WEBVTT 00:00:00.000 --> 00:00:02.000 align:middle line:90% [SQUEAKING] 00:00:02.000 --> 00:00:03.500 align:middle line:90% [RUSTLING] 00:00:03.500 --> 00:00:06.000 align:middle line:90% [CLICKING] 00:00:06.000 --> 00:00:10.000 align:middle line:90% 00:00:10.000 --> 00:00:12.800 align:middle line:84% JOSH MCDERMOTT: So the next topic on our agenda 00:00:12.800 --> 00:00:14.260 align:middle line:90% is attention. 00:00:14.260 --> 00:00:19.240 align:middle line:84% And so attention is a-- it's a big word that can 00:00:19.240 --> 00:00:21.920 align:middle line:90% mean a lot of different things. 00:00:21.920 --> 00:00:24.280 align:middle line:84% Colloquially, you're familiar with the notion 00:00:24.280 --> 00:00:29.905 align:middle line:84% of paying attention to something versus maybe ignoring it. 00:00:29.905 --> 00:00:31.280 align:middle line:84% And we're talking about attention 00:00:31.280 --> 00:00:34.480 align:middle line:84% in a class on perception because attention often 00:00:34.480 --> 00:00:38.000 align:middle line:84% determines what you see and hear and in general 00:00:38.000 --> 00:00:39.680 align:middle line:90% perceive about the world. 00:00:39.680 --> 00:00:43.688 align:middle line:84% So in vision, what you're attending to often is 00:00:43.688 --> 00:00:44.980 align:middle line:90% a function of what you look at. 00:00:44.980 --> 00:00:49.800 align:middle line:84% So attention oftentimes moves with your eyes, 00:00:49.800 --> 00:00:51.400 align:middle line:90% but it doesn't have to. 00:00:51.400 --> 00:00:53.140 align:middle line:84% So if you're being really sneaky, 00:00:53.140 --> 00:00:56.200 align:middle line:84% you could be looking at me, but maybe paying attention 00:00:56.200 --> 00:00:58.460 align:middle line:90% to something that's out here. 00:00:58.460 --> 00:01:00.720 align:middle line:84% So you can pay attention to things 00:01:00.720 --> 00:01:03.240 align:middle line:90% if you're not fixating them. 00:01:03.240 --> 00:01:06.240 align:middle line:84% So I guess, naively and intuitively, 00:01:06.240 --> 00:01:08.880 align:middle line:84% when you pay attention to something, 00:01:08.880 --> 00:01:12.520 align:middle line:84% that seems to improve, in some way, the processing 00:01:12.520 --> 00:01:13.740 align:middle line:90% of whatever's attended. 00:01:13.740 --> 00:01:17.500 align:middle line:84% That's like presumably why we do it, why we use attention. 00:01:17.500 --> 00:01:20.620 align:middle line:84% Like when you pay attention to something, you see it better, 00:01:20.620 --> 00:01:22.260 align:middle line:84% or you hear it better in some sense. 00:01:22.260 --> 00:01:24.123 align:middle line:84% And we're, of course, not specifying 00:01:24.123 --> 00:01:25.540 align:middle line:84% what it means for it to be better, 00:01:25.540 --> 00:01:28.160 align:middle line:90% so that's an important issue. 00:01:28.160 --> 00:01:31.360 align:middle line:84% Again, intuitively, people often think of this maybe as a way 00:01:31.360 --> 00:01:32.355 align:middle line:90% to focus "resources." 00:01:32.355 --> 00:01:34.480 align:middle line:84% Again, it's in scare quotes because we don't really 00:01:34.480 --> 00:01:36.720 align:middle line:84% know what the resources are, but it's perhaps 00:01:36.720 --> 00:01:39.093 align:middle line:84% if resources do exist, and you need them for perception, 00:01:39.093 --> 00:01:40.760 align:middle line:84% and they're limited, this could be a way 00:01:40.760 --> 00:01:43.640 align:middle line:84% to focus them on whatever seems important. 00:01:43.640 --> 00:01:46.800 align:middle line:84% So in this lecture, we're going to talk about how we can measure 00:01:46.800 --> 00:01:49.180 align:middle line:90% and study these phenomena. 00:01:49.180 --> 00:01:54.840 align:middle line:84% And so attention has been studied with a dizzying array 00:01:54.840 --> 00:01:56.080 align:middle line:90% of paradigms. 00:01:56.080 --> 00:01:59.060 align:middle line:84% So this lecture will provide an overview of some of them, 00:01:59.060 --> 00:02:02.750 align:middle line:84% some of the best known, that are very relevant to perception. 00:02:02.750 --> 00:02:05.070 align:middle line:84% But it's going to be a little bit of a grab bag. 00:02:05.070 --> 00:02:06.850 align:middle line:84% So right now, we lack, I would say, 00:02:06.850 --> 00:02:09.330 align:middle line:84% a unifying framework for thinking about attention. 00:02:09.330 --> 00:02:11.172 align:middle line:84% And it's almost surely not one thing. 00:02:11.172 --> 00:02:12.630 align:middle line:84% So there's lots of different things 00:02:12.630 --> 00:02:17.470 align:middle line:84% that often get called attention, and they're not necessarily 00:02:17.470 --> 00:02:19.430 align:middle line:90% the same thing. 00:02:19.430 --> 00:02:22.810 align:middle line:84% And at the moment, we mostly, I would say, 00:02:22.810 --> 00:02:24.930 align:middle line:84% lack rigorous computational understanding. 00:02:24.930 --> 00:02:27.550 align:middle line:84% So that's actually a super exciting direction 00:02:27.550 --> 00:02:29.337 align:middle line:90% for the future. 00:02:29.337 --> 00:02:31.670 align:middle line:84% And I think we're at a point now where we could actually 00:02:31.670 --> 00:02:34.230 align:middle line:84% really start to develop rigorous computational theories 00:02:34.230 --> 00:02:35.950 align:middle line:90% of attention. 00:02:35.950 --> 00:02:38.430 align:middle line:84% But there are nonetheless many interesting phenomena 00:02:38.430 --> 00:02:41.230 align:middle line:84% that place constraints on future theoretical understanding. 00:02:41.230 --> 00:02:43.110 align:middle line:84% So we're going to go through some of them 00:02:43.110 --> 00:02:47.750 align:middle line:84% and talk about what they show us and what they mean. 00:02:47.750 --> 00:02:52.510 align:middle line:90% So here's an initial example. 00:02:52.510 --> 00:02:56.150 align:middle line:84% So very popular paradigm for studying attention 00:02:56.150 --> 00:02:58.270 align:middle line:90% involves cuing people. 00:02:58.270 --> 00:03:00.590 align:middle line:84% So Posner was a cognitive psychologist 00:03:00.590 --> 00:03:02.170 align:middle line:90% who did a lot of work on cuing. 00:03:02.170 --> 00:03:06.390 align:middle line:84% So the essential idea is to focus visual attention 00:03:06.390 --> 00:03:09.772 align:middle line:84% to an area of space using what's called a cue. 00:03:09.772 --> 00:03:11.230 align:middle line:84% So that's like a stimulus that will 00:03:11.230 --> 00:03:12.927 align:middle line:84% occur at some part of space, or that 00:03:12.927 --> 00:03:14.510 align:middle line:84% will indicate some part of space where 00:03:14.510 --> 00:03:16.030 align:middle line:90% something's going to happen. 00:03:16.030 --> 00:03:19.310 align:middle line:84% And then in the classical paradigms for studying this, 00:03:19.310 --> 00:03:21.630 align:middle line:84% they would measure the time that it would take someone 00:03:21.630 --> 00:03:26.750 align:middle line:84% to identify a target stimulus when either the observer does 00:03:26.750 --> 00:03:30.430 align:middle line:84% not know where the item will appear versus 00:03:30.430 --> 00:03:32.750 align:middle line:84% where they do know where the item will appear by virtue 00:03:32.750 --> 00:03:33.630 align:middle line:90% of the cue. 00:03:33.630 --> 00:03:36.510 align:middle line:84% And so in this initial example that I'll show, 00:03:36.510 --> 00:03:38.990 align:middle line:84% the cue might just be a briefly presented dot 00:03:38.990 --> 00:03:40.650 align:middle line:90% at the location of the target. 00:03:40.650 --> 00:03:46.070 align:middle line:84% So look in the middle here at the center of the cross. 00:03:46.070 --> 00:03:47.990 align:middle line:84% And so your task is going to be to say 00:03:47.990 --> 00:03:49.850 align:middle line:90% what letter is going to appear. 00:03:49.850 --> 00:03:52.410 align:middle line:84% So you can all do this task, so do it. 00:03:52.410 --> 00:03:55.550 align:middle line:90% 00:03:55.550 --> 00:03:56.650 align:middle line:90% You're really slow. 00:03:56.650 --> 00:03:57.520 align:middle line:90% What's the letter? 00:03:57.520 --> 00:03:58.122 align:middle line:90% AUDIENCE: A 00:03:58.122 --> 00:03:59.080 align:middle line:90% JOSH MCDERMOTT: A. Wow. 00:03:59.080 --> 00:04:00.497 align:middle line:84% That was the longest reaction time 00:04:00.497 --> 00:04:02.878 align:middle line:84% in the history of attention experiments. 00:04:02.878 --> 00:04:03.920 align:middle line:90% You got it right, though. 00:04:03.920 --> 00:04:05.640 align:middle line:90% It is an A. 00:04:05.640 --> 00:04:07.740 align:middle line:84% So this is the uncued condition, and it 00:04:07.740 --> 00:04:10.380 align:middle line:84% is supposed to be slower than the cued condition. 00:04:10.380 --> 00:04:12.260 align:middle line:90% Here's the cued condition. 00:04:12.260 --> 00:04:13.213 align:middle line:90% AUDIENCE: N. 00:04:13.213 --> 00:04:14.880 align:middle line:84% JOSH MCDERMOTT: N. That was much faster. 00:04:14.880 --> 00:04:18.140 align:middle line:90% So that's the cuing benefit. 