20: Attention

Key Takeaways

So the next topic on our agenda um is attention. Okay. Um and so attention is a it's a big word that can mean a lot of different things. um colloquially you're familiar with the notion of paying attention to something um versus maybe ignoring it. Um and we're talking about attention in a class on perception because um attention often determines what you see and hear and in general perceive about the world. Okay? So in vision what you're attending to often is a function of what you look at. So attention oftenimes moves with your eyes, right? Um, but it doesn't have to. So, like if you're being really sneaky, you could be looking at me, but maybe pulling paying attention to something, you know, that that's out here, right? So, you can pay attention to things if if you're not fixating them. Um, so I guess, you know, naively and intuitively, when you pay attention to something, that seems to improve in some way the processing of whatever is attended, right? That's sort of like presumably why we why we do it, why we use attention, right? like when you pay attention to something like you see it better or you hear it better in some in some sense and we're of course not specifying what it means for it to be better, right? So that's a that's an important issue. Um again, intuitively, you know, people often think of this maybe as a way to focus resources. Again, it's in scare quotes because we don't really know what the resources are, but it's perhaps if resources do exist and you need them for perception and they're limited, this could be a way to focus them on whatever seems important. So in this lecture we're going to we're going to talk about how we can measure and study uh these phenomena. Okay. And so um attention has been studied with a a dizzying array of paradigms. Okay. So this lecture will provide an overview of of some of them. Some of the best known that are very relevant to perception. Um but it it's sort of going to be a little bit of a grabag. Right. So right now we we lack I would say a unifying framework for thinking about attention and and it's almost surely not one thing. So there's lots of different things that often get called attention and they're not necessarily u necessarily the same thing. Um and at the moment uh we we mostly I would say lack rigorous computational understandings. That's actually our super exciting direction um for the future. Um and I think we're at a point now where we could actually really start to develop sort of rigorous computational theories of attention. Um but there nonetheless many interesting phenomena um that place constraints on future theoretical understanding. Okay. So, we'll we're going to go through some of them and talk about uh what what they what they show us and what they mean. Okay. So, here's an an initial example. Um so, very popular paradigm for studying attention involves queuing people. Okay. So, Pner was a cognitive psychologist who did a lot of work on queuing. So, the essential idea um is to focus visual attention to an area of space using what's called a Q. Okay. Okay, so that's like a stimulus that will occur at some part of space or that will indicate some part of space where something's going to happen. Um, and then in the classical paradigms for studying this, um, they would measure the time that it would take someone to identify a target stimulus, um, when either the observer does not know where the item will appear, okay, versus where they do know where the item will appear by virtue of the Q, okay? And so um in this initial example that I'll show the Q might just be a briefly presented dot at the location of the target. Okay. Um so look in the middle here at the center of the cross. Okay. And so your task is going to be to say what letter is going to appear. Okay. So you can all do this task. So do it. You're really slow. What's the letter? >> Hey. Wow. That was like the longest reaction time in the history of attention experiments. Okay. Okay. You got it right though. It is an A. Okay. All right. So this is the uncueded condition. Okay. And it is supposed to be slower than the ceued condition. Here's the ceued condition. >> And that was much faster. See? So that's the queueing benefit. Okay. Okay. All right. So this is the a schematic of the kind of results. So um what is measured here is reaction time. So again you assume that most of the time and and you val you verify that most of the time people get the letter correct. Right? So you measure how long it takes them to identify the letter. Um and there's a reduction in reaction time when you're cued to the location. Okay. Um and so in this sense um advanced knowledge of the location improves performance as measured by reaction time. Okay. Okay. So the other um thing that that is associated with this phenomena um is that there is uh it's spatially localized. Okay. So this is again kind of a fake graph sort of schematizing the results of lots and lots of experiments. Um and the graph is plotting the reduction in reaction time that you get from the Q as a function um of the ceued location um relative to the target. Okay. And so the point is that when the Q is exactly at the location of the target, that's where you get the biggest benefit. Okay? And then if the target is is off of the ceued location by some amount, um there's a reduction in the benefit. Okay? All right. So this type of finding um led to a very influential and popular metaphor of attention as a spotlight. Right? So the idea is that there's this this thing that's spatially localized. It gets cued to a location. It gets moved to that location. Does something that is as of now unspecified to the processing that makes you faster or more accurate or whatever. Okay. So the spotlight metaphor of of attention. So the idea is that whatever's in the spotlight is attended. um the more it's attended, the