align:middle line:84% So you look at a particular place, 00:53:09.350 --> 00:53:12.170 align:middle line:84% you see the stuff that's there, but it 00:53:12.170 --> 00:53:15.470 align:middle line:84% stays built up only as long as attention is at that location. 00:53:15.470 --> 00:53:19.070 align:middle line:84% So it seems like it dissolves once attention is withdrawn. 00:53:19.070 --> 00:53:21.370 align:middle line:84% So again, we talk a lot about these loose metaphors 00:53:21.370 --> 00:53:22.790 align:middle line:90% in the world of attention. 00:53:22.790 --> 00:53:25.190 align:middle line:84% So think of attention acting like a hand. 00:53:25.190 --> 00:53:27.373 align:middle line:84% It grabs the stuff, holds it together. 00:53:27.373 --> 00:53:29.790 align:middle line:84% You can then look at it and make discriminations about it, 00:53:29.790 --> 00:53:32.370 align:middle line:84% but then it moves on, and the hand lets go, 00:53:32.370 --> 00:53:36.490 align:middle line:84% and the things go back to not being 00:53:36.490 --> 00:53:38.110 align:middle line:90% represented in the same way. 00:53:38.110 --> 00:53:41.050 align:middle line:90% 00:53:41.050 --> 00:53:43.130 align:middle line:84% So really interesting question here 00:53:43.130 --> 00:53:47.810 align:middle line:84% is that if the conclusion to these observations about chain 00:53:47.810 --> 00:53:51.000 align:middle line:84% blindness are correct and we're actually only seeing 00:53:51.000 --> 00:53:53.520 align:middle line:84% a few things at a time, why do we 00:53:53.520 --> 00:53:59.920 align:middle line:84% have the impression of seeing such a rich world out there? 00:53:59.920 --> 00:54:02.280 align:middle line:84% So I think one of the really deep insights 00:54:02.280 --> 00:54:06.320 align:middle line:84% to come out of this line of work is 00:54:06.320 --> 00:54:09.500 align:middle line:84% that whenever we need to see something, we look at it. 00:54:09.500 --> 00:54:13.800 align:middle line:90% 00:54:13.800 --> 00:54:15.880 align:middle line:84% So because we can just look at whatever 00:54:15.880 --> 00:54:19.260 align:middle line:84% it is that we want to see, in some sense, 00:54:19.260 --> 00:54:23.360 align:middle line:84% there's no need to have rich representations in your head. 00:54:23.360 --> 00:54:25.860 align:middle line:84% So this is often called "just in time" vision, 00:54:25.860 --> 00:54:29.960 align:middle line:84% another key idea and catchphrase associated with this is the idea 00:54:29.960 --> 00:54:33.160 align:middle line:84% that eye movements are cheaper than memory. 00:54:33.160 --> 00:54:34.640 align:middle line:90% So memory is expensive. 00:54:34.640 --> 00:54:36.680 align:middle line:84% You have to do stuff in your brain, 00:54:36.680 --> 00:54:39.780 align:middle line:84% expend metabolic resources, maybe use up synapses. 00:54:39.780 --> 00:54:43.440 align:middle line:84% So there's a cost to actually remembering stuff. 00:54:43.440 --> 00:54:46.140 align:middle line:84% Whereas eye movements are plausibly pretty cheap. 00:54:46.140 --> 00:54:50.005 align:middle line:84% They don't cost the organism very much. 00:54:50.005 --> 00:54:52.870 align:middle line:84% And so if you're living in a world that's stable 00:54:52.870 --> 00:54:54.630 align:middle line:90% and this world is pretty stable. 00:54:54.630 --> 00:54:57.710 align:middle line:84% Most of you haven't moved around a whole lot over the last 10 00:54:57.710 --> 00:54:59.750 align:middle line:90% minutes. 00:54:59.750 --> 00:55:03.190 align:middle line:84% You can actually use the world as an outside memory. 00:55:03.190 --> 00:55:04.650 align:middle line:90% You want to know what's there. 00:55:04.650 --> 00:55:08.270 align:middle line:90% You just look at it, OK? 00:55:08.270 --> 00:55:12.052 align:middle line:84% And so we have this illusion of seeing all this stuff 00:55:12.052 --> 00:55:14.010 align:middle line:84% because anytime that you want to see something, 00:55:14.010 --> 00:55:15.630 align:middle line:84% you just look at it, and it's there. 00:55:15.630 --> 00:55:19.570 align:middle line:84% So again, another useful analogy is like the refrigerator light, 00:55:19.570 --> 00:55:20.990 align:middle line:90% right? 