00:04:18.140 --> 00:04:18.640 align:middle line:90% all 00:04:18.640 --> 00:04:21.880 align:middle line:84% So this is a schematic of the kind of results. 00:04:21.880 --> 00:04:24.440 align:middle line:84% So what is measured here is reaction time. 00:04:24.440 --> 00:04:26.902 align:middle line:84% So again, you assume that most of the time-- and you 00:04:26.902 --> 00:04:29.360 align:middle line:84% verify that most of the time people get the letter correct. 00:04:29.360 --> 00:04:33.260 align:middle line:84% So you measure how long it takes them to identify the letter. 00:04:33.260 --> 00:04:35.260 align:middle line:84% And there's a reduction in reaction time 00:04:35.260 --> 00:04:37.740 align:middle line:84% when you're cued to the location. 00:04:37.740 --> 00:04:42.260 align:middle line:84% And so in this sense, advanced knowledge of the location 00:04:42.260 --> 00:04:46.000 align:middle line:84% improves performance as measured by reaction time. 00:04:46.000 --> 00:04:49.660 align:middle line:84% So the other thing that is associated with this phenomena 00:04:49.660 --> 00:04:51.060 align:middle line:90% is that there is-- 00:04:51.060 --> 00:04:52.600 align:middle line:90% it's spatially localized. 00:04:52.600 --> 00:04:55.580 align:middle line:84% So this is, again, kind of a fake graph 00:04:55.580 --> 00:04:58.660 align:middle line:84% schematized the results of lots and lots of experiments. 00:04:58.660 --> 00:05:01.860 align:middle line:84% And the graph is plotting the reduction in reaction time 00:05:01.860 --> 00:05:08.540 align:middle line:84% that you get from the cue as a function of the cued location 00:05:08.540 --> 00:05:10.060 align:middle line:90% relative to the target. 00:05:10.060 --> 00:05:12.598 align:middle line:84% And so the point is that when the cue is exactly 00:05:12.598 --> 00:05:14.140 align:middle line:84% at the location of the target, that's 00:05:14.140 --> 00:05:15.680 align:middle line:84% where you get the biggest benefit. 00:05:15.680 --> 00:05:19.180 align:middle line:84% And then if the target is off of the cued location 00:05:19.180 --> 00:05:23.240 align:middle line:84% by some amount, there's a reduction in the benefit. 00:05:23.240 --> 00:05:27.540 align:middle line:84% So this type of finding led to a very influential and popular 00:05:27.540 --> 00:05:30.020 align:middle line:84% metaphor of attention as a spotlight. 00:05:30.020 --> 00:05:32.180 align:middle line:84% So the idea is that there's this thing that's 00:05:32.180 --> 00:05:33.520 align:middle line:90% spatially localized. 00:05:33.520 --> 00:05:36.220 align:middle line:84% It gets cued to a location, it gets moved to that location, 00:05:36.220 --> 00:05:37.920 align:middle line:84% does something that is, as of now, 00:05:37.920 --> 00:05:41.300 align:middle line:84% unspecified to the processing that makes you faster or more 00:05:41.300 --> 00:05:42.760 align:middle line:90% accurate or whatever. 00:05:42.760 --> 00:05:46.720 align:middle line:84% So the spotlight metaphor of attention. 00:05:46.720 --> 00:05:49.460 align:middle line:84% So the idea is that whatever's in the spotlight is attended. 00:05:49.460 --> 00:05:51.628 align:middle line:84% The more it's attended, the better it's processed. 00:05:51.628 --> 00:05:53.420 align:middle line:84% And the size and the shape of the spotlight 00:05:53.420 --> 00:05:56.540 align:middle line:84% can be controlled to some extent. 00:05:56.540 --> 00:05:58.500 align:middle line:84% So now another really important distinction 00:05:58.500 --> 00:06:01.580 align:middle line:84% that came out of this sort of study 00:06:01.580 --> 00:06:04.580 align:middle line:84% is the idea that there can be two types of cues. 00:06:04.580 --> 00:06:08.300 align:middle line:84% So the one that we just saw is what would be normally referred 00:06:08.300 --> 00:06:10.000 align:middle line:90% to as exogenous. 00:06:10.000 --> 00:06:13.380 align:middle line:84% So that means outside generating. 00:06:13.380 --> 00:06:18.280 align:middle line:84% So the dot flashes up, without really thinking about it, 00:06:18.280 --> 00:06:19.780 align:middle line:84% like your attention is kind of drawn 00:06:19.780 --> 00:06:22.380 align:middle line:84% to the location of this thing that pops up on the screen. 00:06:22.380 --> 00:06:24.760 align:middle line:90% It's almost like a reflex. 00:06:24.760 --> 00:06:29.040 align:middle line:84% So typically, exogenous cues would be sudden changes, 00:06:29.040 --> 00:06:30.820 align:middle line:90% so flashes or movement. 00:06:30.820 --> 00:06:33.540 align:middle line:84% They draw attention automatically. 00:06:33.540 --> 00:06:37.140 align:middle line:84% So by comparison, the other kind of cue 00:06:37.140 --> 00:06:40.920 align:middle line:84% that you can get in general is what's called endogenous. 00:06:40.920 --> 00:06:45.460 align:middle line:84% So that stands for inside generating So this typically we 00:06:45.460 --> 00:06:48.300 align:middle line:84% think requires some kind of high-level control, 00:06:48.300 --> 00:06:50.440 align:middle line:84% and typically involves an instruction. 00:06:50.440 --> 00:06:52.300 align:middle line:84% So some kind of visual sign or pattern 00:06:52.300 --> 00:06:57.750 align:middle line:84% that will be a symbol of where the target is going to appear. 00:06:57.750 --> 00:07:01.090 align:middle line:84% So here's an example of exogenous cuing. 00:07:01.090 --> 00:07:03.130 align:middle line:84% Sorry, this is the exogenous one that we just 00:07:03.130 --> 00:07:05.250 align:middle line:90% saw, just to repeat it. 00:07:05.250 --> 00:07:07.895 align:middle line:84% So there's the dot, and then the letter So you 00:07:07.895 --> 00:07:09.270 align:middle line:84% don't have to interpret anything. 00:07:09.270 --> 00:07:10.790 align:middle line:84% Your attention just does its thing. 00:07:10.790 --> 00:07:13.930 align:middle line:84% Here's an example of endogenous cuing. 00:07:13.930 --> 00:07:16.930 align:middle line:84% So we've got an arrow, and then the letter appears. 00:07:16.930 --> 00:07:20.690 align:middle line:84% So the point is that in order for you to benefit from the cue, 00:07:20.690 --> 00:07:22.630 align:middle line:84% you have to understand what arrows mean. 00:07:22.630 --> 00:07:27.210 align:middle line:84% And willfully move your attention over 00:07:27.210 --> 00:07:29.450 align:middle line:84% in that direction, and then you get a benefit. 00:07:29.450 --> 00:07:33.350 align:middle line:90% 00:07:33.350 --> 00:07:36.330 align:middle line:84% So both types of cues seem to control 00:07:36.330 --> 00:07:39.170 align:middle line:84% an attentional mechanism-- maybe not the same one. 00:07:39.170 --> 00:07:41.410 align:middle line:84% But they reflect different strategies. 00:07:41.410 --> 00:07:43.330 align:middle line:84% So we think of exogenous cues as tapping 00:07:43.330 --> 00:07:45.430 align:middle line:84% into some kind of bottom-up control of attention 00:07:45.430 --> 00:07:47.830 align:middle line:84% that's based on physical transients in the environment, 00:07:47.830 --> 00:07:50.650 align:middle line:84% and endogenous cues as involving what's 00:07:50.650 --> 00:07:53.210 align:middle line:84% often called top-down control of attention 00:07:53.210 --> 00:07:57.110 align:middle line:84% based on what an observer believes. 00:07:57.110 --> 00:08:00.290 align:middle line:84% So how do these things actually cash out 00:08:00.290 --> 00:08:01.350 align:middle line:90% in terms of the results? 00:08:01.350 --> 00:08:04.610 align:middle line:84% Well, one of the clear differences that you see 00:08:04.610 --> 00:08:06.570 align:middle line:90% is what's shown here. 00:08:06.570 --> 00:08:11.828 align:middle line:84% So this is a graph that is plotting the same thing 00:08:11.828 --> 00:08:12.870 align:middle line:90% that we looked at before. 00:08:12.870 --> 00:08:17.030 align:middle line:84% So the y-axis is the benefit in reaction time from a valid cue. 00:08:17.030 --> 00:08:20.250 align:middle line:84% So it's how much faster you are if you're cued to the location. 00:08:20.250 --> 00:08:22.850 align:middle line:84% But now what's being manipulated here 00:08:22.850 --> 00:08:25.890 align:middle line:84% is what's called the stimulus onset asynchrony. 00:08:25.890 --> 00:08:29.330 align:middle line:90% And so that's indicated here. 00:08:29.330 --> 00:08:32.650 align:middle line:84% So here's what would happen on a trial of an experiment 00:08:32.650 --> 00:08:34.710 align:middle line:84% where you're measuring this stuff. 00:08:34.710 --> 00:08:36.970 align:middle line:84% This is a case where the cue is exogenous. 00:08:36.970 --> 00:08:41.030 align:middle line:84% So you're fixating that little star, and then the cue pops up. 00:08:41.030 --> 00:08:42.470 align:middle line:90% In this case, it's this red box. 00:08:42.470 --> 00:08:44.169 align:middle line:84% So something changes, your attention 00:08:44.169 --> 00:08:46.130 align:middle line:90% moves over to the left. 00:08:46.130 --> 00:08:49.710 align:middle line:84% And then after some amount of time, the target appears. 00:08:49.710 --> 00:08:53.450 align:middle line:84% And that some amount of time is the SOA, the Stimulus Onset 00:08:53.450 --> 00:08:54.570 align:middle line:90% Asynchrony. 