![[HOT Lab logo]](http://www.carleton.ca/hotlab/hottopics/Images/logo_hotlab.gif){kind=logo}
![[Carleton Log]](http://www.carleton.ca/hotlab/hottopics/Images/logo_carleton.jpg){kind=logo}
|
|
 HOT Topics!
Interview with Donald Norman on Mental Models
(c) 2004 by Avi Parush [Interviewer:] When you go to symposiums, seminars and conferences do you still find yourself talking about mental models? [Norman:] Very seldom. [Interviewer:] Is it because people aren't interested any more, or because it's an old topic? [Norman:] Well, there may be a little bit of that -- it is an old topic. But I think there are two reasons. One has to do with me; the other has to do with the topic. The first reason is the topic. I don't think it's a direct research concern any more; it's an accepted concept that is used uncritically. That doesn't mean it's well understood, but it has passed into the accepted folklore of science. The second reason has to do with me. I don't stick to any research area for very long. I am continually shifting my gaze and moving into new areas. So I have never maintained any particular research area for more than ten years and often only five. My main focus in the last few years has been on applications in cognitive science and in particular to the domain of design. [Interviewer:] Let me ask questions that put together issues to do with mental models and user interface design in the industry. For example, until recently I was working a lot within engineering-oriented environments, industrial-engineering to be exact, and I was just about the only cognitive psychologist there and the industrial engineers kept telling me "there is no such thing as mental models -- it all has to do with performance, what do we care about mental models?" And I have to admit I didn't know how to convince them or explain to them that there is such a thing and that it is important. What would you say to those people? [Norman:] I would ask them how they know what to do when there's an unexpected problem. So what kind of equipment were you working with? [Interviewer:] Process control, medical imaging, network management, or semi-conductor equipment. [Norman:] In process control for instance, there might be an unexpected rise in temperature with a simultaneous decrease in pressure. So, you might ask these engineers "what do you think is going on?" The point being not whether their answer is right or wrong but how they reach their answer. I can tell you the way you answer this question is that you do a mental simulation. You take your knowledge of the way the process is working and imagine what could give rise to an increase in temperature and a decrease in pressure and since they are normally correlated (normally pressure increase is caused by temperature increase) they would have to do some clever simulation. This "imagining" process is what we mean by mental models; how people take their understanding of something to simulate what would happen in a novel situation. [Interviewer:] So mental models are not only for understanding what there is now, but also for being able to cope with unexpected, unfamiliar situations? [Norman:] A good question to ask is "where is the mental model needed?" It's used in learning. I believe that if you have a good conceptual understanding it gives you a framework for learning. And it's of use in unexpected novel situations. But for normal performance you don't need it. Normal performance you do by rote memory. You just simply know that when this happens you do that. Mental models are almost no use for operating a device for routine operations. In fact mental models might get in the way. If you had to derive everything you did it would slow you up. [Interviewer:] Do we develop mental models through routine activities or through guided learning? How do I get to the point where I have a mental model that will help me in totally new and unfamiliar situations? [Norman:] Either: 1) you have been taught or otherwise learned the principles of the mechanisms you are using or 2) you have deduced it from your observations. My point is that if you don't provide someone with a mental model they will make one up. They sort of have to in order to figure out what they're doing in unexpected situations (including first time learning). And of course, if they make their mental models up, they're apt to be wrong. So, that doesn't mean they don't have them; they just have erroneous ones. That's one reason I think it's important to teach appropriate models. [Interviewer:] The most accepted definition of mental models by you or others is that we create a mental model through experience with devices. Is it possible to develop a mental model before I have any experience with a device? [Norman:] Sure. [Interviewer:] Would you call that intuition? [Norman:] No. Intuition is a name that we give to knowledge structures that are so well "routinized" that they occur subconsciously -- but intuition is something that requires a tremendous amount of learning. But give me any new phenomenon - ask me about some topic on which I am not an expert and I will take whatever knowledge I have and derive a conceptual model to predict the behavior. That conceptual model is in my head - it is mental - hence the name, mental model. Basically, we take whatever we know and we expand it. [Interviewer:] As a designer, should I try to figure out what the mental models of users could be before they interact with the system so that when I design I do it in such a way that I will match that mental model or should I design in a way that I will form their mental model? [Norman:] That's a two-part question. The designer has an obligation to provide an appropriate conceptual model for the way that the device works. It doesn't have to be completely accurate but it has to be sufficiently accurate that it will help in both the learning of the operation and also dealing with novel situations. So that's part one of your question. In part two you asked: " should I try to figure out what they come in with and build on it?" Here, the answer is more complex: both yes and no. There is a philosophy of design that uses metaphors and a metaphor is a stepping-stone to a mental model. So if I use a metaphor I am trying to let the user know that this device is like something they already know about. So you form your new mental model by taking something you already know