00:55:20.990 --> 00:55:22.810 align:middle line:84% So when you were little might have thought, 00:55:22.810 --> 00:55:24.350 align:middle line:84% well, the light in the refrigerator is on all the time 00:55:24.350 --> 00:55:26.517 align:middle line:84% because every time you open the door the light's on. 00:55:26.517 --> 00:55:28.807 align:middle line:84% But in actuality, the light is mostly off 00:55:28.807 --> 00:55:30.890 align:middle line:84% and it just turns on every time you open the door. 00:55:30.890 --> 00:55:32.490 align:middle line:90% So it looks like it's always on. 00:55:32.490 --> 00:55:35.150 align:middle line:90% 00:55:35.150 --> 00:55:39.790 align:middle line:84% So that's the general conclusion from that particular literature 00:55:39.790 --> 00:55:43.070 align:middle line:84% which is that you rely on attention in order 00:55:43.070 --> 00:55:45.110 align:middle line:90% to detect changes. 00:55:45.110 --> 00:55:47.800 align:middle line:84% And whenever you attend to something, 00:55:47.800 --> 00:55:51.580 align:middle line:84% you can see that, and represent it, and notice the changes. 00:55:51.580 --> 00:55:53.900 align:middle line:84% But there's a lot of stuff that is not represented 00:55:53.900 --> 00:55:55.360 align:middle line:90% in the absence of attention. 00:55:55.360 --> 00:55:58.420 align:middle line:90% 00:55:58.420 --> 00:55:59.920 align:middle line:84% What questions you got about that? 00:55:59.920 --> 00:56:06.100 align:middle line:90% 00:56:06.100 --> 00:56:07.900 align:middle line:90% Yeah? 00:56:07.900 --> 00:56:10.300 align:middle line:84% AUDIENCE: Is this the representation 00:56:10.300 --> 00:56:13.040 align:middle line:84% of our surroundings, something that's trainable at all, 00:56:13.040 --> 00:56:16.920 align:middle line:84% or are we just incapable of encoding that kind of stuff? 00:56:16.920 --> 00:56:24.380 align:middle line:90% 00:56:24.380 --> 00:56:26.140 align:middle line:84% JOSH MCDERMOTT: Yeah, I don't actually 00:56:26.140 --> 00:56:29.860 align:middle line:84% know of good attempts that have been made 00:56:29.860 --> 00:56:34.100 align:middle line:90% to alter this with training. 00:56:34.100 --> 00:56:38.140 align:middle line:84% My guess is that it would be pretty hard, 00:56:38.140 --> 00:56:40.020 align:middle line:90% but I don't know for sure. 00:56:40.020 --> 00:56:42.660 align:middle line:90% Yeah. 00:56:42.660 --> 00:56:43.240 align:middle line:90% Yeah? 00:56:43.240 --> 00:56:46.530 align:middle line:84% AUDIENCE: Is change blindness limited to the visual modality, 00:56:46.530 --> 00:56:49.050 align:middle line:84% or would you be able to replicate 00:56:49.050 --> 00:56:50.325 align:middle line:90% this in like an auditory way? 00:56:50.325 --> 00:56:52.450 align:middle line:84% JOSH MCDERMOTT: Yeah, there are some demonstrations 00:56:52.450 --> 00:56:54.890 align:middle line:84% of what's called change deafness, where 00:56:54.890 --> 00:56:58.530 align:middle line:84% you can play people an auditory scene with five 00:56:58.530 --> 00:57:01.810 align:middle line:84% sounds in different locations, and show that they're not 00:57:01.810 --> 00:57:04.710 align:middle line:84% great at noticing, one of the sounds disappearing, 00:57:04.710 --> 00:57:05.350 align:middle line:90% for instance. 00:57:05.350 --> 00:57:08.170 align:middle line:84% But then if you, again, you draw their attention to it, 00:57:08.170 --> 00:57:09.670 align:middle line:84% then of course, they hear it, right? 00:57:09.670 --> 00:57:15.150 align:middle line:84% So yeah, I think it's probably pretty general. 00:57:15.150 --> 00:57:15.650 align:middle line:90% Yeah. 00:57:15.650 --> 00:57:19.170 align:middle line:90% 00:57:19.170 --> 00:57:20.550 align:middle line:90% Any other questions on that? 00:57:20.550 --> 00:57:23.770 align:middle line:90% 00:57:23.770 --> 00:57:33.070 align:middle line:84% So all of this suggests that representations are sparse, 00:57:33.070 --> 00:57:36.090 align:middle line:84% that you don't actually remember a whole lot about what 00:57:36.090 --> 00:57:38.090 align:middle line:90% you are seeing. 00:57:38.090 --> 00:57:42.690 align:middle line:84% And I want to give you an interesting example that points 00:57:42.690 --> 00:57:45.060 align:middle line:90% in the opposite direction. 00:57:45.060 --> 00:57:49.240 align:middle line:84% So I want you to look at every single image in this array, just 00:57:49.240 --> 00:58:00.120 align:middle line:84% one at a time, just briefly 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 00:58:00.120 --> 00:58:01.740 align:middle line:90% 12, 13, 14. 00:58:01.740 --> 00:58:02.360 align:middle line:90% 15, 16. 00:58:02.360 --> 00:58:02.860 align:middle line:90% All right. 00:58:02.860 --> 00:58:03.500 align:middle line:90% Got them all. 00:58:03.500 --> 00:58:10.000 align:middle line:84% So this is a brief gap to avoid image transients. 00:58:10.000 --> 00:58:15.560 align:middle line:84% So now I want to ask you which of these did you just see? 00:58:15.560 --> 00:58:19.000 align:middle line:84% So raise your hand if that one was there before. 