00:08:54.570 --> 00:08:56.810 align:middle line:84% So essentially, you can vary the amount of time 00:08:56.810 --> 00:08:59.410 align:middle line:84% from when the cue pops up until when that thing that you 00:08:59.410 --> 00:09:02.930 align:middle line:90% have to detect pops up. 00:09:02.930 --> 00:09:05.110 align:middle line:84% And so what does this graph show? 00:09:05.110 --> 00:09:09.050 align:middle line:90% Well, it shows that initially-- 00:09:09.050 --> 00:09:11.250 align:middle line:84% so if the stimulus onset asynchrony 00:09:11.250 --> 00:09:14.370 align:middle line:84% is really, really short, then you don't get the benefit. 00:09:14.370 --> 00:09:17.010 align:middle line:84% So it's like the cue does something that takes 00:09:17.010 --> 00:09:20.350 align:middle line:90% some amount of time to happen. 00:09:20.350 --> 00:09:24.550 align:middle line:84% And then in the case of what's called the peripheral cue here, 00:09:24.550 --> 00:09:26.750 align:middle line:84% that's the exogenous one, the red box. 00:09:26.750 --> 00:09:31.210 align:middle line:84% By the time you-- there's about 100 milliseconds, 00:09:31.210 --> 00:09:34.010 align:middle line:84% you're close to getting the maximal benefit from the cue. 00:09:34.010 --> 00:09:37.090 align:middle line:84% But when you have a symbolic cue, what 00:09:37.090 --> 00:09:40.110 align:middle line:84% we're calling endogenous, where you have to interpret it, 00:09:40.110 --> 00:09:41.210 align:middle line:90% that's like the arrow. 00:09:41.210 --> 00:09:46.370 align:middle line:84% It takes you longer to get the benefit because something else 00:09:46.370 --> 00:09:48.010 align:middle line:90% has to happen in your head. 00:09:48.010 --> 00:09:51.080 align:middle line:84% You see this symbol, and there's some process of interpretation, 00:09:51.080 --> 00:09:52.540 align:middle line:90% and that takes a bit more time. 00:09:52.540 --> 00:09:56.160 align:middle line:84% So this is some evidence that these two types of cues 00:09:56.160 --> 00:10:00.840 align:middle line:84% are doing different things in your head 00:10:00.840 --> 00:10:02.100 align:middle line:90% Any questions about that? 00:10:02.100 --> 00:10:12.360 align:middle line:90% 00:10:12.360 --> 00:10:13.760 align:middle line:84% And both of these types of things 00:10:13.760 --> 00:10:15.420 align:middle line:84% happen all the time in everyday life. 00:10:15.420 --> 00:10:19.720 align:middle line:84% So there's very frequently transient things happen, 00:10:19.720 --> 00:10:22.520 align:middle line:84% attention is drawn to some location-- 00:10:22.520 --> 00:10:24.230 align:middle line:84% because that's how we talk about it. 00:10:24.230 --> 00:10:26.480 align:middle line:84% Other times there'll be something that's more symbolic 00:10:26.480 --> 00:10:29.978 align:middle line:84% that indicates to you-- so, for instance-- 00:10:29.978 --> 00:10:31.520 align:middle line:84% I don't know-- you see a turn signal. 00:10:31.520 --> 00:10:33.940 align:middle line:84% It indicates the thing's going to turn to the left, 00:10:33.940 --> 00:10:35.220 align:middle line:84% and so you're paying attention over there. 00:10:35.220 --> 00:10:35.720 align:middle line:90% Yeah. 00:10:35.720 --> 00:10:38.200 align:middle line:84% AUDIENCE: Is there are always these binary difference 00:10:38.200 --> 00:10:42.080 align:middle line:84% between the exogenous and the other one? 00:10:42.080 --> 00:10:45.080 align:middle line:84% Or could you imagine an arrow that's 00:10:45.080 --> 00:10:49.640 align:middle line:84% pointing upward, but the visual stimulus is immediately there as 00:10:49.640 --> 00:10:50.140 align:middle line:90% well? 00:10:50.140 --> 00:10:52.240 align:middle line:90% So it's orienting you. 00:10:52.240 --> 00:10:56.085 align:middle line:84% is there some of, I guess, gray area? 00:10:56.085 --> 00:10:56.960 align:middle line:90% JOSH MCDERMOTT: Yeah. 00:10:56.960 --> 00:11:00.980 align:middle line:84% I mean, I think that often in a lot of real-world scenarios, 00:11:00.980 --> 00:11:03.340 align:middle line:84% you're getting combinations of these two things, 00:11:03.340 --> 00:11:06.000 align:middle line:84% where there'll be some transient, 00:11:06.000 --> 00:11:08.680 align:middle line:84% but you also just-- because of your understanding 00:11:08.680 --> 00:11:10.800 align:middle line:84% of the situation, you know that something's going 00:11:10.800 --> 00:11:15.098 align:middle line:90% to happen in a particular area. 00:11:15.098 --> 00:11:15.640 align:middle line:90% I don't know. 00:11:15.640 --> 00:11:17.160 align:middle line:84% Let's say you're watching baseball. 00:11:17.160 --> 00:11:18.827 align:middle line:84% And so the pitcher is throwing the ball. 00:11:18.827 --> 00:11:20.920 align:middle line:84% So you know that it's going to be 00:11:20.920 --> 00:11:23.980 align:middle line:84% caught in some region around the strike zone. 00:11:23.980 --> 00:11:27.280 align:middle line:84% But then the catcher puts up the mitt, 00:11:27.280 --> 00:11:29.860 align:middle line:84% and you hear the sound of the ball slapping in. 00:11:29.860 --> 00:11:32.058 align:middle line:84% And then you get this more exogenous cue 00:11:32.058 --> 00:11:33.100 align:middle line:90% for where it's happening. 00:11:33.100 --> 00:11:35.860 align:middle line:84% So that's like a combination of things right. 00:11:35.860 --> 00:11:38.940 align:middle line:84% So I think that kind of stuff happens a lot. 00:11:38.940 --> 00:11:39.440 align:middle line:90% Yeah. 00:11:39.440 --> 00:11:41.732 align:middle line:84% I don't think these are necessarily mutually exclusive. 00:11:41.732 --> 00:11:44.280 align:middle line:90% 00:11:44.280 --> 00:11:46.720 align:middle line:84% So that's a very common distinction 00:11:46.720 --> 00:11:51.110 align:middle line:84% in the world of attention, exogenous and endogenous 00:11:51.110 --> 00:11:53.130 align:middle line:90% attention. 00:11:53.130 --> 00:11:56.550 align:middle line:84% So these kinds of cuing studies, they gave rise to this idea 00:11:56.550 --> 00:11:59.050 align:middle line:84% that you could think of attention as a spotlight. 00:11:59.050 --> 00:12:03.710 align:middle line:84% And so that was very persistent for several decades. 00:12:03.710 --> 00:12:09.070 align:middle line:84% And then people got very interested in whether or not 00:12:09.070 --> 00:12:11.350 align:middle line:84% you could actually attend to multiple things 00:12:11.350 --> 00:12:12.450 align:middle line:90% at the same time. 00:12:12.450 --> 00:12:15.110 align:middle line:84% And so the paradigm that is shown here 00:12:15.110 --> 00:12:18.270 align:middle line:84% became wildly popular in the world of visual attention. 00:12:18.270 --> 00:12:21.070 align:middle line:84% This is what's often called multiple object tracking. 00:12:21.070 --> 00:12:22.830 align:middle line:84% And so what happens in this paradigm 00:12:22.830 --> 00:12:25.150 align:middle line:84% is there's a bunch of objects that are on a screen. 00:12:25.150 --> 00:12:29.330 align:middle line:84% And you are cued to attend to a subset of them. 00:12:29.330 --> 00:12:31.393 align:middle line:84% So in this case, the red ones are the ones 00:12:31.393 --> 00:12:32.810 align:middle line:84% that you're supposed to attend to. 00:12:32.810 --> 00:12:35.390 align:middle line:90% So you get this initial cue. 00:12:35.390 --> 00:12:37.490 align:middle line:90% Then the cue disappears. 00:12:37.490 --> 00:12:40.090 align:middle line:84% And at this point, all of the objects are basically identical. 00:12:40.090 --> 00:12:42.110 align:middle line:84% So the only thing that distinguishes them 00:12:42.110 --> 00:12:43.950 align:middle line:84% is the fact that you know that you're 00:12:43.950 --> 00:12:46.510 align:middle line:84% supposed to be paying attention to a subset of them. 00:12:46.510 --> 00:12:49.690 align:middle line:90% Then they start moving around. 00:12:49.690 --> 00:12:51.470 align:middle line:90% So they move around for a while. 00:12:51.470 --> 00:12:55.930 align:middle line:84% And then at the end, one of them is going to turn red again. 00:12:55.930 --> 00:12:57.950 align:middle line:84% And you have to say, is the one that 00:12:57.950 --> 00:13:00.990 align:middle line:84% turned red, is that part of the initial set of red balls, 00:13:00.990 --> 00:13:02.470 align:middle line:90% or is it not? 00:13:02.470 --> 00:13:05.150 align:middle line:84% So it's a measure of the extent to which you can keep track 00:13:05.150 --> 00:13:08.250 align:middle line:90% of all these different objects. 00:13:08.250 --> 00:13:10.470 align:middle line:90% Let's see if you can do it. 00:13:10.470 --> 00:13:13.670 align:middle line:84% So you need to fixate that little square at the center. 00:13:13.670 --> 00:13:16.786 align:middle line:90% 00:13:16.786 --> 00:13:20.230 align:middle line:84% The flashing ones are the ones you got to pay attention to. 00:13:20.230 --> 00:13:21.890 align:middle line:84% And now they're going to move around. 00:13:21.890 --> 00:13:31.790 align:middle line:90% 00:13:31.790 --> 00:13:34.350 align:middle line:84% So did you feel like you could pay attention 00:13:34.350 --> 00:13:35.650 align:middle line:90% to a bunch of them? 00:13:35.650 --> 00:13:36.390 align:middle line:90% Yeah. 00:13:36.390 --> 00:13:41.870 align:middle line:84% You also probably noticed that gets pretty hard when 00:13:41.870 --> 00:13:44.470 align:middle line:90% two things kind of overlap. 