better it's processed. Um and the size and the shape of the spotlight can be controlled to some extent. Okay. So now another really important distinction that kind of came out of this sort of study is the idea that there can be two types of cues. Okay. So the one that we just saw is what would be normally referred to as exogenous. So that means outside generating. Okay. So the dot flashes up without really thinking about it. Like your attention is kind of drawn to the location of this thing that pops up on the screen. Okay. Um it's almost like a reflex. Okay. Um so typically exogenous cues would be sudden changes. So flashes or movement. Um they draw attention automatically. Okay. So the the kind by comparison like the other kind of cue that that you can get in general is what's called called endogenous. So that stands for inside generating. Okay. Um, so this typically we think requires some kind of high level control. Um, and typically involves an instruction. So some kind of visual sign or pattern that will be a symbol of where the target is going to um appear. Okay, so here's an example of exogenous queuing or so uh sorry this is this is the exogenous one that we just saw. Okay, just to just to repeat it. So there's the dot and then the letter. Okay, so that you don't have to interpret anything. Your attention just does its thing. Here's an example of endogenous queuing. Okay, so we got an arrow and then the letter appears. Okay, so the point is that in order for you to benefit from the queue, you have to understand what arrows mean, right? Um and willfully move your attention over in that direction and then you get a benefit. Okay. All right. So both types of cues um seem to control an attentional mechanism. Maybe not the same one, okay? But they reflect different strategies. So we think of exogenous cues as tapping into some kind of bottom up control of attention that's based on physical transients in the environment and endogenous cues as involving what's often called top-down control of attention based on what an observer believes. Okay. All right. So how do these things actually cash out in in terms of the results? Well, one of the kind of clear differences that you see um is what's shown here. Um so um this is a graph that is um plotting the same thing that we looked at before. So the y- axis is the benefit in reaction time from a valid Q. So it's like how much faster you are if you're queued to the location. Okay. But now what's being manipulated here is what's called the stimulus onset asynchrony. Okay. And so that's indicated here. Okay. So here's what what would happen on a trial of an experiment where you're measuring this stuff. This is a case where the Q is exogenous. So you're fixating that little star. Okay? And then the Q pops up. In this case, it's this red box, right? So something changes. Your attention kind of moves over to the left. Um and then after some amount of time, the target appears. And that some amount of time is the S SOA, the stimulus onset asynchrony. Okay? So essentially you can vary the amount of time from when the Q pops up until when that thing that you have to detect pops up. Okay. All right. And so what what what does this graph show? Well, it shows that um initially so if this if the stimulus on synchron is really really short then you don't get the benefit, right? So it's like the queue does something that takes some amount of time to kind of happen. Okay. All right. And then you know in in the case of um what's called the peripheral queue here, that's the exogenous one, the red box. Okay. by the time you you know it's there's about a 100 milliseconds you you know you're close to getting the maximal benefit from the que okay but when you have a symbolic Q what we're calling endogenous right where you have to kind of interpret it that's like the arrow okay it takes you longer to get the benefit okay because it's like something else has to happen in your head you see this symbol and there's some process of interpretation okay and that takes a bit more time okay so this is some evidence that these two types of cues are kind of doing different things in your head. Okay, any questions about that? Okay. And both of these types of things happen all the time in like everyday life, right? So there's very frequently like transient things happen, you know, attention's drawn to some location um because that's how we talk about it. Um, other times there'll be something that's more symbolic, you know, that indicates to you, you know, so for instance, like I don't know, you see a turn signal, right? I mean, indicates the thing's going to turn to the the left and so you're paying attention over there. Yeah. Is there like this binary difference between the exogenous and the other one? Um or like could you imagine like an arrow that's pointing upward but like the the visual stimulus is like immediately there as well so it's like orienting you like is is there some sort of I guess gray area? >> Yeah. Yeah, I mean I think that that often in a lot of real world scenarios, you're getting combinations of these two things, right? Where there'll be some transient, but you know, you you also just because of your understanding of the situation, you kind of know that something's going to happen in a particular area. You know, um I don't know, let's say you're watching baseball, right? And so the pitcher's throwing the ball, so you know the you know you know that it's going to be caught in some region, you know, that around the strike zone, right? But then like you know the catcher puts up the mid and you hear you know the sound of the ball slapping in right and then you get this this more exogenous cue for where it's happening right so that's sort of like a combination of things right so I think that kind of stuff happens a lot yeah I don't think these are necessarily mutually exclusive