00:58:19.000 --> 00:58:20.640 align:middle line:90% How about that? 00:58:20.640 --> 00:58:22.440 align:middle line:90% How about that? 00:58:22.440 --> 00:58:24.760 align:middle line:90% How about that? 00:58:24.760 --> 00:58:26.640 align:middle line:90% How about this? 00:58:26.640 --> 00:58:28.760 align:middle line:90% How about this? 00:58:28.760 --> 00:58:30.560 align:middle line:90% How about this? 00:58:30.560 --> 00:58:32.840 align:middle line:90% How about this? 00:58:32.840 --> 00:58:35.400 align:middle line:90% How about this one? 00:58:35.400 --> 00:58:37.120 align:middle line:90% This one? 00:58:37.120 --> 00:58:38.920 align:middle line:90% This one? 00:58:38.920 --> 00:58:40.440 align:middle line:90% This one? 00:58:40.440 --> 00:58:42.560 align:middle line:90% This one? 00:58:42.560 --> 00:58:44.230 align:middle line:90% This one? 00:58:44.230 --> 00:58:45.990 align:middle line:90% This one? 00:58:45.990 --> 00:58:47.070 align:middle line:90% This one? 00:58:47.070 --> 00:58:47.570 align:middle line:90% All right. 00:58:47.570 --> 00:58:49.830 align:middle line:90% So quite a lot of agreement. 00:58:49.830 --> 00:58:52.110 align:middle line:90% This is the answer key. 00:58:52.110 --> 00:58:55.150 align:middle line:90% So you all did very well. 00:58:55.150 --> 00:58:56.430 align:middle line:90% All right. 00:58:56.430 --> 00:59:00.830 align:middle line:84% And what we just observed is the fact 00:59:00.830 --> 00:59:08.670 align:middle line:84% that visual memory for images is spectacularly good. 00:59:08.670 --> 00:59:12.130 align:middle line:84% And that had been known to some extent for a long time. 00:59:12.130 --> 00:59:13.790 align:middle line:84% But there was very influential work 00:59:13.790 --> 00:59:16.830 align:middle line:84% that actually was actually done in this building 00:59:16.830 --> 00:59:20.950 align:middle line:84% by Aude Oliva and Tim Brady and others 00:59:20.950 --> 00:59:25.730 align:middle line:84% to document this rigorously and with modern standards. 00:59:25.730 --> 00:59:28.470 align:middle line:84% Finding that people can remember, 00:59:28.470 --> 00:59:30.665 align:middle line:84% a remarkably large number of images just 00:59:30.665 --> 00:59:32.790 align:middle line:84% from a single exposure and often with a fair degree 00:59:32.790 --> 00:59:34.110 align:middle line:90% of specificity. 00:59:34.110 --> 00:59:37.050 align:middle line:90% 00:59:37.050 --> 00:59:40.710 align:middle line:84% So this is an indication that, well, some image properties 00:59:40.710 --> 00:59:44.780 align:middle line:84% are very easily retained, just from one exposure. 00:59:44.780 --> 00:59:48.380 align:middle line:90% 00:59:48.380 --> 00:59:52.220 align:middle line:84% And that seems at odds with these findings of change 00:59:52.220 --> 00:59:55.140 align:middle line:84% blindness, where we were making these large changes to images 00:59:55.140 --> 00:59:57.820 align:middle line:90% and people weren't noticing. 00:59:57.820 --> 01:00:02.260 align:middle line:84% I think the way to reconcile these things is that typically 01:00:02.260 --> 01:00:07.340 align:middle line:84% the changes that are made in these change blindness demos, 01:00:07.340 --> 01:00:11.600 align:middle line:84% they often don't alter the meaning of the image. 01:00:11.600 --> 01:00:13.540 align:middle line:84% So you have that image of the plane, 01:00:13.540 --> 01:00:16.240 align:middle line:84% and yeah, the engine pops on and off, but it's still a plane. 01:00:16.240 --> 01:00:17.780 align:middle line:90% There's still people. 01:00:17.780 --> 01:00:20.038 align:middle line:84% You've got the image of people having dinner, 01:00:20.038 --> 01:00:21.080 align:middle line:90% the railing moves around. 01:00:21.080 --> 01:00:23.390 align:middle line:84% It's like a visually salient change at some level, 01:00:23.390 --> 01:00:25.140 align:middle line:84% but it doesn't really change what we often 01:00:25.140 --> 01:00:27.340 align:middle line:90% call the gist of a scene. 01:00:27.340 --> 01:00:30.460 align:middle line:84% And so it may well be that what you're 01:00:30.460 --> 01:00:35.620 align:middle line:84% really good at remembering is something related to gist, 01:00:35.620 --> 01:00:38.500 align:middle line:84% and there's a lot of visual details 01:00:38.500 --> 01:00:40.610 align:middle line:84% that are not really critical to that. 01:00:40.610 --> 01:00:43.430 align:middle line:84% And those are things that are actually not really represented. 01:00:43.430 --> 01:00:48.170 align:middle line:84% But again, we don't yet have complete theories about this. 01:00:48.170 --> 01:00:52.250 align:middle line:90% But it's pretty striking. 