00:13:44.470 --> 00:13:46.470 align:middle line:84% So one of the things that you're trying to track 00:13:46.470 --> 00:13:49.772 align:middle line:84% is-- crosses paths with one of the other ones. 00:13:49.772 --> 00:13:50.730 align:middle line:90% That gets kind of hard. 00:13:50.730 --> 00:13:52.438 align:middle line:84% So there's all kinds of interesting stuff 00:13:52.438 --> 00:13:54.870 align:middle line:90% that happens here. 00:13:54.870 --> 00:13:57.190 align:middle line:84% But people have some ability to do this. 00:13:57.190 --> 00:14:00.590 align:middle line:84% This paradigm was introduced by Zenon Pylyshyn, 00:14:00.590 --> 00:14:03.750 align:middle line:84% who is a famous cognitive psychologist at Rutgers. 00:14:03.750 --> 00:14:07.830 align:middle line:84% This is a graph that shows the proportion of errors 00:14:07.830 --> 00:14:10.110 align:middle line:84% that are made as a function of the number of targets 00:14:10.110 --> 00:14:12.310 align:middle line:90% that people are tracking. 00:14:12.310 --> 00:14:14.390 align:middle line:84% And you can see that people get worse 00:14:14.390 --> 00:14:16.073 align:middle line:84% as the number of targets increases. 00:14:16.073 --> 00:14:17.990 align:middle line:84% But the point is that the proportion of errors 00:14:17.990 --> 00:14:21.430 align:middle line:84% is still relatively low, even out to five things. 00:14:21.430 --> 00:14:24.750 align:middle line:84% So people seem to have the ability 00:14:24.750 --> 00:14:30.020 align:middle line:84% to track multiple things at the same time. 00:14:30.020 --> 00:14:33.030 align:middle line:84% And so this paradigm kind of became a little cottage 00:14:33.030 --> 00:14:33.690 align:middle line:90% industry. 00:14:33.690 --> 00:14:37.790 align:middle line:84% And there were many, many, many experiments measuring 00:14:37.790 --> 00:14:41.470 align:middle line:84% how people could do this under different conditions and things 00:14:41.470 --> 00:14:42.310 align:middle line:90% like that. 00:14:42.310 --> 00:14:44.870 align:middle line:84% So the general conclusions from this 00:14:44.870 --> 00:14:50.460 align:middle line:84% are that people can attend to multiple locations at once. 00:14:50.460 --> 00:14:52.800 align:middle line:84% So attention is not just a spotlight, 00:14:52.800 --> 00:14:54.520 align:middle line:84% maybe there are multiple spotlights. 00:14:54.520 --> 00:14:57.540 align:middle line:90% 00:14:57.540 --> 00:14:59.840 align:middle line:84% It's still the case, though, that in a lot of cases, 00:14:59.840 --> 00:15:01.280 align:middle line:84% the spotlight metaphor works pretty well. 00:15:01.280 --> 00:15:02.905 align:middle line:84% So there's often kind of one main thing 00:15:02.905 --> 00:15:05.260 align:middle line:90% that you're attending to. 00:15:05.260 --> 00:15:07.320 align:middle line:84% Any questions about multiple object tracking? 00:15:07.320 --> 00:15:14.940 align:middle line:90% 00:15:14.940 --> 00:15:15.540 align:middle line:90% Yeah. 00:15:15.540 --> 00:15:16.623 align:middle line:90% AUDIENCE: Is it learnable. 00:15:16.623 --> 00:15:18.940 align:middle line:84% Like can you train to get better at it? 00:15:18.940 --> 00:15:20.940 align:middle line:84% JOSH MCDERMOTT: So can you train to get better? 00:15:20.940 --> 00:15:26.300 align:middle line:90% So-- probably. 00:15:26.300 --> 00:15:31.880 align:middle line:84% I'm pretty sure that there's a study that I know about, 00:15:31.880 --> 00:15:35.780 align:middle line:84% which I hope is not apocryphal, but I believe 00:15:35.780 --> 00:15:39.940 align:middle line:90% these experiments were ran on-- 00:15:39.940 --> 00:15:42.900 align:middle line:84% I think it was like the Canadian Olympic basketball team. 00:15:42.900 --> 00:15:45.568 align:middle line:84% So basketball is a great example where you maybe 00:15:45.568 --> 00:15:47.860 align:middle line:84% have to do something kind of like this, because there's 00:15:47.860 --> 00:15:48.940 align:middle line:84% all these different players, and you 00:15:48.940 --> 00:15:51.520 align:middle line:84% need to keep track of who's where and stuff like that. 00:15:51.520 --> 00:15:54.940 align:middle line:84% And so I believe that there was a finding 00:15:54.940 --> 00:15:57.620 align:middle line:84% that the capacity for multiple object tracking 00:15:57.620 --> 00:16:00.080 align:middle line:84% was better on people who were really good at basketball. 00:16:00.080 --> 00:16:03.140 align:middle line:90% 00:16:03.140 --> 00:16:05.620 align:middle line:84% So I think there's been some studies like that showing 00:16:05.620 --> 00:16:07.620 align:middle line:84% that some people are better than others. 00:16:07.620 --> 00:16:09.800 align:middle line:84% I don't know how trainable that is. 00:16:09.800 --> 00:16:12.632 align:middle line:84% It could also just be that the people who are good at multiple 00:16:12.632 --> 00:16:15.340 align:middle line:84% object tracking end up being the ones that are good at basketball 00:16:15.340 --> 00:16:15.882 align:middle line:90% or something. 00:16:15.882 --> 00:16:17.320 align:middle line:90% So yeah. 00:16:17.320 --> 00:16:19.880 align:middle line:90% 00:16:19.880 --> 00:16:23.140 align:middle line:84% The other thing that is related to this 00:16:23.140 --> 00:16:26.460 align:middle line:84% is there's, again, another small cottage 00:16:26.460 --> 00:16:30.820 align:middle line:84% industry of studies on people who play lots of video games. 00:16:30.820 --> 00:16:36.260 align:middle line:84% And in general, the finding is that people 00:16:36.260 --> 00:16:37.660 align:middle line:84% who play lots of video games seem 00:16:37.660 --> 00:16:40.038 align:middle line:84% to do better on a lot of visual attention tasks. 00:16:40.038 --> 00:16:41.580 align:middle line:84% I don't know if they were ever tested 00:16:41.580 --> 00:16:43.860 align:middle line:84% on this one in particular, but some 00:16:43.860 --> 00:16:47.940 align:middle line:84% of the other effects that we will see later in the lecture 00:16:47.940 --> 00:16:51.080 align:middle line:84% show benefits from video game playing. 00:16:51.080 --> 00:16:56.780 align:middle line:90% 00:16:56.780 --> 00:17:01.240 align:middle line:84% So lots of attention research has been conducted in vision, 00:17:01.240 --> 00:17:05.780 align:middle line:84% but it's also-- it's really important for hearing as well. 00:17:05.780 --> 00:17:08.618 align:middle line:84% You remember back when we were talking about audition, 00:17:08.618 --> 00:17:10.660 align:middle line:84% we talked a lot about the cocktail party problem, 00:17:10.660 --> 00:17:13.040 align:middle line:84% where there's one person you're trying to understand, 00:17:13.040 --> 00:17:14.900 align:middle line:84% so you have to pay attention to their voice, 00:17:14.900 --> 00:17:17.660 align:middle line:90% omit these distractors. 00:17:17.660 --> 00:17:21.900 align:middle line:84% And so one situation where attention may be important 00:17:21.900 --> 00:17:25.520 align:middle line:84% is where there are concurrent sources that are similar, 00:17:25.520 --> 00:17:26.925 align:middle line:84% and you have to keep track of one 00:17:26.925 --> 00:17:28.300 align:middle line:84% that you're trying to understand. 00:17:28.300 --> 00:17:32.820 align:middle line:84% And so this is a graph that is showing you 00:17:32.820 --> 00:17:37.340 align:middle line:84% the trajectories that are taken by two voices that are-- 00:17:37.340 --> 00:17:41.780 align:middle line:84% two people talking at the same time through a space of features 00:17:41.780 --> 00:17:44.410 align:middle line:84% that we think are important for voices. 00:17:44.410 --> 00:17:47.502 align:middle line:84% So F0 is the fundamental frequency. 00:17:47.502 --> 00:17:49.710 align:middle line:84% Everybody remembers the fundamental frequency, right? 00:17:49.710 --> 00:17:51.668 align:middle line:84% It's thought to be like the correlate of pitch. 00:17:51.668 --> 00:17:53.823 align:middle line:90% It's the rate of repetition. 00:17:53.823 --> 00:17:55.490 align:middle line:84% And then the other thing-- another thing 00:17:55.490 --> 00:17:57.740 align:middle line:84% that's really important about voices are the formants. 00:17:57.740 --> 00:18:02.890 align:middle line:84% So the first two formants define vowels, the first order, so F1 00:18:02.890 --> 00:18:04.490 align:middle line:90% and F2. 00:18:04.490 --> 00:18:08.050 align:middle line:84% And so when you talk, you're constantly 00:18:08.050 --> 00:18:10.810 align:middle line:84% modulating your fundamental frequency and your formants. 00:18:10.810 --> 00:18:12.810 align:middle line:84% And so you can think of a voice as kind 00:18:12.810 --> 00:18:15.790 align:middle line:84% of moving around in that three-dimensional feature space. 00:18:15.790 --> 00:18:18.770 align:middle line:84% And so these are example trajectories. 00:18:18.770 --> 00:18:20.605 align:middle line:84% The yellow line is for one voice, 00:18:20.605 --> 00:18:22.230 align:middle line:84% and the blue line is for another voice. 