and modifying it slightly to understand the new device or interface. Hence the goal of metaphorical design is to make it much easier to learn. I'm not convinced that's necessarily a good thing. I agree that a metaphor makes the device easier to learn but it may also bring in as many difficulties as advantages because a new device is always different from the old device in some way (perhaps new features) and a metaphor doesn't necessary help here. In fact, it may get in the way of understanding the new features. [Interviewer:] So a metaphor may get in the way of developing an appropriate conceptual model? [Norman:] Yes, I think so. It brings excess baggage - some things that may not be appropriate to the new design. [Interviewer:] I have concerns that the user's primary or original mental model may conflict with what I, as the designer, would think should be the appropriate mental model for the user. [Norman:] This does not mean that you should ignore what the user already knows. At times, if you design something that really conflicts with what they understand that may cause great difficulty. Ideally you should make something that is consistent with what they understand unless there is some strong reason not to do so. [Interviewer:] I want to try to relate the topic of mental models to another topic that you eluded to in one of your earlier articles Some observations on mental models where you suggested mental models are parsimonious and one of the reasons is that I could kind of generate one simplified rule to apply to various devices. I would like to ask you about cross-device mental models. Is there such a thing? I'm asking about it because we have a lot of practical design problems that have to do with cross-device applications. For example, suppose I build a web site and users learn to use it with their desktop computer but then they start to use it with mobile devices like cell-phones or PDAs. I could be concerned about their mental models but I'm also practically concerned about how to design a particular device with access to the web site. Are users going to come with the same mental model for that web site regardless of the device or should I assume that the mental model is device dependent and I should design the web site a bit differently to match the device? [Norman:] I suspect there are multiple models at play here. One group is for the different components of the display and selection device, another is the website. So, on my desktop computer I have a mouse, a keyboard and a very large display screen and I have developed an understanding about how these three things work and the relationships among them. When I go to a web site I have an understanding of the traditions of web site design - what a hyperlink means, how to access hyperlinks (whether I click once or click twice), what happens when I do that and so on. I suspect the models we have of web sites are a vast amount of confusion because web sites are not consistent. When I click on something should I click once or twice? Well, the computer requires twice but the web site requires once. When I click on something, do I move to that location (that is the window changes) or do I open up a new window for the new location but the old window is still there? This varies from web site to web site and it's impossible to predict what will happen. There is no way of telling just by looking at the site. So web sites themselves are confusing, and, as a result, our mental models of websites are also confusing, or perhaps, site-specific. Now consider the impact of the technology used to view a website, and the need for different mental models. If I switch to a new device such as a cellphone or another, small, portable browser, and I go to the same web site, and I don't have a mouse anymore and my screen is tiny; it may or may not have graphics; it may have colour or it may just have black and white; it may have a touch screen or it may be I just have arrows that go up and down. And as a result of these differences in the context, there is often great confusion over what it means to select a hyperlink. But this not because of my mental model of the web site. This is because of my mental model of the device and how it interacts with a web site. So, to answer your question, I believe that the design should be device specific, but because the model of the device is different, not because of the model of the web site. [Interviewer:] Would it be fair of me to conclude that mental models tend to be device-dependent rather than device-independent? [Norman:] No. How about: mental models are context sensitive. The model I use depends on the context. When I drive an automobile I probably have a whole bunch of mental models that are involved. I have a model about how the radio works, I have a model about how the brakes work and how the steering works. In fact, it took me a lot of training to understand how the model of the steering works with the model of the brakes. I took a special driving course to learn about that relationship. For example, most people think that the accelerator makes you go faster, steering makes you turn and brakes make you go slower. So there's a whole set of mental models that we have about driving that have fairly simple modes for the accelerator (speeding up or not), the brakes (slowing down or not) and the steering wheel (turning left or turning right). Well, I took a two-day driving course and learned a very different way of thinking about driving, so I developed a very different mental model. So, the accelerator and brakes and, to some extent, the steering wheel are ways of shifting weight. If you are going into a turn it helps to de-accelerate a little bit to put more weight on the front wheels so that when you turn you are less apt to skid. [Interviewer:] So this new mental model changed the way you drive? [Norman:] Absolutely. And if you talk to racing car drivers that's what they talk about -- they talk about weight control and contact patches (how many square inches of a tire are contacting the road). That's what determines whether you are going to skid or not. Mind you, if you really want to get around a turn in the fastest possible way, as you get towards the end of the turn, you want to skid on purpose, so you brake and turn your wheel hard, which takes the weight off the rear wheels and therefore causes them to skid, turning