01:00:52.250 --> 01:00:56.750 align:middle line:84% So another angle on attention, in fact, 01:00:56.750 --> 01:00:59.970 align:middle line:84% one of the very first attempts to study attention 01:00:59.970 --> 01:01:03.530 align:middle line:84% was in the auditory domain with selective listening. 01:01:03.530 --> 01:01:07.410 align:middle line:84% This was pioneered by Colin Cherry. 01:01:07.410 --> 01:01:11.250 align:middle line:84% And this used a paradigm called speech shadowing. 01:01:11.250 --> 01:01:15.110 align:middle line:84% So in this experiment you got headphones on. 01:01:15.110 --> 01:01:18.090 align:middle line:84% And there are two different speech streams 01:01:18.090 --> 01:01:20.930 align:middle line:84% or two audio streams, one to the left ear 01:01:20.930 --> 01:01:22.530 align:middle line:90% and one to the right ear. 01:01:22.530 --> 01:01:25.130 align:middle line:84% And the task is to repeat or what's 01:01:25.130 --> 01:01:29.270 align:middle line:84% called shadow, what's being said in one of the ears. 01:01:29.270 --> 01:01:33.330 align:middle line:90% 01:01:33.330 --> 01:01:35.810 align:middle line:84% And the two audio streams are independent of each other. 01:01:35.810 --> 01:01:37.832 align:middle line:84% So you might hear this and if this 01:01:37.832 --> 01:01:39.290 align:middle line:84% is the ear you've got a shadow, you 01:01:39.290 --> 01:01:41.068 align:middle line:84% have to say fruit, morning, planet, door, 01:01:41.068 --> 01:01:41.860 align:middle line:90% swan, for instance. 01:01:41.860 --> 01:01:44.720 align:middle line:84% And then there's all this stuff coming in the other ear. 01:01:44.720 --> 01:01:48.080 align:middle line:84% And the point is that this is a really hard task. 01:01:48.080 --> 01:01:49.560 align:middle line:84% It's so hard that you really have 01:01:49.560 --> 01:01:52.320 align:middle line:84% to focus your attention on the stream 01:01:52.320 --> 01:01:56.940 align:middle line:84% that you're supposed to be paying attention to. 01:01:56.940 --> 01:01:59.280 align:middle line:84% So this is the title page from the original papers 01:01:59.280 --> 01:02:02.760 align:middle line:90% from back in 1953. 01:02:02.760 --> 01:02:06.280 align:middle line:84% It's called "Some Experiments on the Recognition of Speech with 01:02:06.280 --> 01:02:08.400 align:middle line:90% One and with Two Ears." 01:02:08.400 --> 01:02:10.520 align:middle line:90% OK. 01:02:10.520 --> 01:02:14.000 align:middle line:84% Using speech shadowing, and the kind 01:02:14.000 --> 01:02:16.320 align:middle line:84% of key result from these experiments 01:02:16.320 --> 01:02:18.720 align:middle line:84% is that when people are successfully 01:02:18.720 --> 01:02:21.720 align:middle line:84% performing this task, they're shadowing one of the streams. 01:02:21.720 --> 01:02:24.760 align:middle line:84% They recall almost nothing from the unattended stream. 01:02:24.760 --> 01:02:33.320 align:middle line:84% So here's an excerpt from the results section. 01:02:33.320 --> 01:02:36.327 align:middle line:84% It says, what factors of the rejected message are recognized? 01:02:36.327 --> 01:02:37.910 align:middle line:84% Rejected means the one that you're not 01:02:37.910 --> 01:02:40.710 align:middle line:84% like supposed to be listening to and shadowing. 01:02:40.710 --> 01:02:44.030 align:middle line:90% 01:02:44.030 --> 01:02:46.310 align:middle line:84% "In this series of tests, the listening subjects 01:02:46.310 --> 01:02:48.710 align:middle line:84% were presented at their right hand ears 01:02:48.710 --> 01:02:51.630 align:middle line:84% with spoken passages from newspapers, chosen carefully 01:02:51.630 --> 01:02:54.310 align:middle line:84% to avoid proper names or difficult words, 01:02:54.310 --> 01:02:56.970 align:middle line:84% and again instructed to repeat these passages concurrently 01:02:56.970 --> 01:02:58.150 align:middle line:90% without omission or error. 01:02:58.150 --> 01:03:01.750 align:middle line:84% Into their left ears were fed signals of different kinds 01:03:01.750 --> 01:03:04.990 align:middle line:84% for different tests, but each of which started and ended with 01:03:04.990 --> 01:03:08.132 align:middle line:84% a short passage of normal English speech in order to avoid 01:03:08.132 --> 01:03:09.590 align:middle line:84% any troubles that might be involved 01:03:09.590 --> 01:03:12.270 align:middle line:84% in the listener's 'getting going' on the test. 01:03:12.270 --> 01:03:15.870 align:middle line:84% The center, major portions of these rejected left ear signals 01:03:15.870 --> 01:03:19.350 align:middle line:84% thus reached the listener, while he or she was steadily repeating 01:03:19.350 --> 01:03:21.870 align:middle line:90% the right ear message." 