00:18:22.230 --> 00:18:25.370 align:middle line:84% And so the point is just that they're all kind of tangled up. 00:18:25.370 --> 00:18:27.840 align:middle line:90% So this is like two-- 00:18:27.840 --> 00:18:30.010 align:middle line:84% I think there are two female speakers that 00:18:30.010 --> 00:18:31.350 align:middle line:90% are talking at the same time. 00:18:31.350 --> 00:18:35.702 align:middle line:84% And so naively, you might think that if you 00:18:35.702 --> 00:18:37.410 align:middle line:84% want to follow what one person is saying, 00:18:37.410 --> 00:18:40.290 align:middle line:84% you might need to track that voice as it moves 00:18:40.290 --> 00:18:41.690 align:middle line:90% through the feature space. 00:18:41.690 --> 00:18:44.770 align:middle line:84% And so this is an example task that 00:18:44.770 --> 00:18:48.548 align:middle line:90% was devised to look at this. 00:18:48.548 --> 00:18:50.590 align:middle line:84% I'm going to play you an example of the stimulus, 00:18:50.590 --> 00:18:52.048 align:middle line:84% but these are synthetic voices that 00:18:52.048 --> 00:18:54.790 align:middle line:90% are continuously voiced vowels. 00:18:54.790 --> 00:18:57.590 align:middle line:84% So they move around in this space. 00:18:57.590 --> 00:18:59.490 align:middle line:84% And there's going to be two of them. 00:18:59.490 --> 00:19:02.250 align:middle line:84% And this is a task that was devised 00:19:02.250 --> 00:19:06.650 align:middle line:84% to measure whether or not you can track one of the two voices. 00:19:06.650 --> 00:19:11.030 align:middle line:84% And so the way that this works is you initially get a cue, 00:19:11.030 --> 00:19:12.970 align:middle line:84% and that tells you which of the two voices 00:19:12.970 --> 00:19:14.430 align:middle line:90% you're supposed to listen to. 00:19:14.430 --> 00:19:19.130 align:middle line:84% And that cue is the very initial part of the target voice, 00:19:19.130 --> 00:19:21.070 align:middle line:90% so the green one in this case. 00:19:21.070 --> 00:19:23.553 align:middle line:84% So then you get a mixture of two voices. 00:19:23.553 --> 00:19:25.970 align:middle line:84% And so you can see this is like two harmonic things on top 00:19:25.970 --> 00:19:26.630 align:middle line:90% of each other. 00:19:26.630 --> 00:19:29.173 align:middle line:84% It's just spectrograms, obviously. 00:19:29.173 --> 00:19:31.590 align:middle line:84% And then after the mixture, you get what's called a probe. 00:19:31.590 --> 00:19:34.970 align:middle line:84% And so the probe is the end portion of one 00:19:34.970 --> 00:19:36.410 align:middle line:90% of the two voices. 00:19:36.410 --> 00:19:38.330 align:middle line:84% So half the time, it will be the end portion 00:19:38.330 --> 00:19:40.658 align:middle line:84% from the green voice, the target voice. 00:19:40.658 --> 00:19:43.200 align:middle line:84% Half the time it will be the end portion from the other voice 00:19:43.200 --> 00:19:44.492 align:middle line:90% that you're supposed to ignore. 00:19:44.492 --> 00:19:47.520 align:middle line:84% And so you just have to say, yes, 00:19:47.520 --> 00:19:50.240 align:middle line:84% the probe is from the cued voice, or, no, it isn't. 00:19:50.240 --> 00:19:54.040 align:middle line:84% And the idea is that because these things kind of are 00:19:54.040 --> 00:19:56.700 align:middle line:84% circling around each other in the same part of the space, 00:19:56.700 --> 00:19:58.920 align:middle line:84% in order to perform this task, you 00:19:58.920 --> 00:20:00.860 align:middle line:84% have to be able to track this thing over time 00:20:00.860 --> 00:20:02.440 align:middle line:84% to essentially be able to connect 00:20:02.440 --> 00:20:06.160 align:middle line:84% the cued portion of the voice to the end portion of the voice. 00:20:06.160 --> 00:20:09.480 align:middle line:90% So you can try it for yourself. 00:20:09.480 --> 00:20:11.180 align:middle line:84% You're initially going to hear the cue, 00:20:11.180 --> 00:20:13.688 align:middle line:84% and then you're supposed to track the cued voice. 00:20:13.688 --> 00:20:14.980 align:middle line:90% And then listen to how it ends. 00:20:14.980 --> 00:20:16.360 align:middle line:84% And then you get a probe, and you 00:20:16.360 --> 00:20:19.120 align:middle line:84% have to say whether the probe was the one at the end or not-- 00:20:19.120 --> 00:20:22.013 align:middle line:84% was the end of the target voice, rather. 00:20:22.013 --> 00:20:22.680 align:middle line:90% [AUDIO PLAYBACK] 00:20:22.680 --> 00:20:24.885 align:middle line:90% [OVERLAPPING VOCALIZATIONS] 00:20:24.885 --> 00:20:25.760 align:middle line:90% [SINGLE VOCALIZATION] 00:20:25.760 --> 00:20:26.577 align:middle line:90% [END PLAYBACK] 00:20:26.577 --> 00:20:27.660 align:middle line:90% JOSH MCDERMOTT: Yes or no? 00:20:27.660 --> 00:20:28.520 align:middle line:90% AUDIENCE: No. 00:20:28.520 --> 00:20:29.395 align:middle line:90% JOSH MCDERMOTT: Yeah? 00:20:29.395 --> 00:20:29.960 align:middle line:90% OK, good. 00:20:29.960 --> 00:20:31.020 align:middle line:90% Some of you're not sure. 00:20:31.020 --> 00:20:31.820 align:middle line:90% Here's another one. 00:20:31.820 --> 00:20:32.487 align:middle line:90% [AUDIO PLAYBACK] 00:20:32.487 --> 00:20:35.415 align:middle line:90% [OVERLAPPING VOCALIZATIONS] 00:20:35.415 --> 00:20:36.325 align:middle line:90% [SINGLE VOCALIZATION] 00:20:36.325 --> 00:20:36.985 align:middle line:90% [END PLAYBACK] 00:20:36.985 --> 00:20:37.860 align:middle line:90% JOSH MCDERMOTT: Yeah. 00:20:37.860 --> 00:20:40.960 align:middle line:84% So in both of those cases, it was. 00:20:40.960 --> 00:20:45.440 align:middle line:84% So, interestingly, it turns out that this task 00:20:45.440 --> 00:20:49.080 align:middle line:90% is much easier for musicians. 00:20:49.080 --> 00:20:53.200 align:middle line:84% And I think many of you who were smiling 00:20:53.200 --> 00:20:55.880 align:middle line:90% are probably musically trained. 00:20:55.880 --> 00:20:57.100 align:middle line:90% But people can do this. 00:20:57.100 --> 00:21:00.560 align:middle line:90% 00:21:00.560 --> 00:21:03.320 align:middle line:84% So one question is like, well, it 00:21:03.320 --> 00:21:05.007 align:middle line:84% seems like in order to do this task, 00:21:05.007 --> 00:21:06.840 align:middle line:84% you would have to track this thing over time 00:21:06.840 --> 00:21:07.820 align:middle line:90% with your attention. 00:21:07.820 --> 00:21:10.343 align:middle line:84% And one way to potentially measure this, 00:21:10.343 --> 00:21:12.760 align:middle line:84% and a conventional way to measure whether you're attending 00:21:12.760 --> 00:21:15.520 align:middle line:84% to something, is to see whether there 00:21:15.520 --> 00:21:18.800 align:middle line:84% is some other kind of attentional benefit to attending 00:21:18.800 --> 00:21:19.660 align:middle line:90% to that thing. 00:21:19.660 --> 00:21:21.960 align:middle line:84% So in other words, are you better at noticing stuff 00:21:21.960 --> 00:21:23.185 align:middle line:90% about that thing? 00:21:23.185 --> 00:21:24.560 align:middle line:84% And so a way that this was looked 00:21:24.560 --> 00:21:26.520 align:middle line:84% at in this particular paradigm was 00:21:26.520 --> 00:21:32.400 align:middle line:84% by adding a little bit of vibrato to one of the voices. 00:21:32.400 --> 00:21:36.240 align:middle line:84% So the vibrato could either appear on the cued voice 00:21:36.240 --> 00:21:38.040 align:middle line:90% or on the uncued voice. 00:21:38.040 --> 00:21:41.300 align:middle line:84% So you have to perform this tracking task, 00:21:41.300 --> 00:21:44.025 align:middle line:84% but then in addition, you have to say, yes, there was vibrato 00:21:44.025 --> 00:21:44.900 align:middle line:90% or no, there was not. 00:21:44.900 --> 00:21:45.760 align:middle line:84% And so half the time, there would 00:21:45.760 --> 00:21:47.300 align:middle line:84% be vibrato on one of the two voices, 00:21:47.300 --> 00:21:48.800 align:middle line:84% and half the time there wouldn't be. 00:21:48.800 --> 00:21:52.560 align:middle line:84% And so what this graph is showing-- 00:21:52.560 --> 00:21:55.780 align:middle line:84% so this is how good people are at detecting vibrato. 00:21:55.780 --> 00:22:00.080 align:middle line:84% So this is d prime-- remember, sensitivity, right? 00:22:00.080 --> 00:22:03.400 align:middle line:84% And when the vibrato appears on the cued voice, 00:22:03.400 --> 00:22:06.563 align:middle line:84% performance is better than when it appears on the uncued voice. 00:22:06.563 --> 00:22:08.480 align:middle line:84% But the other thing that's kind of interesting 00:22:08.480 --> 00:22:12.535 align:middle line:84% is that if you take the people who are performing this task 00:22:12.535 --> 00:22:14.160 align:middle line:84% and you split them up into the ones who 00:22:14.160 --> 00:22:17.120 align:middle line:84% are good at the tracking task, and not so good-- 00:22:17.120 --> 00:22:20.360 align:middle line:84% so that's the good streamers and the poor streamers. 