okay so that's like a very common distinction in the world of attention um e exogenous and endogenous um attention all right so the these kinds of queuing studies um they gave rise to this idea that you could kind of think of attention as a spotlight. Okay. And so that was sort of very persistent for several decades. Um and then um people got very interested in in whether or not um you could actually attend to multiple things at the same time. Okay. And so this the paradigm that is shown here became wildly popular in the world of visual attention. Um this is what's often called multiple object tracking. Okay. And so what happens in this paradigm is there's a bunch of objects that are on a screen. Okay. And you are cued to attend to um a subset of them. So in this case um the red ones are the ones that you're supposed to attend to. Okay? So you get this initial queue. Okay? Then the queue disappears. All right? And at this point all of the objects are basically identical, right? So the only thing that distinguishes them was the is the fact that you know that you're supposed to be paying attention to a subset of them. Okay? Then they start moving around. Okay? So they move around for a while, okay? And then at the end, one of them is going to turn red again. And you have to say, um, is the one that turned red, is that part of the initial set of red balls or or is it not? Okay. So it's a measure of the extent to which you can kind of keep track of all these different objects. Okay. All right. Let's see if you can do it. Um, so you need to fixate that little square at the center. Okay. Um, the flashing ones are the ones you got to pay attention to. And now they're going to move around. Okay. So, did you feel like you could pay attention to a bunch of them? Yeah. You also probably noticed that gets pretty hard when two things kind of overlap, right? So, one of the things that you're trying to track is sort of crosses paths with one of the other ones, you know, that gets kind of hard. So, there's all kinds of interesting stuff that happens here. Okay. Um, but people have some ability to do this. Um, this paradigm was introduced by Xenon um pollition uh who was a famous cognitive psychologist at Ruckers. Um this is a graph that shows the proportion of errors that are made as a function of the number of targets that um that people are tracking and you can see that people get worse as the number of targets increases. Um but the point is that the proportion of errors is still relatively low you know even out to kind of five things. Okay. Um so people have the seem to have the ability to track um multiple things at the same time. And so this um this paradigm kind of became a little cottage industry and there were you know many many many experiments like measuring how people could do this under different conditions and um and things like that. Okay. Um so the general conclusions from this are that people can attend to multiple locations at once. Um so attention is not just a spotlight. Maybe there are multiple spotlights. Um, it's still the case though that in a lot of cases, um, the spotlight metaphor works pretty well. So there's often kind of one main thing that you're you're attending to. Um, any questions about multiple object tracking? >> Yeah. >> Is it learnable? Like you train to get better. >> So can you train to get better? Um, so probably um I'm pretty sure that there's a study that I I know about which I hope is not apocryphal, but I believe um these experiments were ran on um I think it was like the Canadian Olympic basketball team, you know? So like, you know, basketball is a great example where you maybe have to do something kind of like this because there's all these different players and you sort of need to keep track of like who's where and stuff like that, right? Um and so I believe it there was a finding that um the capacity for multiple object tracking was better on people who are really good at basketball. Um um so I think there's been some things some studies like that showing that some people are better than others. Um I don't know how trainable that is. It could also just be that you know the people who are good at multiple object tracking end up being the ones that are good at basketball or something, right? So yeah. Um the other okay the other thing that is related to this is there's a again another sort of small cottage industry of studies on people who play lots of video games. Um and in general like the finding is that people who play lots of video games seem to do better on a lot of visual attention tasks. I don't know if they were ever tested on this one in particular. U but the some of the other effects that we will kind of see later in the lecture show benefits um from video game playing. Okay. Um so lots of attention research has been conducted in vision. Um but it's also it's really important for hearing as as well. Um, you remember back when we were um talking about audition, we talked a lot about the cocktail party problem where there's like one person you're trying to understand. So, you have to pay attention to their voice with amid these distractors. Um, and so, you know, one situation where attention uh may be important is where there are um concurrent sources that are similar and you kind of have to keep track of one that you're trying to understand. Okay. Okay. And so this is a graph um that is um showing you the trajectories that are taken by two voices that are people two people talking at the same time um through a space of features that we think are important for voices. Okay. So f0 is the fundamental frequency. Everybody remembers the fundamental frequency, right? It's thought to be like the correlative pitch. Okay, it's the rate of repetition. Um and then the other thing another thing that's really important about voices are the formats, right? So the first two