01:03:21.870 --> 01:03:24.550 align:middle line:84% OK, so these were some of the conditions in this experiment. 01:03:24.550 --> 01:03:27.510 align:middle line:84% The center major portions of the left ear signals 01:03:27.510 --> 01:03:31.750 align:middle line:84% were they could be normal male spoken English, which 01:03:31.750 --> 01:03:33.730 align:middle line:84% was what was going in the attended ear; 01:03:33.730 --> 01:03:36.710 align:middle line:84% female spoken English, so this notably has a higher pitch so it 01:03:36.710 --> 01:03:40.860 align:middle line:84% sounds really different; reversed male speech, so 01:03:40.860 --> 01:03:43.380 align:middle line:84% same kind of frequency content but without words 01:03:43.380 --> 01:03:48.010 align:middle line:84% or semantic content; a steady 400 cycle per second oscillator, 01:03:48.010 --> 01:03:52.020 align:middle line:84% so that's a pure tone at 400 hertz, so a beep. 01:03:52.020 --> 01:03:54.920 align:middle line:84% And so after any one of these tests, the subject was asked, 01:03:54.920 --> 01:03:57.980 align:middle line:84% did the left ear signal consist of human speech or not. 01:03:57.980 --> 01:04:02.080 align:middle line:84% If yes, can you say what it was about or even quote any words? 01:04:02.080 --> 01:04:06.260 align:middle line:84% Was it a male or female speaker, and what language was it in? 01:04:06.260 --> 01:04:08.340 align:middle line:84% So here's the description of the results. 01:04:08.340 --> 01:04:10.980 align:middle line:84% "The responses varied only slightly. 01:04:10.980 --> 01:04:14.020 align:middle line:84% In no case in which normal human speech was used 01:04:14.020 --> 01:04:17.460 align:middle line:84% did the listening subjects fail to identify it as speech; 01:04:17.460 --> 01:04:19.020 align:middle line:84% in every such instance, they were 01:04:19.020 --> 01:04:23.480 align:middle line:84% unable to identify any word or phrase heard in the rejected ear 01:04:23.480 --> 01:04:26.620 align:middle line:84% and, furthermore, unable to make definite identification 01:04:26.620 --> 01:04:29.420 align:middle line:84% of the language as being English. 01:04:29.420 --> 01:04:32.500 align:middle line:84% On the other hand, the change of voice-- male to female-- 01:04:32.500 --> 01:04:35.540 align:middle line:84% was nearly always identified, while the 400 hertz 01:04:35.540 --> 01:04:38.130 align:middle line:90% pure tone was always observed. 01:04:38.130 --> 01:04:39.850 align:middle line:84% The reversed speech was identified 01:04:39.850 --> 01:04:42.590 align:middle line:84% as having something funny about it by a few listeners, 01:04:42.590 --> 01:04:45.490 align:middle line:84% but was thought to be normal speech by others. 01:04:45.490 --> 01:04:48.370 align:middle line:84% The broad conclusions are that the 'rejected' signal has 01:04:48.370 --> 01:04:51.030 align:middle line:84% certain statistical properties recognized, 01:04:51.030 --> 01:04:53.550 align:middle line:84% but that detailed aspects such as the language, 01:04:53.550 --> 01:04:58.570 align:middle line:84% individual words, or semantic content are unnoticed." 01:04:58.570 --> 01:04:59.070 align:middle line:90% All right. 01:04:59.070 --> 01:05:01.530 align:middle line:84% So you're able to hear some aspects 01:05:01.530 --> 01:05:04.370 align:middle line:84% of the unattended signal, but they're kind of basic 01:05:04.370 --> 01:05:10.150 align:middle line:84% and don't include things like language. 01:05:10.150 --> 01:05:10.650 align:middle line:90% OK? 01:05:10.650 --> 01:05:11.633 align:middle line:90% Yeah? 01:05:11.633 --> 01:05:12.550 align:middle line:90% AUDIENCE: I'm curious. 01:05:12.550 --> 01:05:14.490 align:middle line:84% Did they measure all the performance 01:05:14.490 --> 01:05:18.010 align:middle line:84% of the shadowing task because they have a start 01:05:18.010 --> 01:05:21.490 align:middle line:84% and then like a middle important [INAUDIBLE]. 01:05:21.490 --> 01:05:26.090 align:middle line:84% I'm curious whether they know when that change happens, when 01:05:26.090 --> 01:05:29.150 align:middle line:84% the shadowing tasks kind of slowed down for a second 01:05:29.150 --> 01:05:31.170 align:middle line:84% like to pick it up and then switch back. 01:05:31.170 --> 01:05:32.930 align:middle line:84% JOSH MCDERMOTT: Yeah, I don't think 01:05:32.930 --> 01:05:36.640 align:middle line:90% that was explicitly measured. 01:05:36.640 --> 01:05:38.280 align:middle line:84% I mean, they made sure that people 01:05:38.280 --> 01:05:41.600 align:middle line:90% were doing the task right. 