On Learning – Part 2
So we now have a simple model of learning that we can summarize : We have an experience that causes us to recall a previous similar experience, and use it as a guide as to what to expect. In other words:
EXPERIENCE -> RECALL -> EXPECT
Although this model seems to make intuitive sense, it is wrong—or at least, incomplete—in a very important way. To see that, think about what happens when you have a lot of experiences in the same essential situation. For example, you may have been fishing hundreds of times. If so, you will have experienced many different episodes in which you learned important things. So what happens when you go fishing again? Do you remember all of these episodes at once? If so, how do you manage to mentally process all those memories rapidly enough to extract and use the knowledge they contain? This issue has been referred to as the “paradox of the expert”—the notion that an experienced person might have so many relevant memories that she would be paralyzed by the effort to make sense of them all. The label is meant to be tongue-in-cheek, because of course real experts are not paralyzed in this way. The question is, why not?
The answer lies in something that psychologists call a schema. A schema is an organizing structure in memory that collects the set of things you should expect in a particular kind of situation. So for example, if you have been fishing a lot, you will have a fishing schema in memory that amalgamates all of the things you have learned from all of the fishing experiences you have had. For example, you may remember which spots are best to fish from, which kinds of fish you are most likely to catch at different times of year, what kind of bait works best for each kind of fish, and on, along with miscellaneous details such as remembering that you should stay out of the water in the winter months, and that if you leave your bait unguarded a raccoon may steal it. By collecting this knowledge in one place, the schema helps you avoid the paradox of the expert.
The simplest model of how schemas are built is that they are based on the first episode of a particular sort that you experience. So for example, the first time you went fishing, you created a schema for this new experience that basically told you to expect everything that happened this first time to happen every time (this is a little oversimplified since you probably would have understood you first fishing trip in terms of some pre-existing scheme—your schema for hunting, for example—but I’ll ignore this detail). From there, your fishing schema gets refined and amended whenever you learn something new from a fishing trip. This raises the quexstion, how do you know that you are learning something new? The answer is deceptively simple: you know you are learning something new when the expectations provided by your schema fail in the current situation.
For example, suppose that your fishing schema includes the expectation that when there is ice on the river, you can cut a hole in the ice and fish through it, but that on this occasion the ice cracks and you fall through. Your expectation that you can fish through the ice has failed. But simply removing this expectation would be too extreme, because sometimes you can fish through the ice; sometimes you can’t. To create an appropriate revised expectation, you have to explain the why this works sometimes but not others. Then you can revise your expectation—for example, to note that you can fish through the ice as long as it is thick enough to support your weight.
So expectation failure—the failure of your schema to accurately predict what is going to happen—is the trigger for learning. This is not exactly obvious, but if you think about it you can see that there really is no other way it could work. After all, when your schema tells you exactly what to expect in a situation, what is there to learn? It’s analogous to a scientific theory, which, if confirmed by experiment, remains unchanged, but that must be modified when a violation occurs.
This is of learning from expectation failures, by the way, is where the notion of “surprise” comes into the picture. As I noted above, surprise and emotion are the two features that tend to make an episode stick in memory. It is interesting to note that surprise triggers the production of the neurotransmitter dopamine in your brain, and dopamine has the effect of strengthening the active connections between your neurons. This is thought to have the impact of making it more likely that you will remember in the future whatever it is that you are thinking about now. Emotions—especially pleasure and pain—also trigger dopamine production. So there is at least a plausible neurological theory of how surprise and emotion determine what you remember.
By repeatedly experiencing and explaining expectation failures, you create progressively more sophisticated theories over time. Think, for example, about how you learn to get along with a particular person over time. At first, your schema for interacting with them may be more or less a copy of your generic schema for interacting with anyone. But over time various expectations in that schema will inevitably fail. To take a simple example, you may have a supply of jokes that make most people laugh, but you may discover your new acquaintance doesn’t find those jokes funny. From other interactions you may decide also decide that this person is quick-tempered, or sentimental, or iconoclastic, or shy, or any of a hundred other things. These hypotheses allow you to replace very specific expectations—say, that person X doesn’t like your joke about the three Neanderthals that walk into a bar—with a much more general expectation, for example, that person X doesn’t like anything having to do with negative stereotyping.
