Take a look at the short movie I've linked below (Click on the picture to play. QuickTime required). The movie shows a virtual gripping device (two red balls) lifting rectangular objects and placing them on a conveyor belt. Do you notice anything unusual happening as the objects are being moved?
This is a recording of the image seen by a volunteer using a virtual-reality headset and two force-feedback devices to simulate moving objects by picking them up with a thumb and index finger. The force-feedback devices offer resistance so that the volunteer feels as if he is lifting real objects:
So, did you spot the anomaly in the movie? I'll spare you the suspense and tell you that the fourth block actually changed shape during the course of the motion. What's really interesting about this is that volunteer viewer never noticed -- even though he was "carrying" it at the time. You might have noticed it, though, especially since I hinted to you that a change was going to occur. Incredibly, nearly all the time, the volunteers in this experiment did not notice a change.
This is an amazing example of a phenomenon we've discussed several times before on Cognitive Daily: change blindness. In the traditional change blindness paradigm, which is amazing enough, an object must be briefly removed from view for the change not to be noticed. As this demonstration shows, when the change occurs right before our eyes, it's much easier to spot. So why were the volunteers in the virtual reality experiment so bad at spotting the changes? The research team that conducted the experiment, led by Jochen Triesch, has some ideas. But first, let's take a look at what they actually did in their study.
Fifty-nine college students agreed to don virtual reality helmets and attach their hands to the force-feedback device while they did one of three tasks:
- "Pick up the bricks in front to back order and place them on the closer conveyor belt."
- "Pick up the tall bricks first and put them on the closer conveyor belt. Then, pick up the small bricks and also put them on the closer conveyor belt."
- "Pick up the tall bricks first and put them on the closer conveyor belt. Then, pick up the small bricks and put them on the distant conveyor belt."
In each task, ten percent of the time one of the objects they are asked to move changes shape while it's being carried to the conveyer belt, as you saw in the movie above.
Notice that in the first task, the size of the object doesn't matter at all. In task 2, size only matters when the objects are being picked up. In task 3, size matters both when the objects are being picked up and when they are put on the belts. The students weren't told that the objects could change, but they were told that the computer system was just upgraded, so they should let the experimenter know right away if they spotted anything unusual during the experiment. Each student moved 100 blocks as specified by their task group. Did they notice the objects changing shape? Here are the results:
The vast majority of students in Task 1 and most students in Task 2 didn't ever stop the experimenter to say a shape had changed. Even in Task 3, when participants had to pay attention to the shape both when picking up the objects and when setting them down (and the object was now a different shape), fewer than half of the shape changes were noticed.
Even after they finally noticed a change, participants still often didn't notice when other objects changed, as this graph illustrates:
So what's going on -- why do viewers miss so many of the changes? One possibility is that they were simply looking away from the objects at the time the shape change occurred. But Triesch's team actually monitored the eye movements of the students as they performed the task; they could use this information to determine where they were looking when the shape changed. In Task 1 and 2, it turns out, even when participants were looking directly at the objects, they still failed to notice the change more than fifty percent of the time! In Task 3, accuracy improved across the board, but many shape changes were still missed when viewers looked right at the objects!
The authors argue that the visual system computes only just enough information to complete the task at hand at any given moment. Everything else is automatically ignored. Of course, we're generally not conscious that this is what our visual systems are doing, which is why the phenomenon of change blindness is so surprising.
Triesch, J., Ballard, D.H., Hayhoe, M.M., & Sullivan, B.T. (2003). What you see is what you need. Journal of Vision, 3(1), 86-94. DOI: 10.1167/3.1.9
The change in shape is actually REALLY subtle.. so much that the subjects may be considering it an artifact of the degraded visuals in the VR environment. Or a result of differing perspectives while object is moved in the virtual envment.
It'd be interesting to see if the change would be detected with less subtle changes of shape.. or if the subject was given a much larger number of only small blocks with one of them changing to a tall block, then would a change from a small to tall block be detected?
Part of the reason the change looks pretty subtle is that this is a low-res version of the movie. The object actually transforms from one class to another, as defined by the experimental paradigm itself. It's not a meaningless transformation.