01:05:41.600 --> 01:05:47.520 align:middle line:84% And my guess is that if they mess up, they just start over. 01:05:47.520 --> 01:05:48.440 align:middle line:90% Yeah. 01:05:48.440 --> 01:05:51.160 align:middle line:84% And I should say, I mean, this is 01:05:51.160 --> 01:05:54.960 align:middle line:90% a massively influential study. 01:05:54.960 --> 01:05:57.080 align:middle line:84% As an auditory experiment, it's pretty complicated 01:05:57.080 --> 01:05:59.262 align:middle line:84% because the person's talking at the same time 01:05:59.262 --> 01:06:00.720 align:middle line:84% as they're listening to this stuff. 01:06:00.720 --> 01:06:02.080 align:middle line:84% And so, I mean, they can probably 01:06:02.080 --> 01:06:03.580 align:middle line:84% hear themselves talk to some extent. 01:06:03.580 --> 01:06:08.480 align:middle line:84% And what effect that has on things is hard to know. 01:06:08.480 --> 01:06:12.660 align:middle line:84% And this has been revisited a few times, but yeah, 01:06:12.660 --> 01:06:17.760 align:middle line:84% probably not as much as maybe it should be. 01:06:17.760 --> 01:06:19.760 align:middle line:84% So from the unattended channel you 01:06:19.760 --> 01:06:22.488 align:middle line:84% notice large changes in volume, change 01:06:22.488 --> 01:06:24.780 align:middle line:84% in the gender of the speaker, not a change in language. 01:06:24.780 --> 01:06:27.440 align:middle line:84% So some things seem to draw attention 01:06:27.440 --> 01:06:29.680 align:middle line:84% to themselves, these very salient low level 01:06:29.680 --> 01:06:31.120 align:middle line:90% properties of the signal. 01:06:31.120 --> 01:06:33.670 align:middle line:84% And we think of them as being processed more or less 01:06:33.670 --> 01:06:35.110 align:middle line:90% without attention. 01:06:35.110 --> 01:06:36.970 align:middle line:90% Other things are not. 01:06:36.970 --> 01:06:39.430 align:middle line:90% 01:06:39.430 --> 01:06:39.930 align:middle line:90% All right. 01:06:39.930 --> 01:06:42.138 align:middle line:84% Let's talk a little bit about attention in the brain. 01:06:42.138 --> 01:06:45.095 align:middle line:90% 01:06:45.095 --> 01:06:49.288 align:middle line:84% So in general, when you pay attention to something, 01:06:49.288 --> 01:06:50.830 align:middle line:84% it could be a part of space, it could 01:06:50.830 --> 01:06:56.830 align:middle line:84% be a particular type of feature, it could be a type of object, 01:06:56.830 --> 01:07:00.750 align:middle line:84% we generally observe enhancement of neural activity 01:07:00.750 --> 01:07:05.150 align:middle line:84% in the region that is processing the thing that is attended. 01:07:05.150 --> 01:07:11.110 align:middle line:84% So this is an example of attention to spatial locations. 01:07:11.110 --> 01:07:12.890 align:middle line:90% This is an fMRI experiment. 01:07:12.890 --> 01:07:16.510 align:middle line:84% And there are two conditions attend 2, 01:07:16.510 --> 01:07:19.270 align:middle line:84% where there are these two locations on the screen that you 01:07:19.270 --> 01:07:20.830 align:middle line:84% have to attend to because there's 01:07:20.830 --> 01:07:24.830 align:middle line:84% stuff happening that you're going to have to respond to. 01:07:24.830 --> 01:07:28.650 align:middle line:84% Here's a case where there's just one location that's attended. 01:07:28.650 --> 01:07:30.970 align:middle line:84% And this location is different from these two. 01:07:30.970 --> 01:07:34.420 align:middle line:90% 01:07:34.420 --> 01:07:37.320 align:middle line:90% So this is a retinotopic map. 01:07:37.320 --> 01:07:38.240 align:middle line:90% So it's a flat map. 01:07:38.240 --> 01:07:43.100 align:middle line:84% It's obtained with fMRI showing the primary visual cortex, 01:07:43.100 --> 01:07:45.220 align:middle line:84% the left hemisphere and the right hemisphere. 01:07:45.220 --> 01:07:48.020 align:middle line:84% This is the foveal representation. 01:07:48.020 --> 01:07:51.940 align:middle line:84% And then these are the regions of the visual cortex that 01:07:51.940 --> 01:07:56.100 align:middle line:84% are representing the four locations at which things 01:07:56.100 --> 01:07:57.800 align:middle line:90% could happen in the experiment. 01:07:57.800 --> 01:08:03.660 align:middle line:84% And the yellow, things are color coded yellow 01:08:03.660 --> 01:08:07.580 align:middle line:84% when there is an increase in the attend 2 01:08:07.580 --> 01:08:09.120 align:middle line:84% condition relative to some baseline. 01:08:09.120 --> 01:08:10.860 align:middle line:84% And they're colored blue when there's 01:08:10.860 --> 01:08:14.580 align:middle line:84% an increase in the attend 1 condition relative to baseline. 