A mental model...
 |
| Hamster-powered search (P/C Gemini) |
.. is just the collection of what you know about some thing: what that thing is, the way it looks, how to use it. Just as importantly, a mental model also tells you (a) what a thing normally looks like, (b) its normal functioning, and (c) how to fix it when things go wrong.
That’s usually the way people define MENTAL MODEL. But wait… there’s more!
You should know that the mental model doesn’t have to be accurate, in fact, they rarely are. Few of us have a deep and accurate mental model of the way our car works, yet we use them all the time. (Can you really tell me how antilock brakes work?)
For most things, you don’t need a completely correct model—you just need to know enough to know what it is, how to work it, how it behaves, and how fix things when the process goes awry.
A mental model is whatever a person believes to be true about that thing--it's appearance, the way it acts as it operates, even how long it takes to do its normal behavior.
Does your fancy AI thing all run on top of a stack of hamsters? Sure. Whatever. As long as you can use it and as long as your mental model makes accurate predictions. (See image above, and remember that famous April Fool's Day joke about Google's Pigeon Rank. I'm sure there are people who believe this is true.)
Explaining how something works in detail is often very, very hard and to use your thing, people usually don’t need all of those details. Almost nobody can explain how electricity actually works, yet we all use it successfully every day. The trick is to create enough of a mental model to be useful.
A mental model guides you in your use of a tool or system. The mental model tells you “If you do X, then Y will happen.” If you turn on a light switch, the light bulb goes on. And if the bulb doesn’t go on, the mental model will give you some ideas about what to do next to fix it.
Having said that, people can have very different mental models of online search… and in the age of AI, it matters if your mental model is correct enough to be useful. The reason to have a mental model is to be able to use a system AND make simple predictions about what the system will do.
In other words, I want to talk about mental models for information seeking, and along the way, discuss mental models for information and sensemaking.
Let’s start by talking about mental models people have for online search systems.
Over my years of doing research at Google, I’ve seen a lot of interesting mental models. Here are the three most common mental models that I've seen out there.
1. Database: when you issue a query, you get back everything that it available on the internet that’s on the topic of your query. If you think about it for a bit, one will realize that this can’t possibly be the case—there are too many hits for any search engine to return, and the idea of a “complete set” of results is just dreaming. By design, a database WILL return everything that matches the query. They’re superficially similar, but the model falls apart if you test it carefully.
2. File search: some people have used a file search tool like Apple’s Spotlight or Windows’ Start Menu search tool. They both search the file system for matches in filenames or file contents. But these tools search a slowly-changing set of local content. Internet search systems do their best to index a vast, seething mass of constantly churning content.
3. Library index: people who grew up in libraries love to think of internet search as a kind of library catalog. These folks believe that everything is described somewhere on a small card by a talented librarian who lists the title, the date, the author, and the major topic areas of the document. Wikipedia is great, with their articles on nearly every possible topic, but that’s not really complete in any sense. (It’s fairly easy to find documents on the internet that don’t have a title or even a clear topical focus.)
The thing about these kinds of mental models of internet search is that people have formed expectations for what the system is doing when they use it.
People with a database-model expect complete results and are surprised when they don’t get it. And since most real databases have very well-defined operators like NOT, these folks are surprised when a search result seems to defy the query.
If you’re in the business of creating new applications (including AI-based apps), you—the designer or system builder—have to know what MENTAL MODELs your users have of the thing you’re building. And if it’s a totally new kind of thing, realize that people are going to use the closest MENTAL MODEL they can think of and make predictions with that.
So, one of the most important things you have to do as a designer is to communicate a good mental model.
The mental model should be clear about what your thing is (and what it isn’t), as well as set strong expectations about what normal operation is, and what to do when that operation goes off track.
We’ve written before about the lack of understanding of Control-F (i.e., find in text, the most important skill). The root cause of the problem is that people’s mental model of a web page does NOT include text-findability as a legitimate operation.
(It’s an interesting self-check: think for a second right now. When you’re looking at a webpage in your browser, what kinds of things CAN you do? We’re talking about finding text, so you know that. Can you find a synonym for the target of your search? Is that even possible? If so, how? You probably know you can export or save the web page as a file, but can you somehow extract and save all of the images on the page? What about getting the definition of an unfamiliar word? Possible?)
What to do about giving useful mental models to our users?
UX master Jakob Nielsen wrote about this a while back in his article about mental models. (This is my paraphrase of his advice.)
If your user has a different MENTAL MODEL than you, there are two different options:
1 Make the system align with the users' mental models — assuming most users’ models are similar. Your fancy new document editor better look / operate like other document editors, or you’re fighting an uphill battle. This is the best approach: if people look for something in the wrong place, then move it to the place where they look for it. If your system doesn’t offer a function that all of the other systems offer, you’d better have a great reason why not.
2. Improve users' mental models so that they more accurately reflect your system. You can, for example, explain things more clearly, make labels clearer, and generally offer a great communication system in order to guide users to form the right model. Analogies and metaphors work well here--just be sure you know how your user population understands your metaphorical language. (Caution: It's easy to get your metaphors wrong. See this paper: "On the problem and promise of metaphor use in science and science communication")
In my next post, I’ll talk about how people think the underlying search systems are built, and what happens when users don’t understand the data over which the search is run. It’s more complicated than you might expect.
Keep searching.
No comments:
Post a Comment