00:22:20.360 --> 00:22:24.200 align:middle line:84% The good streamers show a pretty big advantage for detecting 00:22:24.200 --> 00:22:26.820 align:middle line:84% the vibrato in the cued voice compared to the uncued voice, 00:22:26.820 --> 00:22:29.040 align:middle line:84% whereas the poor streamers don't. 00:22:29.040 --> 00:22:32.480 align:middle line:84% So that really indicates that your ability 00:22:32.480 --> 00:22:35.160 align:middle line:84% to track this thing over time does really 00:22:35.160 --> 00:22:37.360 align:middle line:84% depend on being able to selectively attend just 00:22:37.360 --> 00:22:38.690 align:middle line:90% to that thing. 00:22:38.690 --> 00:22:41.070 align:middle line:84% And that makes you more sensitive to its features 00:22:41.070 --> 00:22:42.630 align:middle line:90% compared to the other stuff. 00:22:42.630 --> 00:22:46.558 align:middle line:84% And this is just showing that advantage 00:22:46.558 --> 00:22:48.350 align:middle line:84% for detecting the vibrato in the cued voice 00:22:48.350 --> 00:22:53.230 align:middle line:84% is present throughout the whole time the stimulus is on. 00:22:53.230 --> 00:22:55.870 align:middle line:84% So another kind of paradigm for studying this kind of stuff, 00:22:55.870 --> 00:22:58.910 align:middle line:90% in this case in hearing. 00:22:58.910 --> 00:23:02.990 align:middle line:84% So in vision, you often are attending 00:23:02.990 --> 00:23:05.930 align:middle line:84% to particular spatial locations, but not always. 00:23:05.930 --> 00:23:09.070 align:middle line:90% You can also attend to features. 00:23:09.070 --> 00:23:13.802 align:middle line:84% So what I want you to do here is attend to the blue elements. 00:23:13.802 --> 00:23:15.510 align:middle line:84% And when you attend to the blue elements, 00:23:15.510 --> 00:23:17.590 align:middle line:84% you become very aware that there's 00:23:17.590 --> 00:23:20.090 align:middle line:84% this diagonal organization to the blue elements. 00:23:20.090 --> 00:23:25.870 align:middle line:84% Now I want you to attend to the red elements. 00:23:25.870 --> 00:23:29.670 align:middle line:84% And you become aware that there's a circular organization. 00:23:29.670 --> 00:23:33.295 align:middle line:84% Now I want you to attend to horizontal. 00:23:33.295 --> 00:23:34.670 align:middle line:84% And you become aware that there's 00:23:34.670 --> 00:23:39.670 align:middle line:84% another diagonal organization in the opposite place. 00:23:39.670 --> 00:23:41.190 align:middle line:90% So attention is pretty flexible. 00:23:41.190 --> 00:23:43.990 align:middle line:84% You can attend to locations, also 00:23:43.990 --> 00:23:46.990 align:middle line:90% to certain classes of features. 00:23:46.990 --> 00:23:53.550 align:middle line:84% And one paradigm for studying this type of attention 00:23:53.550 --> 00:23:56.250 align:middle line:84% and its interaction with location is visual search. 00:23:56.250 --> 00:23:59.070 align:middle line:84% This is another kind of classic paradigm 00:23:59.070 --> 00:24:02.530 align:middle line:84% So you're all going to do these tasks. 00:24:02.530 --> 00:24:04.710 align:middle line:84% In this particular task, you have 00:24:04.710 --> 00:24:07.390 align:middle line:84% to say whether there is a blue dot. 00:24:07.390 --> 00:24:10.727 align:middle line:90% 00:24:10.727 --> 00:24:11.310 align:middle line:90% AUDIENCE: Yes. 00:24:11.310 --> 00:24:12.102 align:middle line:90% JOSH MCDERMOTT: OK. 00:24:12.102 --> 00:24:16.012 align:middle line:84% A little slow, but you got it right So the result 00:24:16.012 --> 00:24:18.470 align:middle line:84% of doing experiments like this, and this is kind of obvious 00:24:18.470 --> 00:24:19.720 align:middle line:90% when you look at this display. 00:24:19.720 --> 00:24:20.967 align:middle line:90% This is super easy. 00:24:20.967 --> 00:24:23.050 align:middle line:84% As soon as it pops up, you see there's a blue dot. 00:24:23.050 --> 00:24:26.990 align:middle line:84% So for some kinds of targets and scenes, 00:24:26.990 --> 00:24:30.190 align:middle line:90% visual search is always fast. 00:24:30.190 --> 00:24:32.330 align:middle line:84% And we say that in these situations, 00:24:32.330 --> 00:24:34.330 align:middle line:84% that the target kind of pops out of the display. 00:24:34.330 --> 00:24:35.630 align:middle line:84% Remember back when we were talking about grouping, 00:24:35.630 --> 00:24:37.570 align:middle line:84% we also talked about this phenomenon of pop-out, 00:24:37.570 --> 00:24:39.528 align:middle line:84% where you have one element that kind of differs 00:24:39.528 --> 00:24:44.370 align:middle line:84% from all the rest on some simple dimension, and it pops out. 00:24:44.370 --> 00:24:46.350 align:middle line:90% It's really obvious. 00:24:46.350 --> 00:24:48.750 align:middle line:84% So the way that you objectively measure this 00:24:48.750 --> 00:24:52.670 align:middle line:84% is by measuring how long it takes 00:24:52.670 --> 00:24:56.990 align:middle line:84% you to detect the target as a function of the number of items 00:24:56.990 --> 00:25:00.830 align:middle line:84% that are in the display, which is often called the set size. 00:25:00.830 --> 00:25:03.430 align:middle line:84% So this is how you analyze visual search experiments, 00:25:03.430 --> 00:25:06.230 align:middle line:90% reaction time versus set size. 00:25:06.230 --> 00:25:10.550 align:middle line:84% And in a situation like this, where pop-out occurs, 00:25:10.550 --> 00:25:11.770 align:middle line:90% the graph is flat. 00:25:11.770 --> 00:25:16.570 align:middle line:90% 00:25:16.570 --> 00:25:18.070 align:middle line:84% So the standard explanation of this 00:25:18.070 --> 00:25:21.110 align:middle line:84% is that for some kinds of properties, 00:25:21.110 --> 00:25:25.090 align:middle line:84% a unique value in the image will draw attention. 00:25:25.090 --> 00:25:27.203 align:middle line:84% So it serves as an exogenous cue. 00:25:27.203 --> 00:25:28.870 align:middle line:84% So the fact that that one thing was blue 00:25:28.870 --> 00:25:31.870 align:middle line:84% and everything else was yellow immediately draws your attention 00:25:31.870 --> 00:25:33.110 align:middle line:90% to that location. 00:25:33.110 --> 00:25:36.520 align:middle line:84% So other kinds of things that have the same effect 00:25:36.520 --> 00:25:38.920 align:middle line:84% are a single large item among small ones, 00:25:38.920 --> 00:25:41.145 align:middle line:84% or a single curved item among straight ones. 00:25:41.145 --> 00:25:42.520 align:middle line:84% There's lots of things like that. 00:25:42.520 --> 00:25:44.860 align:middle line:84% And the diagnostic property is that the search 00:25:44.860 --> 00:25:47.060 align:middle line:84% slope of that results graph that you 00:25:47.060 --> 00:25:50.557 align:middle line:84% make of reaction time versus set size, the search slope is zero. 00:25:50.557 --> 00:25:52.140 align:middle line:84% So that's what corresponds to pop-out. 00:25:52.140 --> 00:25:54.700 align:middle line:90% 00:25:54.700 --> 00:25:59.220 align:middle line:84% But not everything is-- not everything pops out. 00:25:59.220 --> 00:26:02.440 align:middle line:90% 00:26:02.440 --> 00:26:07.420 align:middle line:84% So is there a blue vertical line? 00:26:07.420 --> 00:26:08.480 align:middle line:90% That was slow. 00:26:08.480 --> 00:26:11.420 align:middle line:84% So that's much harder than saying 00:26:11.420 --> 00:26:14.780 align:middle line:84% whether there was a blue thing amongst yellow 00:26:14.780 --> 00:26:17.920 align:middle line:84% So this is what is called a conjunction search. 00:26:17.920 --> 00:26:20.660 align:middle line:84% So conjunction searches are often pretty slow. 00:26:20.660 --> 00:26:22.140 align:middle line:84% So conjunction searches, that just 00:26:22.140 --> 00:26:24.260 align:middle line:84% means that the target is not defined 00:26:24.260 --> 00:26:26.160 align:middle line:84% by a unique value of one feature, 00:26:26.160 --> 00:26:28.800 align:middle line:84% it's defined by a conjunction of features. 00:26:28.800 --> 00:26:33.200 align:middle line:84% So you can see that there's blue stuff and vertical stuff, 00:26:33.200 --> 00:26:38.900 align:middle line:84% and the target is just the only one that is blue and vertical. 00:26:38.900 --> 00:26:44.300 align:middle line:84% So in some kinds of situations like this, search is slow. 00:26:44.300 --> 00:26:48.020 align:middle line:84% And the diagnostic results graph that you would get in this 00:26:48.020 --> 00:26:51.580 align:middle line:84% situation-- again, it's reaction time versus set size-- 00:26:51.580 --> 00:26:53.740 align:middle line:84% is that the reaction time increases 00:26:53.740 --> 00:26:56.060 align:middle line:84% with the number of elements in the display with-- that 00:26:56.060 --> 00:26:58.180 align:middle line:90% is the set size 00:26:58.180 --> 00:27:00.460 align:middle line:84% So it's almost as though you have 00:27:00.460 --> 00:27:02.020 align:middle line:84% to look around at all of the elements 00:27:02.020 --> 00:27:04.620 align:middle line:84% until you find the one that contains 00:27:04.620 --> 00:27:07.465 align:middle line:90% this combination of properties. 