Even after being told what to look for I could see no change in the object in question. My estimate is that any conclusions based on this work are highly suspect in my book. You say that the transformation was not meaningless, but you had to resort to a virtual world to get that result. Could the novelty of working in that environment distract the brain in certain situations? I think so. Try this again with real blocks that change shape when touched or moved.
I would echo Anuj's comment on the subtle change. Even in the higher def picture you have on this page the block that is being 'picked up' could fit either small or large sizes depending on where the block is in the virtual space.
I'm more interested in the setup of the experiment. All of the mention of the changes are visual... Was there no tactile differentation between the big and small blocks (one heavier than the other)??? If so, did the 'weight' of the block change as it changed size? If not, then isn't there the possibility that the tactile sensation of the block staying the same size grossly outweigh the visual information that the block is getting larger as far as the brain is concerned?
In the book "Stumbling on Happiness" some similar results are mentioned.
The study involved CaMeL cAsEd text. The eyes of the participants were monitored and when the eyes move away the cAmEl CaSiNg is reversed.
Apparently, most participants did not notice this change, which happens as they are reading the text ;-)
The object got taller but it didn't really change shape did it? It was still a rectangle but just taller. There was an artifact in shape #3 where it actually kind of splits in half for a moment. I thought that was the change you were asking for. So if that was not the significant yet unnoticed change, and the subtle height change in #4 was, then how can the subject tell it's not just another animation artifact if the video is full of unimportant artifacts? I should also point out that object #4 got taller as it was moved closer to the subject. It can easily be mistaken for a perspective shift since these graphics are pretty basic and almost crude.
Hello, my first time comment here :)
I suspect that this ignorance phenomenon is specific to the change in *size*. That means, if there was a change in shape (rectangular block into cylindar), then the user should notice it. Why? I guess... this operation is a *translation* task, so the dimensional information of the block is hard to compute, thus considered unreliable and ignored by the brain. If it was another type of task (e.g. rotation) the change in size may be easily noticed.
Another reason maybe the low quality of real-time video (you can see the fluctuations!) and the unreal sense of virtual world, so the brain didn't take it seriously and has higher tolerance on unsual things.
Actually, I noticed the change, but percieved it as "glitchy" beviour of the video, based on my experience with glitchiness on the internet. Even with that I still would've ignored it to focus on the task getting done, as it wasn't enough to interupt judgement of manipulation.
I'd just like to point out once again that the video example you see is a *much* lower-res version of what was in the actual experiment. The video is 200 × 150 -- in the actual study, it was 640 × 480. That's over three times as large in each dimension, and more than 10 times as many pixels. Even the image you click on to view the video is a smaller image, just 400 × 300. For comparison, standard cable TV is about 440 pixels wide.
Not only that, but the virtual reality display used in the study is stereoscopic, meaning a separate 640 × 480 image is displayed in each eye, each offering a different perspective on the image and giving it a three-dimensional quality.
Furthermore, the sample movie was degraded even more in order to be compressed for Internet viewing, much like YouTube videos aren't usually as clear as a TV picture, even though they are higher-resolution. There would have been no such degradation in the actual images participants saw.
Finally, the change in the objects was precisely the relevant change for participants in the study: they transformed from short to tall or tall to short, and participants had to do different things depending on whether the objects were tall or short. This was not a subtle change in the context of the study.
Like other commenters I couldn't decide whether the blocks had changed shape even after I had read the post and was watching the blocks to see if they changed.
How about a further control? Ask some students to do the exercise and ask them which blocks change shape and how confident they are in their answer?
They were not actually directly handling a physical object. I'm sure more of them would have noticed if they were and it changed shape.
I think other commenters who mention the quality of the video environment are on to something too.
Plus, unless the participants were told that shape change was important, maybe they simply chose to ignore it rather than didn't see it. I mean, were they ever tested by being told that if the object changes shape they should discard it, as a control?
I don't think this experiment had enough controls in place ot be conclusive about this.
You bring up some good points. The researchers actually did directly ask participants after the study was over if they had noticed any of the bricks changing shapes. In general this added about 15 percent to each graph, but this still means that a very large number of changes went unnoticed.
I'm still surprised that people continue to discuss the quality of the environment. Yes, it's a possibility that the quality of the video was the deciding factor. Yet the viewers had no trouble performing the task itself, which required them to notice precisely the difference that occurred while they were moving the objects.
