Daniel Simons interview, continued: Change Blindness

Posted by Dave on May 18, 2010 | 7 Comments

Last week I interviewed Daniel Simons for Seed Magazine. But the interview that ran there actually only included a little over half of our discussion. In addition to his work on inattentional blindness, Simons is also a leading figure in a different line of research, called change blindness.

Here’s a quick video demonstrating the phenomenon of change blindness:

It’s uncanny how difficult it can be to spot changes occurring right before our eyes. In this case, it seems impossible that anyone would miss it, but Simons and Daniel Levin found people missed this sort of change around 50 percent of the time!

Simons and Christopher Chabris’ book The Invisible Gorilla, which covers both change blindness and inattentional blindness, comes out today (Amazon, Barnes & Noble, Powell’s). I’ve read an advance copy provided by the publisher, and I highly recommend it.

Munger: You’ve studied change blindness for over 13 years. Tell me a bit about that field and your contributions.

Simons: I got into it indirectly. I was studying how kids can categorize different kinds of objects, and I was trying to see if they would notice some kinds of changes more than others. It turned out they weren’t noticing anything. Adults too. I was also studying motion picture perception with Dan Levin, and we were looking at whether people notice editing mistakes in movies. There was kind of a convergence of research on it in the early days. Ron Rensink was starting to use his flicker task, we were starting to do motion picture perception research, and most of it built on some work by George McConkey some years earlier.

Ron Rensink’s work was triggered by a presentation at a conference in Vancouver in 1991 that involved making changes as people were moving their eyes from one place to another. John Grimes showed his demo to illustrate the changes, but of course you can’t have eye tracking for everybody in the audience. They were showing how the stimuli changed, and on occasion, somebody in the audience would miss the change. For Rensink, that suggested that maybe you could do this without eye tracking. So he developed his flicker task as a way of doing that.

That was happening when I was in grad school; I didn’t know anything about that presentation, I didn’t know anything about those studies. I was doing object recognition work, and motion picture perception work independently, and Levin and I started pushing this idea that we don’t notice changes in movies, which people had known for decades in movies.

The early days, I think, of the change blindness work, were much more of a demo nature: It was, how big an effect could you make, how much could you make people miss? The more parametric studies started in the mid to late 1990s, when people were interested in how much you were actually retaining from one moment to the next, or what kinds of things would you miss, what kinds of things wouldn’t you miss, what role would attention play, and that sort of thing. So that was all in the change blindness world.

Munger: Can you briefly explain what a parametric study is?

Simons: Sure. So in change blindness, there are really two different kinds of phenomena that are probably related but they might be somewhat different. There’s incidental change blindness, where the change happens unexpectedly, you’re not looking for it, and you’re asked whether you noticed it after the fact. A lot of my early studies worked that way: we showed someone a movie, then we unexpectedly changed something in the movie, say someone’s clothes, or the actual actor in the scene, and find that people didn’t notice, and we asked them immediately afterwards to describe what they’d seen, and they didn’t know anything had gone wrong. With those sorts of studies you just get one shot, it’s just one trial. But you can also do studies when you intentionally change something in the scene, where people know there’s going to be a change, they do their best to find what’s changing, and they still sometimes miss it. The nice thing about that is you can do it over and over again, and systematically vary the nature of the change, the size of the change, the location of the change, across trials to see what factors matter.

Rensink’s flicker task fits that bill, where you have two images that alternate repeatedly with a blank between them. It takes a while to find the changes in most contexts. And you can look to see if there are systematic differences in who notices which change, how quickly people notice different kinds of changes. You can systematically or parametrically vary how long the displays are on, how long the gap is between them, how many things there are on the display. It’s hard to do that with an incidental task because you really only get one trial per subject. So you could do it but you’d need thousands of subjects to do the same thing that you could do in a thousand trials with one subject.

[Here's an example of the "flicker" paradigm]

Munger: So how has that work changed over the years?