00:27:07.465 --> 00:27:08.840 align:middle line:84% And so the more things there are, 00:27:08.840 --> 00:27:10.340 align:middle line:84% the more things you have to look at. 00:27:10.340 --> 00:27:15.500 align:middle line:84% And so the reaction time kind of gets higher 00:27:15.500 --> 00:27:19.180 align:middle line:90% So here are some more examples. 00:27:19.180 --> 00:27:22.420 align:middle line:84% So in these feature search examples, the three on the left, 00:27:22.420 --> 00:27:24.303 align:middle line:90% again, it's kind of immediate. 00:27:24.303 --> 00:27:26.220 align:middle line:84% In these different conjunction search examples 00:27:26.220 --> 00:27:29.540 align:middle line:84% where you have to find the red vertical, each time it takes you 00:27:29.540 --> 00:27:33.770 align:middle line:90% takes you a minute to see it. 00:27:33.770 --> 00:27:36.130 align:middle line:84% These spatial configuration searches, again, it's 00:27:36.130 --> 00:27:40.890 align:middle line:84% a sort of conjunction, a certain combination of features. 00:27:40.890 --> 00:27:44.530 align:middle line:84% Finding the T takes a little while. 00:27:44.530 --> 00:27:47.210 align:middle line:84% So again, the results of these experiments 00:27:47.210 --> 00:27:49.693 align:middle line:84% would be expressed as reaction time versus set size graphs. 00:27:49.693 --> 00:27:52.110 align:middle line:84% In the cases where there's what's called a feature search, 00:27:52.110 --> 00:27:54.730 align:middle line:84% you get pop-out and the slope is zero. 00:27:54.730 --> 00:27:56.770 align:middle line:84% In these other cases, the slope is positive. 00:27:56.770 --> 00:27:59.887 align:middle line:84% So there's two lines in these graphs, 00:27:59.887 --> 00:28:02.470 align:middle line:84% because in order to actually do this experiment-- so normally, 00:28:02.470 --> 00:28:04.210 align:middle line:84% the way the experiment is done is 00:28:04.210 --> 00:28:06.823 align:middle line:84% you say whether the target is there or not. 00:28:06.823 --> 00:28:08.490 align:middle line:84% So sometimes the target will be present, 00:28:08.490 --> 00:28:11.810 align:middle line:90% and sometimes it will be absent. 00:28:11.810 --> 00:28:14.730 align:middle line:84% And so there are two results graphs here. 00:28:14.730 --> 00:28:17.490 align:middle line:84% One for the trials where the target is present, 00:28:17.490 --> 00:28:20.390 align:middle line:84% and one for where the target is absent. 00:28:20.390 --> 00:28:23.650 align:middle line:84% And you will notice that the slopes 00:28:23.650 --> 00:28:27.490 align:middle line:84% are twice as big when the target is absent as when 00:28:27.490 --> 00:28:29.050 align:middle line:90% the target is present. 00:28:29.050 --> 00:28:32.810 align:middle line:84% Does anyone want to posit an explanation 00:28:32.810 --> 00:28:34.590 align:middle line:90% for why that would be the case? 00:28:34.590 --> 00:28:36.442 align:middle line:90% AUDIENCE: To double-check. 00:28:36.442 --> 00:28:37.650 align:middle line:90% JOSH MCDERMOTT: Say it again. 00:28:37.650 --> 00:28:38.770 align:middle line:90% AUDIENCE: Double-check. 00:28:38.770 --> 00:28:39.390 align:middle line:90% JOSH MCDERMOTT: Double-check. 00:28:39.390 --> 00:28:40.250 align:middle line:90% What do you mean by that? 00:28:40.250 --> 00:28:40.650 align:middle line:84% AUDIENCE: You have to double-check 00:28:40.650 --> 00:28:41.667 align:middle line:90% that it's not there. 00:28:41.667 --> 00:28:42.750 align:middle line:90% JOSH MCDERMOTT: All right. 00:28:42.750 --> 00:28:44.610 align:middle line:84% What's double-check mean, though? 00:28:44.610 --> 00:28:47.233 align:middle line:90% AUDIENCE: Look once, look twice. 00:28:47.233 --> 00:28:48.650 align:middle line:84% JOSH MCDERMOTT: What do you think? 00:28:48.650 --> 00:28:50.290 align:middle line:84% AUDIENCE: Well, you can short circuit and finish 00:28:50.290 --> 00:28:51.810 align:middle line:84% searching early once you find it. 00:28:51.810 --> 00:28:54.290 align:middle line:84% You only have to look through half the image, potentially. 00:28:54.290 --> 00:28:55.150 align:middle line:90% JOSH MCDERMOTT: On average. 00:28:55.150 --> 00:28:55.450 align:middle line:90% AUDIENCE: Yeah. 00:28:55.450 --> 00:28:56.110 align:middle line:90% JOSH MCDERMOTT: Yeah. 00:28:56.110 --> 00:28:56.652 align:middle line:90% That's right. 00:28:56.652 --> 00:28:59.930 align:middle line:84% So under a model where you're kind of looking around 00:28:59.930 --> 00:29:03.943 align:middle line:84% at every element in the image, then if the target is there, 00:29:03.943 --> 00:29:05.610 align:middle line:84% on average, you'll find it after looking 00:29:05.610 --> 00:29:06.735 align:middle line:90% through half of the things. 00:29:06.735 --> 00:29:08.770 align:middle line:84% Whereas if it's not there, to be sure, 00:29:08.770 --> 00:29:10.210 align:middle line:90% you got to look at everything. 00:29:10.210 --> 00:29:15.470 align:middle line:84% That's probably what you meant by double-check. 00:29:15.470 --> 00:29:20.215 align:middle line:84% So this is the phenomenon, that in some cases, 00:29:20.215 --> 00:29:22.590 align:middle line:84% the reaction time doesn't scale with the number of items, 00:29:22.590 --> 00:29:24.370 align:middle line:90% and in other cases, it does. 00:29:24.370 --> 00:29:28.130 align:middle line:84% So the standard explanation for these slow searches 00:29:28.130 --> 00:29:31.690 align:middle line:84% is that you have to combine properties in order 00:29:31.690 --> 00:29:34.170 align:middle line:84% to detect the target, and the combination process 00:29:34.170 --> 00:29:35.430 align:middle line:90% is not automatic. 00:29:35.430 --> 00:29:38.376 align:middle line:90% 00:29:38.376 --> 00:29:42.210 align:middle line:84% So it's commonly proposed that you need attention in order 00:29:42.210 --> 00:29:43.550 align:middle line:90% to do that combination. 00:29:43.550 --> 00:29:45.330 align:middle line:84% So the spotlight of attention is often 00:29:45.330 --> 00:29:50.850 align:middle line:84% proposed to weld these different elementary features together. 00:29:50.850 --> 00:29:52.850 align:middle line:84% And this makes search a serial process 00:29:52.850 --> 00:29:56.410 align:middle line:84% because the spotlight of attention under this explanation 00:29:56.410 --> 00:29:58.370 align:middle line:84% can only be in one place at a time, 00:29:58.370 --> 00:30:00.515 align:middle line:90% so you have to move that around. 00:30:00.515 --> 00:30:01.890 align:middle line:84% And every time it's on something, 00:30:01.890 --> 00:30:06.760 align:middle line:90% you get the features combined. 00:30:06.760 --> 00:30:10.450 align:middle line:84% And so if you look at what these slopes are like, 00:30:10.450 --> 00:30:13.930 align:middle line:84% you can infer that the spotlight travels at about 50 milliseconds 00:30:13.930 --> 00:30:14.670 align:middle line:90% per item. 00:30:14.670 --> 00:30:17.630 align:middle line:90% 00:30:17.630 --> 00:30:18.630 align:middle line:90% Here's another example. 00:30:18.630 --> 00:30:21.130 align:middle line:90% Find the red verticals. 00:30:21.130 --> 00:30:24.330 align:middle line:90% This one is really annoying. 00:30:24.330 --> 00:30:25.910 align:middle line:90% There's a couple of them there. 00:30:25.910 --> 00:30:26.410 align:middle line:90% Yeah. 00:30:26.410 --> 00:30:30.180 align:middle line:90% 00:30:30.180 --> 00:30:30.680 align:middle line:90% Yeah. 00:30:30.680 --> 00:30:34.060 align:middle line:90% It's a tough conjunction search. 00:30:34.060 --> 00:30:38.500 align:middle line:84% So this, on the one hand-- it's a very interesting phenomenon. 00:30:38.500 --> 00:30:42.932 align:middle line:84% So we're finding that there's this structure 00:30:42.932 --> 00:30:45.140 align:middle line:84% in the visual system, and that's kind of interesting. 00:30:45.140 --> 00:30:47.920 align:middle line:84% But it's important also because it has 00:30:47.920 --> 00:30:49.260 align:middle line:90% a lot of real-world relevance. 00:30:49.260 --> 00:30:51.960 align:middle line:84% So oftentimes the things that you're 00:30:51.960 --> 00:30:54.080 align:middle line:84% looking for in the real world-- and you 00:30:54.080 --> 00:30:56.680 align:middle line:90% do visual search all the time. 00:30:56.680 --> 00:30:57.740 align:middle line:90% Where are my keys? 00:30:57.740 --> 00:30:59.400 align:middle line:84% Like where did I leave my coffee cup? 00:30:59.400 --> 00:31:02.140 align:middle line:90% Where's my kid's left shoe? 00:31:02.140 --> 00:31:04.377 align:middle line:84% It's like visual search is a big part of life. 00:31:04.377 --> 00:31:05.960 align:middle line:84% And oftentimes what you're looking for 00:31:05.960 --> 00:31:07.500 align:middle line:90% is defined by conjunctions. 00:31:07.500 --> 00:31:12.640 align:middle line:84% So if I ask you to find the faucet here, 00:31:12.640 --> 00:31:14.380 align:middle line:84% raise your hand when you see the faucet. 00:31:14.380 --> 00:31:17.220 align:middle line:90% 00:31:17.220 --> 00:31:17.720 align:middle line:90% Yeah. 00:31:17.720 --> 00:31:21.000 align:middle line:90% See, it takes a little while. 00:31:21.000 --> 00:31:24.600 align:middle line:84% So visual search, it's an important thing, 00:31:24.600 --> 00:31:27.780 align:middle line:90% and a classic paradigm. 