01:08:14.580 --> 01:08:22.700 align:middle line:84% And so the locations here, we've got the upper left hemifield 01:08:22.700 --> 01:08:24.279 align:middle line:90% and the lower right hemifield. 01:08:24.279 --> 01:08:28.740 align:middle line:84% So remember how in the visual cortex 01:08:28.740 --> 01:08:30.040 align:middle line:90% things are flipped, right? 01:08:30.040 --> 01:08:33.990 align:middle line:84% So you have the left hemifield represented on the right. 01:08:33.990 --> 01:08:35.689 align:middle line:84% And they're also up-down reversed, 01:08:35.689 --> 01:08:36.790 align:middle line:90% like in the visual cortex. 01:08:36.790 --> 01:08:39.410 align:middle line:84% So the upper hemifield, the representation of that 01:08:39.410 --> 01:08:41.990 align:middle line:84% is below the representation of the lower hemifield. 01:08:41.990 --> 01:08:44.529 align:middle line:84% So that's why you get a blob here 01:08:44.529 --> 01:08:48.149 align:middle line:84% and here that correspond to these two locations. 01:08:48.149 --> 01:08:50.850 align:middle line:84% And then, in this case where there's just this one location 01:08:50.850 --> 01:08:53.770 align:middle line:84% that you're attending, the blob is up here. 01:08:53.770 --> 01:08:57.689 align:middle line:84% So in this case, you're getting these spatially specific 01:08:57.689 --> 01:09:01.689 align:middle line:84% increases in response to attending 01:09:01.689 --> 01:09:03.609 align:middle line:90% to particular locations. 01:09:03.609 --> 01:09:06.970 align:middle line:90% So that's one example. 01:09:06.970 --> 01:09:09.850 align:middle line:84% This is another pretty cool example 01:09:09.850 --> 01:09:14.210 align:middle line:84% from our own Nancy Kanwisher a long time ago at this point. 01:09:14.210 --> 01:09:20.310 align:middle line:84% So in this experiment, people were shown these stimuli here. 01:09:20.310 --> 01:09:23.930 align:middle line:84% So what these are, are superpositions of images 01:09:23.930 --> 01:09:26.490 align:middle line:90% of faces and images of houses. 01:09:26.490 --> 01:09:29.859 align:middle line:84% And so these two types of stimuli, 01:09:29.859 --> 01:09:35.840 align:middle line:84% they were chosen because Nancy had identified these two brain 01:09:35.840 --> 01:09:39.540 align:middle line:84% regions that selectively respond to faces and houses, 01:09:39.540 --> 01:09:41.680 align:middle line:84% so the fusiform face area responds a lot more to 01:09:41.680 --> 01:09:43.439 align:middle line:90% faces than to other stuff. 01:09:43.439 --> 01:09:46.359 align:middle line:84% The parahippocampal place area, or PPA, 01:09:46.359 --> 01:09:48.479 align:middle line:84% it responds a lot more to houses and other kind 01:09:48.479 --> 01:09:54.272 align:middle line:84% of landmark-ish things than other kinds of images. 01:09:54.272 --> 01:09:59.160 align:middle line:84% And in the experiment here, people, 01:09:59.160 --> 01:10:04.880 align:middle line:84% they had to make discriminations about one of the two stimuli. 01:10:04.880 --> 01:10:10.000 align:middle line:84% And in this particular case, either the face or the house 01:10:10.000 --> 01:10:12.780 align:middle line:84% moved back and forth along one axis. 01:10:12.780 --> 01:10:16.800 align:middle line:84% And so people were instructed to either attend 01:10:16.800 --> 01:10:18.560 align:middle line:84% to the thing that was moving or to attend 01:10:18.560 --> 01:10:20.173 align:middle line:90% to the thing that was static. 01:10:20.173 --> 01:10:22.340 align:middle line:84% Because then they were performing some task on them, 01:10:22.340 --> 01:10:24.400 align:middle line:84% probably like detecting whether it repeated 01:10:24.400 --> 01:10:26.680 align:middle line:90% compared to the previous trial. 01:10:26.680 --> 01:10:27.800 align:middle line:90% OK. 01:10:27.800 --> 01:10:28.930 align:middle line:90% What do these graphs show? 01:10:28.930 --> 01:10:32.190 align:middle line:84% They are plotting the responses to these stimuli 01:10:32.190 --> 01:10:35.450 align:middle line:84% in these two different brain regions, the FFA and the PPA. 01:10:35.450 --> 01:10:38.330 align:middle line:84% So the y-axis here is percent signal change. 01:10:38.330 --> 01:10:40.830 align:middle line:84% So when it's higher, that means there's a larger brain 01:10:40.830 --> 01:10:42.530 align:middle line:90% response in that region. 01:10:42.530 --> 01:10:45.490 align:middle line:90% 01:10:45.490 --> 01:10:49.190 align:middle line:84% So there's always both a face and a house on the screen. 