Simons: The early days of this sort of research in the 90s was very much of this demo character: Can we change the person you’re talking to in an interview? That phase for change blindness is done now. We spent about four years or so in my lab and in lots of other labs one-upping, showing what you could miss. But the theoretical implications of that were no longer quite as critical. We now knew people would certainly miss things they were not focusing attention on, but they also missed things they were focusing attention on if they weren’t actively comparing it to something else that had changed. But the demo phase of change blindness more or less wrapped up in the early 2000s. There are still occasional demo, one-trial change blindness studies out there, but for the most part that field has moved toward more systematic studies of how much information is preserved when people fail to notice changes, or, are there systematic individual or group differences that determine what people notice or what they don’t. Or, are there individual differences in attention, personality, or memory, that will predict whether people will notice. What role does salience in the scene play in whether or not people are drawn to the change?

So those are the sorts of things people are looking at now. As with any phenomenon like this there’s the underlying phenomenon, and the phenomenology of failing to notice change. And there’s still work on that, but there’s a lot more studying of the task too. Also, because it’s easy to do, I think some people who don’t really know the literature and the methods have started trying to do these experiments and publish the results. There’s more work that’s less theoretically interesting. And there’s still plenty that is.

Munger: What are some real-world applications of this research?

Simons: We all experience this sort of stuff. For example, an experience I have all the time is that I’m working on my computer, I’ve got multiple applications open, and I don’t realize that I’ve actually switched applications, and I don’t realize the menu has changed. That’s just a simple practical example that doesn’t have any consequence, but the same sort of mechanism is at play there.

Munger: In someways change blindness is similar to inattentional blindness studies like your “Gorillas in our Midst” study because you’re not noticing something, but as you pointed out they are different because it’s a one-shot deal with inattentional blindness—

Simons: —and sometimes with change blindness, too

Munger: Yes, but are there similarities as well? Do you think there are some cognitive processes that are the same between the two phenomena?

Simons: Yes. I’m sure. Change blindness is more complicated in many ways. Inattentional blindness is in some ways a purer example of a failure of perception, or awareness of perception. It’s a cleaner example of the mechanisms involved—it is due to attention being focused on something wrong, or in a limited way. Change blindness is more complicated because it can happen for any number of reasons. In change blindness, you’re failing to notice that something is different from the way it was a moment ago, which means you have to encode it in the first place, then you have to remember what it was that you encoded, and then you have to compare what’s there later to what was there before. Any one of those stages could break down. When you fail to notice a change, you don’t know whether people never encoded it in the first place, or forgot it, or remembered it perfectly and never bothered to make the right comparison. So it makes it much more difficult to say exactly what went wrong. It also makes it interesting to study. But I think both are typically attributed to limits of attention in some form. The interesting question for change blindness is how much actually is represented. How much have have you actually stored but just not bothered to access? Or not compared? Or stored in the wrong format to access, that would have allowed you to detect the change? Say you look at a scene. If you look at the pre-change scene, there’s nothing about that scene that tells you what changed. If you look at the post-change scene there’s nothing about that scene that tells you there’s anything wrong, so it has to be a comparison between the two.

For inattentional blindness you can look right at that scene and the thing that’s odd is right there. So there’s nothing to compare to, there’s no memory demand. I think there are common processes, there are limits on awareness, there are limits on attention, that contribute to both, but there are more mechanisms that can go wrong to cause change blindness.

The change blindness studies, the gorilla studies, they do the same thing. When we change the person you’re talking to in the middle of an interaction, that’s not something that ever happens. But if we didn’t make that change we’d have no way of knowing whether you were actually encoding details about that person before or afterwards. We’d have no way of knowing how much information you were retaining. So it’s a way of seeing what the default assumptions are of attention and awareness. Metaphorically it’s the same approach. You break the system, in this case the cognitive parts of the system, not vision, but you’re breaking down what typically happens in order to see what happens by default. Do you, by default keep track of the height, sex, hair color of the person you’re talking to? Does that happen automatically? Do you keep track of their identity? We’d really have no way of knowing that unless we test whether or not you had.

Munger: But there are changes in individuals that people do notice pretty readily, right?

Simons: Yes. We tried changing the sex of a person you’re interacting with, or the race, and those are almost always noticed. I think that’s interesting. When people successfully detect a change, it tells you what sorts of information they’re keeping track of. When they fail, you don’t quite know. If you detect a change to the race of a person, but not other sorts of changes, that tells you something about what was encoded.

Munger: I always think it’s interesting that parents of little kids, so small that you really can’t tell what sex they are just looking at them, tend get really offended if you don’t get the sex right.

Simons: I always find that really funny. That’s why people dress their kids in pink or blue.

Comments

4 Tweets