formants like define vowels first order. So F1 and F2. Um and so when you talk um you're constantly modulating your fundamental frequency and your formance. Okay? And so you can think of a voice as kind of moving around in that three-dimensional feature space. And so these are uh example trajectories. The yellow line is for one voice and the blue line is for another voice. And so the point is just that they're all kind of tangled up. Okay. So this is like two two I think there are two female speakers that are talking at the same time. Okay. Um and so naively you might think that if you want if you want to follow what one person is saying, you might need to kind of track that voice um as it moves through the feature space. Okay. And so this is like an example task that was devised to look at this. Okay. So these are I'm going to play an example of the stimulus, but these are synthetic voices that are continuously voiced vowels. Okay. And so they they move around in this space and there's going to be two of them. And this is a task that was devised to sort of measure whether or not you can track one of the two voices. Okay? And so the way that this works is you initially get a Q and that tells you which of the two voices you're supposed to listen to. Okay? And that Q is the very initial part of the target voice. Okay? So like the green one in this case. Okay? So then you get a mixture of two voices. And so you can see this is like two harmonic things on top of each other. It's just spectrograms obviously. Um and then after the mixture you get what's called a probe. And so the probe is the end portion of one of the two voices. Okay? So half the time it will be the end portion from the green voice, the target voice. Half the time it will be the end portion from the other voice that you're supposed to ignore. Okay? And so you just have to say yes, this the probe is from the cued voice or no it isn't. Okay? And the idea is that like because these things kind of are circling around each other in the same part of the space, in order to to to perform this task, you have to be able to track this thing over time to essentially be able to connect the the cued portion of the voice to the end portion of the voice. Okay. Um, so you can try it for yourself. You're initially going to hear the Q and then you're supposed to track the cued voice. Okay. And then listen to how it ends and then you get a probe and you have to say whether the probe was the one at the end or not. Was was the end of the target voice rather? >> Yes or no? >> Yeah. Okay. Good. Some of you not sure. Here's another one. >> Yeah. Okay. So, both in both those cases it was um so interestingly um it turns out that that this task is much easier for musicians. Okay. Um and uh I think many of you who were smiling are probably musically trained. Um but people can do this. Um and um one so one question is like well it seems like in order to do this task you would have to track this thing over time with your attention. And um one way to potentially measure this and a conventional way to sort of measure whether you're attending to something is to to see whether there is um some other kind of attentional benefit to attending to that thing. Right? So in other words, are you better at kind of noticing stuff about that thing. And so way that that this was looked at in this particular paradigm was by adding a little bit of vo um to one of the voices. Right? So the VAR could either appear on the cued voice or on the uncued voice. Okay. So you perfor you have to perform this this tracking task but then in addition you have to say yes there was VBR or no there was not. And so half the time there would be VBR on one of the two voices and half the time there wouldn't be. Okay. Um and so what what this graph is showing this is how um how good people are at detecting the VA. So this is D prime remember sensitivity right? Um and when the VA appear appears on the cued voice um performance is better than when it appears on the uncued voice. Okay. Um but the other thing that's kind of interesting is that if you if you take um the people who are performing this task and you split them up into the ones who are good at the tracking task and not so good. So that's the good streamers and the poor streamers. The good streamers kind of show a pretty big advantage for detecting the VBR in the cued voice compared to the uncued voice whereas the poor streamers don't. Okay, so that really sort of indicates that your ability to kind of track this thing over time does really depend on being able to selectively attend just to that thing, right? And that makes you more sensitive to kind of its features compared to like the other stuff. Okay. Um, and this is just showing that that advantage for detecting the VAR in the cued voice is sort of present throughout the the whole um time the stimulus is on. Okay. So another kind of paradigm for kind of studying this kind of stuff in this case in hearing. All right. So in vision you often are attending to particular spatial locations um but not always is you can also off also attend to to features. Okay. So what I want you to do here is attend to the blue elements. Okay. And when you attend to the blue elements you become very aware that there's kind of this diagonal organization to the blue elements. Okay. Now I want you to attend to the red elements. and you become aware that there's sort of a circular organization. Now I want you to attend to horizontal and you become aware that there's another diagonal organization in the opposite place. Right? Um so attention is pretty flexible. You know you can attend to locations um also to certain classes of features. Um and one paradigm for kind of studying this type of attention and its interaction with location is v is visual search. This is another kind of classic paradigm. Okay. So um you're all going