your car a little faster than otherwise. After this you must get the weight back on the rear wheels, and you do this by straightening out the turn and accelerating. So there's a heavy use of a weight shifting model for advanced driving. [Interviewer:] Do you think that would be an appropriate way to teach beginner drivers to drive? Should you right away convey the correct mental model? [Norman:] No, it might be a bad idea for teaching beginner drivers. Let me explain something. Another example. You were probably taught what to do when a car skids. [Interviewer:] Yes, I was taught to turn the wheel into the skids. [Norman:] But I bet you were not taught that there are two kinds of skids. As a result, you have an incomplete model of what is going on and if you ever got into a skid, you are quite apt to discover that your knowledge would not help you one bit. First, you probably wouldn't remember it and secondly, if you did try to use it, it would transfer the skid you were in into a skid in the opposite direction - just as dangerous. I spent four hours skidding in this course. First we were given lectures about skids. Then we went to skid pads - big wet pieces of pavements that were oily and wet. It was easy to skid and therefore it was easy to experiment and practice. The basic point is that there are two kinds of skids -- you could be skidding with your front wheels or your rear wheels. Skidding with the front wheels is called under-steering and skidding with your rear wheels is called over-steering. It's called under-steering because when the front wheels skid when you try to turn, the car turns slowly, or not at all. If the rear wheels skid, you turn even more sharply than intended -you oversteer. When it is the front wheels that are skidding, you need to get more weight on the front so braking is useful. If your rear wheels are skidding it turns out the worst thing you can do is brake. What you need to do is accelerate because you need to put more weight on the rear wheels. This is consistent with the model. (Note that putting this into practice is difficult. I assure you that my two-day course wasn't nearly enough time to learn these skills very well.) [Interviewer:] Why do you suggest not to teach young drivers all that to make them good drives to start with? [Norman: ] Beginners have enough trouble just controlling the car. But now we're not talking about mental models we're talking about instructional philosophy. [Interviewer: ] Aren't they related? [Norman: ] They are related but somewhat different topics. Instructional philosophy is how you teach a topic. I think you have to teach knowledge where it is appropriate. If you teach something before people are ready it's not useful and it may even be harmful, because it may make the topic too complicated. Drivers should first learn the mechanics of driving. They should practice, they should get to the skill stage where they're no longer worried about how they hold their hands on their steering wheel and how they can synchronize the steering and the shifting and the signaling and all the stuff you have to do. All this basic stuff overwhelms novice drivers. You have to get to the point where that's automatic. Then you can start talking about the more complex things - what happens when things go wrong. [Interviewer:] So complex conceptual models are gradually acquired through instruction or experience? [Norman:] Right, so if you were to learn to fly fighter planes, training could be five or six years long. You can't just throw everything at people right away. You have to choose the appropriate time. [Interviewer:] Now I'd like to tie this topic of mental models to more recent work on emotional design. If I understand correctly, mental models are more of a cognitive thing -- about understanding or being in control. Do emotions play a role in the development of mental models? Do I have to worry about it as a designer? [Norman:] That's an interesting question. I have argued that emotion cannot be separated from cognition. One (cognition), is understanding and the other (emotion), is evaluating and making an assessment. I had not thought about this relationship to mental models so that's a good question. I think that the answer has to be yes, as a designer you should think about the relationship between emotion and mental models. My colleagues William Revelle, Andrew Ortony and I at Northwestern Unviersity, have developed a three level theory of emotion and affect (Ortony, Norman, & Revelle, In press). I have built on that work in my book on design (Norman, D. A., 2004), where I renamed the levels "Visceral," "Behavioral," and "Reflective." In a paper I just finished with Andrew Ortony (Norman, Donald A. & Ortony, 2003) we said that the visceral level is perceptually-based, the behavioral level expectation-based, and the reflective level is intellectually-based. These three levels play very different roles in the design of products. Now let me think about the roles they might play in mental models. Most important of all would be the expectation- and intellectually-based emotions, behavioral and reflective level processing. A mental model provides an immediate expectation about what you think is going to happen and the emotional system will evaluate that positively (positive affect or valence) or negatively (negative affect or valence). As a result, you will experience hope or anxiety (hope in the positive case and anxiety for the negative case). I've often noticed a trivial example of this when I am driving my car and coming to a traffic light. If I suspect it will turn yellow and then red before I make it through the intersection, I get tense. I think this is strange and amusing - that I tense up because I think the green light will go away before I get there -- that's an expectation-based emotion. That initial feeling is then translated into a full-scale emotion, depending on what actually happens - whether my expectation is fulfilled or not. Consider the impact on our feelings towards the products we use. It's good if we expect something bad that doesn't happen and it's bad if we expect something good that doesn't happen. That impacts the way we feel if we continue using something (like a device) and it may impact what we do about it (continue use or abandon the device). At the intellectual level there is also the causality issue. That is, do we