This criticism was leveled against other studies on change blindness, and in response the researchers actually used live actors. People still didn't notice the difference when a real person wearing different clothes was substituted for a person they were giving directions to.
Video link [WARNING: The video may temporarily freeze your browser. Give it a minute or so and it should play properly]
The effect would be a lot more striking if there were a qualitative difference in the object being moved (e.g. from red to green, from box to sphere). This work was done 5 years ago...were there any follow-ups to test the limits of this effect?
Yes, there have been follow-up studies. The most recent one I could find is this poster, which looks like it may answer some of your questions, Derek.
This is a test of following directions properly.
group 3 reported because it was the only group that would care, the size change would affect groups threes results in the task.
the other people couldnt be bothered. they were told the computer was new, it didnt crash-why get the researcher.
Its not that people didnt notice, its that they didnt care
such flawed experimental design is shocking in an estimed scientific journal such as "Journal of vision research"
1. The perspective from which the objects are viewed helps to mask the height change, so the change is partially disguised.
2. Normally, solid objects don't magically change shape when moving them. So, how do you factor in the parameters of using an improbable scenario to test true real-life "change blindness"?
3. Was there also an increase of weight as the object was being moved? If not, then the touch sense would not be triggered to cause the worker to look at what was happening to the object as it changed. A vertical weight change would shift the balance of a "real" object in motion.
4. The visual focus and mental attention was from point to point. People wouldn't expect to have to actually look at the object while it was traveling unless it was originally perceived as being capable of changing; like a wriggling worm or milk-soaked gram cracker (see #2).
5. "why do viewers miss so many of the changes?"
There weren't many changes, but many viewers missed the change. That is because the design of the experiment may be either ignoring or missing some of the variables that occur during actual "change". Maybe the results are caused more from a blindness of the experimenter than from a blindness of the experimentee?
The perspective changes as it moves, it seemed. So that's causing confusion. The lines seemed to get distorted at times by the crappiness that is online animation. So I'm really not surprised. I did see the changes in height, but that was only among other distortions that occured at the same time.
Interesting stuff. I watched the video twice before reading this post beyond the first paragraph and caught the block growing taller on that second viewing. As such I'm going to have to disagree with those calling out the virtual environment as an excuse for why these changes went unnoticed by the experiment participants.
You might consider doing an experiment on how many readers are convinced of an experimental result based on the resolution of the video you post :)
Was the thin set of crosshairs that moved around the video present in the virtual reality scene?
It seems to me that far too many of the commenters are allowing their responses to this experiment to be influenced by the lousy quality of the online video (which, on my setup, displays at about the size of a postage stamp). There is probably some psychological term (perhaps Dave or Greta or someone would know) for being biased by perceptual first impressions like this, even when it is explicitly explained (Dave has said it more than once now) how and why that visual impression is misleading.
Yes, on this video there are other, artifactual, apparent changes, and the change of interest is not strikingly obvious (unlike most of the very obvious changes, once you know, used in other types of change blindness experiments) even when you know about it. However, the change is only a few millimeters in the image on my screen. In the real experiment, the subjects were in a VR environment, so presumably the display filled their visual field, which means that everything, including the length change, would have looked much larger.
Also, quite apart from the generally very degraded quality and small size of the video, it has had cross-hairs superimposed on it to indicate the direction of the subject's gaze. The paper makes it clear that these were not present during the actual experiment; the subjects did not see them. It looks to me very much as though the worst artifact, the apparent splitting of the third block, is actually caused by the cross hairs (presumably in interaction with the compression of the video stream). What is more, the flickering and skittering cross hairs are distracting in such a way that I strongly suspect they make the length change look even less salient to us than it otherwise would, whereas, of course, the real experimental subjects were not subject to this distraction.
pi* says the results are due to task demands. Clearly demand is affecting performance because the rate of noticing of change is different in the three conditions. However, that is surely (part of) the point of the experiment, not a flaw in the design. Yes, when you are told that height is task relevant you are more likely to notice it changing. (And I don't doubt that if subjects were given the task of reporting when a block changes in height as they move it, they would succeed most of the time. The change blindness phenomenon is not a matter of blindness to all changes, but rather of relative blindness to changes to which ones attention is not very explicitly drawn.) What is important, however, is that even when there is a demand to pay attention to height (in conditions 2 and 3) most of the time the change is still missed, and even when subjects have already seen a height change and are expecting one in later trials, it still generally takes them a couple of tries before they see it, even in condition 3.