00:31:27.780 --> 00:31:31.900 align:middle line:84% So visual search became very popular around 1980, 00:31:31.900 --> 00:31:35.520 align:middle line:84% and the popularity in part was due to some very influential 00:31:35.520 --> 00:31:38.680 align:middle line:84% papers by Anne Treisman, who was a very important cognitive 00:31:38.680 --> 00:31:42.720 align:middle line:84% psychologist who worked at Berkeley, and then Princeton. 00:31:42.720 --> 00:31:44.600 align:middle line:90% She died a few years ago. 00:31:44.600 --> 00:31:51.440 align:middle line:84% And the general proposal was that there are these maps 00:31:51.440 --> 00:31:53.660 align:middle line:84% of different features in the brain, 00:31:53.660 --> 00:31:58.120 align:middle line:84% these elementary features, orientation, color, size, 00:31:58.120 --> 00:32:02.460 align:middle line:84% stuff like that-- spatial frequency-- 00:32:02.460 --> 00:32:04.880 align:middle line:84% that we're kind of loosely associated with the kinds 00:32:04.880 --> 00:32:07.700 align:middle line:84% of things that we think of as being extracted by early vision. 00:32:07.700 --> 00:32:10.560 align:middle line:84% So you have these different maps, 00:32:10.560 --> 00:32:13.400 align:middle line:84% and attention kind of serves to bind different features 00:32:13.400 --> 00:32:14.320 align:middle line:90% together. 00:32:14.320 --> 00:32:16.900 align:middle line:84% So without attention, all you have are these separate maps 00:32:16.900 --> 00:32:19.400 align:middle line:84% of features, and that makes it difficult to tell whether you 00:32:19.400 --> 00:32:21.080 align:middle line:90% have a conjunction of features. 00:32:21.080 --> 00:32:23.960 align:middle line:84% So the proposal was that somehow-- again, these were not 00:32:23.960 --> 00:32:25.540 align:middle line:90% very mechanistic explanations. 00:32:25.540 --> 00:32:28.520 align:middle line:90% They're very abstract. 00:32:28.520 --> 00:32:30.880 align:middle line:84% But somehow or another, attention kind of magically 00:32:30.880 --> 00:32:33.300 align:middle line:84% enables you to combine these properties. 00:32:33.300 --> 00:32:36.640 align:middle line:84% And the conjunction search and feature search data 00:32:36.640 --> 00:32:38.980 align:middle line:90% were consistent with that. 00:32:38.980 --> 00:32:43.500 align:middle line:84% So leaving aside the question of how this would actually work, 00:32:43.500 --> 00:32:47.920 align:middle line:84% this raised this question of what 00:32:47.920 --> 00:32:51.320 align:middle line:84% will happen in conditions where attention is not 00:32:51.320 --> 00:32:53.460 align:middle line:84% available to glue stuff together. 00:32:53.460 --> 00:32:56.206 align:middle line:90% 00:32:56.206 --> 00:33:00.120 align:middle line:84% And so this was studied with experiments like this. 00:33:00.120 --> 00:33:03.020 align:middle line:84% And it led to this phenomenon called illusory conjunctions. 00:33:03.020 --> 00:33:05.500 align:middle line:84% So you're looking at displays like this. 00:33:05.500 --> 00:33:07.320 align:middle line:90% They're flashed up very quickly. 00:33:07.320 --> 00:33:10.948 align:middle line:84% Your task here is to report the digits. 00:33:10.948 --> 00:33:13.240 align:middle line:84% So you're looking at the middle, but you got to report, 00:33:13.240 --> 00:33:15.260 align:middle line:90% in this case, 3 and 5. 00:33:15.260 --> 00:33:17.080 align:middle line:84% So that's your main responsibility 00:33:17.080 --> 00:33:19.900 align:middle line:84% in the context of this task, is get the digits right. 00:33:19.900 --> 00:33:22.800 align:middle line:84% So the idea is that causes people to focus their attention 00:33:22.800 --> 00:33:25.560 align:middle line:90% kind out in the periphery. 00:33:25.560 --> 00:33:29.410 align:middle line:84% But you're also supposed to report the colors and letters. 00:33:29.410 --> 00:33:32.903 align:middle line:90% 00:33:32.903 --> 00:33:35.070 align:middle line:84% And so you make sure-- you set this up in such a way 00:33:35.070 --> 00:33:38.010 align:middle line:84% that people have had a cup of coffee, they're motivated. 00:33:38.010 --> 00:33:40.363 align:middle line:84% And you can confirm that they're doing what they're 00:33:40.363 --> 00:33:42.030 align:middle line:84% supposed to do by verifying that they're 00:33:42.030 --> 00:33:43.490 align:middle line:90% good at reporting the digits. 00:33:43.490 --> 00:33:45.830 align:middle line:84% And if you do this, then you find 00:33:45.830 --> 00:33:49.150 align:middle line:84% that the subjects will make errors at reporting the letters 00:33:49.150 --> 00:33:49.810 align:middle line:90% and colors. 00:33:49.810 --> 00:33:51.770 align:middle line:90% But the errors are not random. 00:33:51.770 --> 00:33:54.510 align:middle line:84% Most of the errors, or more of the errors, 00:33:54.510 --> 00:33:56.570 align:middle line:84% are what are called conjunction errors, 00:33:56.570 --> 00:33:58.590 align:middle line:84% where you get the wrong combination 00:33:58.590 --> 00:34:00.188 align:middle line:90% of the color and the letter. 00:34:00.188 --> 00:34:02.230 align:middle line:84% So people generally-- they're not going to-- they 00:34:02.230 --> 00:34:03.690 align:middle line:84% don't report a color that's not there. 00:34:03.690 --> 00:34:05.810 align:middle line:84% They tend to not report a letter that's not there. 00:34:05.810 --> 00:34:08.670 align:middle line:84% But they'll get the letters and color combinations incorrect. 00:34:08.670 --> 00:34:14.670 align:middle line:84% So red letter T and a blue X. So this phenomenon 00:34:14.670 --> 00:34:19.110 align:middle line:84% of illusory conjunctions was discovered by Anne Treisman, 00:34:19.110 --> 00:34:23.130 align:middle line:84% and was taken as support for feature integration theory. 00:34:23.130 --> 00:34:26.050 align:middle line:90% 00:34:26.050 --> 00:34:28.389 align:middle line:84% Any questions about illusory conjunctions? 00:34:28.389 --> 00:34:33.989 align:middle line:90% 00:34:33.989 --> 00:34:39.590 align:middle line:84% So this was very, very influential in the 1980s. 00:34:39.590 --> 00:34:41.690 align:middle line:84% Nowadays, I would say the star has fallen. 00:34:41.690 --> 00:34:45.590 align:middle line:84% I mean, most people don't really believe very strongly 00:34:45.590 --> 00:34:47.790 align:middle line:90% in feature integration theory. 00:34:47.790 --> 00:34:50.215 align:middle line:84% One reason for this is that over the years, 00:34:50.215 --> 00:34:52.590 align:middle line:84% through studying lots of different types of visual search 00:34:52.590 --> 00:34:54.989 align:middle line:84% tasks, lots of complexities emerged 00:34:54.989 --> 00:34:58.070 align:middle line:84% that seemed challenging to account for. 00:34:58.070 --> 00:35:00.470 align:middle line:84% So one such complexity is that you 00:35:00.470 --> 00:35:03.590 align:middle line:84% can get pop-out for things that probably 00:35:03.590 --> 00:35:05.350 align:middle line:90% aren't represented in maps. 00:35:05.350 --> 00:35:07.510 align:middle line:90% So 3D shape is a good example. 00:35:07.510 --> 00:35:09.750 align:middle line:84% So if you look at this, the one on top, 00:35:09.750 --> 00:35:11.550 align:middle line:84% it immediately pops out to you that there's 00:35:11.550 --> 00:35:13.508 align:middle line:84% one that's kind of different from all the rest. 00:35:13.508 --> 00:35:18.910 align:middle line:84% And the thing at the bottom is an alteration 00:35:18.910 --> 00:35:20.350 align:middle line:84% of that display that kind of kills 00:35:20.350 --> 00:35:22.630 align:middle line:84% the three-dimensional interpretation. 00:35:22.630 --> 00:35:26.000 align:middle line:84% And there, it's much harder to actually detect the target. 00:35:26.000 --> 00:35:30.423 align:middle line:84% So there's lots of examples of pop-out occurring 00:35:30.423 --> 00:35:32.840 align:middle line:84% for these kind of complicated, slightly high-level things, 00:35:32.840 --> 00:35:38.340 align:middle line:84% and that seemed inconsistent with it. 00:35:38.340 --> 00:35:41.180 align:middle line:84% But it remains the fact-- so it remains the case that-- 00:35:41.180 --> 00:35:43.460 align:middle line:84% I mean, illusory conjunctions are a phenomenon 00:35:43.460 --> 00:35:44.960 align:middle line:90% that demands an explanation. 00:35:44.960 --> 00:35:49.940 align:middle line:84% And I think, in general, it remains the case that attention 00:35:49.940 --> 00:35:52.740 align:middle line:84% seems to change the way that collections 00:35:52.740 --> 00:35:55.638 align:middle line:84% of what we might consider to be features are represented. 00:35:55.638 --> 00:35:57.180 align:middle line:84% And I would say we still don't really 00:35:57.180 --> 00:36:01.700 align:middle line:84% have great quantitative or rigorous explanations of why 00:36:01.700 --> 00:36:07.200 align:middle line:84% that is, but it's a real phenomenon. 00:36:07.200 --> 00:36:08.940 align:middle line:90% I'm going to end there. 00:36:08.940 --> 00:36:12.820 align:middle line:84% When we resume on Tuesday, we will finish up 00:36:12.820 --> 00:36:15.350 align:middle line:90% talking about attention. 00:36:15.350 --> 00:36:25.000 align:middle line:90%