01:10:49.190 --> 01:10:51.270 align:middle line:84% But people are attending to one of them. 01:10:51.270 --> 01:10:53.985 align:middle line:90% 01:10:53.985 --> 01:10:57.070 align:middle line:84% So in the stimuli, either the face could be moving 01:10:57.070 --> 01:10:59.550 align:middle line:84% or the house could be moving, and the task 01:10:59.550 --> 01:11:01.430 align:middle line:84% was to pay attention to the moving thing 01:11:01.430 --> 01:11:04.430 align:middle line:84% or to attend to the static thing. 01:11:04.430 --> 01:11:04.930 align:middle line:90% All right. 01:11:04.930 --> 01:11:05.970 align:middle line:90% So what happens here? 01:11:05.970 --> 01:11:09.110 align:middle line:84% So in the situation where the face 01:11:09.110 --> 01:11:13.710 align:middle line:84% is moving, when you have to attend to the moving thing, 01:11:13.710 --> 01:11:15.910 align:middle line:84% the response in the fusiform face area 01:11:15.910 --> 01:11:20.110 align:middle line:84% is higher than when you have to attend to the static thing. 01:11:20.110 --> 01:11:23.350 align:middle line:84% So same stimulus, but you direct attention to one or the other, 01:11:23.350 --> 01:11:26.820 align:middle line:90% and the response gets boosted. 01:11:26.820 --> 01:11:29.680 align:middle line:84% By comparison, in that exact same stimulus condition, 01:11:29.680 --> 01:11:32.400 align:middle line:84% when the face is moving in the PPA, 01:11:32.400 --> 01:11:33.640 align:middle line:90% you see the opposite effect. 01:11:33.640 --> 01:11:35.343 align:middle line:84% So there the response is higher when 01:11:35.343 --> 01:11:37.260 align:middle line:84% you have to attend to the static than when you 01:11:37.260 --> 01:11:38.520 align:middle line:90% have to attend to the moving. 01:11:38.520 --> 01:11:41.546 align:middle line:90% 01:11:41.546 --> 01:11:45.360 align:middle line:84% By comparison, when the house is moving and the face is static, 01:11:45.360 --> 01:11:46.780 align:middle line:90% you get the opposite effect. 01:11:46.780 --> 01:11:49.820 align:middle line:84% So there in the fusiform face area, 01:11:49.820 --> 01:11:52.340 align:middle line:84% the response is higher when you're attending to the static 01:11:52.340 --> 01:11:55.680 align:middle line:84% because that's the one that's the face. 01:11:55.680 --> 01:11:58.260 align:middle line:84% And the response is lower when you're attending to the moving 01:11:58.260 --> 01:12:01.700 align:middle line:90% one because that's the house. 01:12:01.700 --> 01:12:02.200 align:middle line:90% OK. 01:12:02.200 --> 01:12:04.477 align:middle line:84% So the idea is that attention is directed 01:12:04.477 --> 01:12:07.060 align:middle line:84% to either the face or the house, even though they're spatially 01:12:07.060 --> 01:12:07.920 align:middle line:90% superimposed. 01:12:07.920 --> 01:12:09.920 align:middle line:84% So that's what's kind of interesting about this. 01:12:09.920 --> 01:12:12.740 align:middle line:84% So you can select one of these two things 01:12:12.740 --> 01:12:15.400 align:middle line:84% even though they're at the same spatial location. 01:12:15.400 --> 01:12:18.420 align:middle line:90% 01:12:18.420 --> 01:12:20.860 align:middle line:84% So attention is directed to one or the other. 01:12:20.860 --> 01:12:24.220 align:middle line:84% And the response in the brain region that we believe 01:12:24.220 --> 01:12:29.570 align:middle line:84% represents that type of thing then gets boosted. 01:12:29.570 --> 01:12:30.170 align:middle line:90% OK. 01:12:30.170 --> 01:12:33.950 align:middle line:84% So big picture here is that attention to something, 01:12:33.950 --> 01:12:36.450 align:middle line:84% so part of space or a type of feature 01:12:36.450 --> 01:12:38.370 align:middle line:84% or a type of object in this case, 01:12:38.370 --> 01:12:40.450 align:middle line:84% generally enhances neural activity 01:12:40.450 --> 01:12:46.130 align:middle line:84% in the region that's processing whatever is attended. 01:12:46.130 --> 01:12:47.010 align:middle line:90% OK. 01:12:47.010 --> 01:12:51.410 align:middle line:84% So the next thing that I have to tell you about 01:12:51.410 --> 01:12:55.250 align:middle line:84% is going to involve effects of attention 01:12:55.250 --> 01:12:57.110 align:middle line:90% at the single neuron level. 01:12:57.110 --> 01:12:59.770 align:middle line:90% 01:12:59.770 --> 01:13:01.750 align:middle line:84% I think I'm going to call it quits for today. 01:13:01.750 --> 01:13:04.470 align:middle line:84% And we will talk about that on Thursday. 01:13:04.470 --> 01:13:07.570 align:middle line:84% So on Thursday we'll wrap up talking about attention 01:13:07.570 --> 01:13:09.920 align:middle line:90% and then get into touch. 01:13:09.920 --> 01:13:25.000 align:middle line:90%