to do this these tasks. Okay. In this particular task you have to say whether there is a blue dot. Okay. >> Yes. >> Okay. A little slow but you got it right. Okay. All right. Um, so the result of doing experiments like this, and this is kind of obvious when you look at this display, right? Like this is super easy, right? As soon as it pops up, you see there's a blue dot. Okay. So for some kinds of targets and scenes, visual search is always fast. All right. Um, and we say that in these situations that the target kind of pops out of the display. Remember back when we were talking about grouping, we also talked about this phenomenon of pop out where you have one element that kind of differs from all the rest on some simple dimension. Um, and it it pops out. It's really obvious. So the way that you objectively measure this is by measuring um how long it takes you to detect the target as a function of the number of items that are in the display which is often called the set size. Okay, so this is how you analyze visual search experiments, right? Reaction time versus set size. And in a situation like this where popout occurs, the graph is flat. Okay. All right. So the standard explanation of this is that for some kinds of properties a unique value in the image will draw attention. So it serves as an exogenous cube. All right. So the fact that that one thing was blue and everything else was yellow immediately draws your attention to that location. Okay. So other kinds of things that do the same have the same effect are a single large item among small ones or a single curved item among straight ones. There's lots of things like that. And the diagnostic property is that the search slope of that results graph that you make of reaction time versus set size, the search slope is zero. So that's what corresponds to pop out. Okay. But um not everything is not everything pops out. Okay. Um so is there a blue vertical line? That was slow. Okay. So that's much harder than saying whether there was a blue thing amongst yellow. Okay. So this is what is called a conjunction search. Okay. So conjunction searches are often pretty slow. So conjunction searches that just means that the target is not defined by a unique value of one feature. It's defined by a conjunction of features. Okay? So you can see that there's um there's blue stuff and vertical stuff, right? And the target is just the only one that is blue and vertical. Okay. So in in some kinds of of situations like this search is slow. Um so and and the the diagnostic results graph that you would get in this situation again it's reaction time versus set size is that the reaction time increases with the number of elements in the display with that is the set size. Okay. So it's almost as though you kind of have to look around at all of the elements until you find the one um that contains this combination of properties. Okay. And so the more things there are, the more things you have to look at. And so the reaction time kind of gets higher. Okay. Okay. So here are some some more examples. Um so in these feature search examples, the the three on the left, again, it's kind of immediate in these different conjunction search examples where you have to find the red vertical each time. Takes you takes you a minute to kind of to see it. um these spatial configuration searches again it's kind it's a sort of conjunction a certain combination of of features um finding the takes a little while okay so again the the results of these experiments would be expressed as reaction time versus set size graphs in the cases where there's what's called a feature search you get pop out and the slope is zero in these other cases the slope is positive so there's two lines in these graphs okay because in order to actually do do this experiment, right? So, normally the way the experiment is done is you say whether the target is there or not. Okay? So, sometimes the target will be present and sometimes it will be absent. Okay? Um and so there are two results graphs here. One for the trials where the the target is present and one for the one for where the the target is absent. And you will notice um that the slopes are twice as big when the target is absent as when the target is present. Does anyone want to posit an explanation for why that would be the case? >> Double check. >> Say it again. >> Double check. >> Double check. What do you mean by that? >> Double check that it's not there. >> All right. What's double check mean, though? >> Once or twice. >> Um, what what do you think? >> Like you can like short circuit and finish searching early once you find it. Like you only have to look through half the image potentially >> on average. >> Yeah. >> Yeah, that's right. So like under a model where you're kind of looking around at every element in the image, okay, then if the target is there on average, you'll find it after looking through half of the things, right? Whereas if it's not there to be sure, you got to look at everything, right? That's probably what you meant by double check, right? Okay. All right. So this is the phenomena, right? That in some cases um the reaction time doesn't scale with the number of items and in other cases it does. So the standard explanation for these slow searches is that you have to combine properties in order to detect the target and the combination process is not automatic. Okay. So it's commonly proposed that you need attention in order to do that combination. Right? So the spotlight of attention is often proposed to kind of weld these different elementary features together. Um, and this makes search a serial process because the spotlight of attention under the under this explanation can only be in one place at a time. Okay? So, you kind of have to move that around. Um, and every time it's on something, you get you get um the features kind of combined. Okay? Um, and