decide to blame the device or ourselves? I just heard a really interesting case study of the way a robot delivery machine was used in a hospital (She, 2004). They eventually had to remove it because it didn't work well. But when you analyzed why it didn't work well, many of the times it was because they blamed the robot for failing even though the robot was performing perfectly. One example occurred when the robot was delivering medicine to nurses' stations. The medicine is in a locked compartment. The robot actually goes to the hospital, goes in the elevator, goes to the third floor, gets out of the elevator and goes to the nurses' station and waits for a nurse to go and take out the medicine and load it up. In the example cited, a nurse goes over and types in a password: nothing happens. And so she thinks that this robot is broken and calls up the service people saying, "Take it away, it's not working again". It turned out she was typing in the wrong password. But it was the robot that was blamed. Because of numerous incidents like this they had to remove the robot. This is in many ways a mental model issue. It's an interesting example of emotional evaluation and functional operation. [Interviewer:] Could it be that in the design of a system where you have machines and people working together to perform a specific set of tasks that you need a much higher level of mental model -- more of a group level mental model? [Norman:] Oh yes, many of these things are social issues. In this case, I think they didn't introduce the robot appropriately. The introduction of the robot should have been treated as a social issue. They needed better training and I think they should have introduced the robot more gradually. The robot should have been used in a really simple situation for a few months to make everybody happy with it, and then it should have been used to do something slightly more complex and so on; but instead they tried to introduce it to the entire hospital all at once. As a result, the robot did not fit into the structural, social, and organizational operation of the hospital. No wonder it failed. [Interviewer:] So, we should have good mental models for any context, it doesn't have to do with machines necessarily? [Norman:] Right, in fact, I have a mental model about how a new technology should be introduced into institutions. [Interviewer:] When integrating new systems into existing systems, we should build a good mental model for that situation. [Norman:] When you told me your first story that you worked on your devices and you said that your engineers said: "What's this mental model crap? It's just a matter of performance. " That violated my mental model of engineers. I expected that engineers would be the ones to get the concept of mental model right away, because that's what they do in their heads -- they figure out how things work in their heads. I was surprised that it was the engineers that were in trouble with the idea of mental models. [Interviewer:] I think the resistance was because it wasn't something that could be observed, measured, or quantified. That was my impression of why they kept their distance from this concept. Is there a good way to measure or map mental models? [Norman:] Amusingly, I had this same kind of discussion yesterday with a group of students who were asking me about usability and they were saying, "Usability was just one opinion against another and what was this usability thing all about?" I was trying to explain to them there was a lot known and it wasn't just opinions. The same issue arises with mental models. That is a legitimate criticism. It isn't really hard to understand the concept of a mental model, especially for engineers. But the important questions remain: What use is it? What do I do with the concept? I think that's what they're asking. On the one hand I think that mental models offer a really valuable qualitative framework for design. On the other hand, it would be of far greater use were it to have a quantitative side. I don't know of any quantitative measures. This gets to the question about where the work on mental models has gone. When it was first introduced, it was an interesting topic, and a number of people tried to do experiments and studies. But these didn't lead very far, so the interest died down. As far as I know, there has been very, very little work in experimental psychology on mental models. And it really hasn't been incorporated into any quantitative work. At this stage, I think the engineer's question to you is legitimate. [Interviewer:] On a personal note, I've been trying recently to find ways to measure and quantify mental models using techniques such as cluster analysis and multi-dimensional scaling so that's why this question is interesting to me personally. This is one of my lines of my research. [Norman:] I think I would also try behavioral measures. If you believe people are using models, then they must have an inference process, and you might be able to find evidence for this. I'd look at reaction time and errors of reasoning - presumably systematic errors that result from the particular models being used. All of these methods, cluster analysis, categorization techniques etc. have to be used. It would be good to incorporate any quantitative results in some formal model like the ACT models that Anderson does or the GOMS models. It would be interesting to see what role mental models play in those simulation tools. [Interviewer:] Thank you so much for your time. [Norman:] You're welcome. Good luck with your research. References Norman, D. A. (2004). Emotional Design: Why We Love (or Hate) Everyday Things. New York: Basic Books. Subscribe/Unsubscribe
|
|
Don Norman's goal is to help companies make products that appeal to the emotions as well as to reason. Norman is cofounder of the Nielsen Norman Group, an executive consulting firm that helps companies produce human-centered products and services. He serves on numerous boards and advisory committees for companies and education. He is Prof. of Computer Science and Psychology at Northwestern University and former Vice President of Apple Computer's Advanced Technology Group. Norman is the author of "The Design of Everyday Things," "Things That Make Us Smart" and, "The Invisible Computer," His newest book "Emotional Design" emphasizes the role of emotions in product design. He lives at