Final point: change blindness is now a very well established phenomenon; that has been demonstrated through many different sorts of experimental design. The fact that it appears in this experiment too is not really all that startling. What this experiment adds to previous work is that change blindness can occur even when subjects are looking directly at the relevant place when the change occurs, even when task demands bias them toward attending to the relevant feature (but does not positively bias them to pay that attention at precisely the right moment), and even in relatively simple visual environments. (There is independent evidence to support the first, and, I think, the second of these points; very possibly the third too.) As the authors point out, this undermines explanations of change blindness in terms either of gaze direction or of short term memory load; but for someone who already knows about change blindness (which, admittedly, is a surprising phenomenon when you first find out about it), and is not antecedently committed to those sorts of explanation of it, these results are interesting but should not be particularly unexpected.
Sorry, more (in response to JK):
The experimenters are not being "blind" to anything, rather, JK seems not to understand the point of change blindness experiments. Of course in natural conditions we usually do notice changes. The point is that when you artificially take away or mask certain of the cues that normally draw our attention to changes then people don't see them, even though they are still available to be seen (and are seen once attention actually is drawn to them). Weight change might, under some circumstances, function as such a cue, although my guess would be that, unless the weight change was sudden and large enough to produce a "haptic transient' (analogous to the sudden visual transients that normally draw attention to visual changes) people would be blind to the weight change too. ("Tactile" change blindness has been demonstrated (Gallace, Tan & Spence, 2007)*.)
In fact, one of the main points of interest of this experiment is that in certain respects (obviously not others) it uses more naturalistic conditions than many other change blindness experiments. It does not momentarily blank the screen when the change takes place, or have people carrying something opaque walk between the subject and the person who gets changed. It does not arrange for the change to happen when the subject is saccading, or looking away from the changing object (indeed, the cross-hairs indicating gaze in the sample video show that the subject was looking directly at the block during at lest part of the time that it was changing). It does not require subjects to focus all their attention on something else (like counting ball tosses so you don't see when the gorilla appears). It doesn't even arrange to have a bunch of other distracting stuff appear at just the moment when the change takes place. The change is masked only inasmuch as (a) it is relatively gradual, (b) happens while the block is moving anyway, and (c) does not happen at the very moment when the task requires subjects to be paying attention to the height of the block.
Nearly all change blindness demonstrations involve "magical" type changes. They are not designed, as JK appears to think, to show that we don't notice most real world changes. Generally we do notice them. Rather, they are designed to reveal something about the workings of our visual system (for one thing, that visual experience it is a lot more dependent on attention than is commonly realized, and, perhaps more arguably, that it is not the experiencing of a visual representation formed in the brain) by exploring the conditions (thankfully rare outside the lab) under which it fails.
* Gallace, A., Tan, H.Z., & Spence, C. (2007). Do "mudsplashes" induce tactile change blindness? Perception & Psychophysics, 69 (4), 477-486.
i think the flaw here is in the post itself. with a provocative title like "Objects changing right before our eyes -- and still not being seen," one expects to see something dramatic demonstrated, like the basketball gorilla video or the follow-up link dave posted of the person switching (i'm curious - did they use a set of identical twins for that experiment?). those amazed and astonished me and i was expecting another example of that.
if you offer up video proof it helps if it actually resembles something.
thank you for your interest in our research. I'm glad it stimulated a lively discussion (although much of it focused on video quality which was very good inside the head mounted display). Some questions have been addressed in subsequent work such as the paper: "Task Demands Control Acquisition and Maintenance of Visual Information", Jason A. Droll, Mary M. Hayhoe, J. Triesch, and Brian T. Sullivan. J. Exp. Psychology: Human Perception and Performance 31(6):1416-1438, 2005.
Here are some quick points:
- if subjects are explicitly told at the start of the experiment to look for such changes and to put the changing objects into a "trash bin" they will still miss the changes
- the effect is still there if you change categorical properties such as the color of an object from blue to red (but it may get weaker)
I had to smile at the comment on the "flawed experimental design". It's always good to be critical! ;-)