so if you kind of look at like, you know, how what these slopes are like, you can infer that the spotlight travels at about 50 milliseconds per item. All right, here's another example. will find the red verticals. This one is really annoying. There a couple of them there, right? Yeah, right. Yeah, it's a tough conjunction search. And okay, so this on the one hand is sort of it's a a very interesting phenomena, right? So we're like we're finding that there's all there's this structure in the visual system, right? And that's kind of interesting, right? But it's also it's important also because it has a lot of real world relevance, right? So often times the things that you're looking for in the real world and you do visual search all the time, right? Where you know where are my keys? Like where do I leave my coffee cup? You know where's my kid's left shoe, right? It's like visual search is a big part of life. And oftentimes what you're looking for is defined by conjunctions, right? So um if I ask you to find the faucet here, raise your hand when you see the faucet. Yeah, see it takes a little while. Okay. Okay. So, visual search is it's an important thing and and um and a classic paradigm. So, the so visual search became very popular um around 1980 and um the popularity in part was due to some very influential um papers by Anne Trezman um who was a very important cognitive psychologist uh who's worked at Berkeley and then Princeton. um she died a few years ago. Um and the the general proposal um was that there are these maps of different features in the in the brain, right? These elementary features, orientation, color, size, stuff like that, right? Spatial frequency, okay? Um you know, that were kind of loosely associated with the kinds of things that we think of as being extracted by early vision. Okay? So, you have these different maps. um and attention kind of serves to bind different features together. Okay, so without attention all you have are these like separate maps of features and that makes it difficult to tell whether you have a conjunction to features. Okay, so the proposal was that somehow again these are not very mechanistic explanations, right? They're sort of um very abstract but somehow or another attention kind of magically enables you to combine these properties. Okay. All right. And the the conjunction search and and feature search data were sort of consistent with that. All right. So leaving aside the question of like what how this would actually work, right? Um this kind of raises the raised this question of what will happen in conditions where attention is not available to glue stuff together. Okay. Um and so this was studied um with experiments like this. Okay. And it led to this phenomenon called elusory conjunctions. Okay. So you you're looking at displays like this. They're flashed up very quickly. Okay. Um your task here is to report the digits. So you're looking at the middle, but you got to report in this case three and five. Okay? So that's your main responsibility in the context of this task is get the digits right. Okay? So the idea is that that causes people to focus their attention kind of out in the periphery. All right? Um but you're also supposed to report the colors and letters. Okay. All right. And so you make sure you set this up in such a way that like you know people have had a cup of coffee, they're motivated, right? Um and you can confirm that they're doing what they're supposed to do by verifying that they're good at reporting the digits. Okay? Um and if you do this then you find that the the subjects will make errors at reporting the letters and colors. But the errors are not random. Okay? Most of the errors or more of the errors are what are called conjunction errors where you get the wrong combination of the color and the letter. All right? So people generally they're not going to they don't report a color that's not there. They tend to not report a letter that's not there, but they'll get the letters and color combinations incorrect. Okay. So red letter T and a blue X. All right. Um so this phenomenon of illusory conjunctions uh was discovered by in treesman and was taken as support for um feature integration theory. Okay. All right. Any questions about illusory conjunctions? Okay. So this was very very influential um in the in the 1980s. Um nowadays I would say the star has fallen. I mean you know people most people don't really believe very strongly in in feature integration theory. One reason for this is that over the years through studying lots of different types of visual search tasks lots of complexities emerged that seemed sort of challenging to account for. Um so one such complexity is that you can get pop out for things that that probably aren't represented in maps. Um so 3D shape is a good example. So if you look at this the one on top like it immediately pops out to you that there's one that's kind of different from all the rest. Okay. Um and the thing at the bottom is sort of an alteration of that display that kind of kills the three-dimensional interpretation. Um and there it's much harder to actually detect the target. Okay, so there's lots of examples of like pop out occurring for these like kind of complicated slightly high level things, right? Um and that sort of seemed inconsistent with it. Um but it remains the fact so the it remains the case that um I mean illusory conjunctions are a phenomenon that demands an explanation and um I think in general it it remains the case that attention seems to change the way that collections of what we might consider to be features are represented. Right? So and I would say we still don't really have great quantitative or rigorous explanations of why that is but um it's a real phenomenon. Okay. All right. I'm going to end there. Uh when we resume on Tuesday, we will finish up talking about attention.