As I noted above, the theory of how this kind of reasoning works is far from worked out scientifically. The key point to hang onto, though, is that explanatory reasoning is triggered when an expectation fails. I want to make one other quick point about this. You might ask why, if a person knows enough to explain an expectation failure, they wouldn’t have anticipated that failure in the first place. The answer is that expectation failures tell you what to focus on. I’ll give a quick, slightly silly example from a more modern, though still somewhat ancient era: my graduate school career. During this time, I sprained my ankle playing football and ended up on crutches for the first and only time in my life. As you might imagine, there was a lot to learn from this. One thing I learned quickly was about getting sodas. It was my habit, when I got my daily diet coke out of the soda machine in the lab building where my office was housed, to pop the top immediately and take a quick drink. The first time I did that on crutches, I soon realized that now that the soda was open, I was not going to be able to get it back to my desk without spilling it all over. Notice that while I could in principle have figured that out in advance, it was actually pretty unlikely that I would happen to think of it. Once the failure focused my attention, however, it was trivial to understand what was wrong. That’s why expectation failure in general plays such a vital role in learning: expectation failures show you where to focus your thinking in order to improve your understanding.
So we can summarize our theory so far in something like this way: Learning is based on the ability to remember our experiences and recall them in similar circumstances. When an experience is repeated, our mind creates a schema to represent what we can expect in the typical instance of that experience. We use the expectations from this schema to guide our subsequent behavior. When an expectation fails, we seek an explanation, and based on that explanation we revise the expectations in our schema. Through the action of this process over time, our schema become more accurate and more detailed, and we will become better able to function in the environment that it describes. We might summarize this schematically as follows:
EXPERIENCE -> RECALL -> EXPECT -> FAIL -> FIX
(Where “fix” means explain and then revise the schema.)
The last point I want to make about learning is that in general, people seem to learn much more from doing things themselves than from watching others do them. As it stands, the model I’ve outlined here doesn’t do much to explain that difference. After all, you can form expectations and revise them in response to failures whether or not you are involved in whatever activity is going on. I think there are three big reasons why you learn better by doing:
- First, when you do something and the expected result doesn’t occur, that almost by definition means that one of your goals is thwarted. That, in turn, as I discussed above, means that you are more likely to learn and remember the lesson that when your goals are unaffected.
- Second, when you have to make a decision, you attend to a lot of things you might not attend to when you aren’t involved. For example, you can watch people fish all day without really paying attention to when they cast their lines, how far they cast them, how they use their equipment, and so on. When you are forced to make decisions, you are forced to notice a lot of things that you wouldn’t have noticed otherwise, and thus you have a much richer memory of the experience
- Third, to make a decision, you have to settle on a theory of what you expect to happen, even if it is only a wild guess. When you are just observing, you don’t have to have any theory at all. So, for example, when you watch someone fish you don’t need to think about how hard they should pull on the line once they get a strike. But when you are fishing, you are forced to make some kind of guess about this, and your guess will be either confirmed or refuted by what happens next. In other words, your guess becomes an expectation, which, even if wildly off base, can start you on the cycle of expectation failure and explanation that will lead you to a better theory.
Taking all this into account, we are most likely to learn effectively when we make a decision in service of a goal, which forces us to adopt expectations, which can then be revised through experience and failure as described above. We might render this as:
GOAL -> EXPERIENCE -> DECIDE -> FAIL -> FIX
In other words, we start with a goal we are trying to achieve, we experience an episode in which we are trying to achieve that goal, we are forced to make a decision about how to address the goal, and that decisions is based on expectations that may fail and require fixing. Of course, this model is really a cycle–we would go around the cycle numerous times in any complex situation, pursuing multiple goals, making multiple decisions, and so on.
So that is my answer to the question “how does learning work.”
There are a number of things to say about this model, but probably the most important is that in this model we learn by interacting with the world and refining out expectations about it, especially in the course of practicing the skills we need to learn. This model looks nothing like a lot of what goes on in a typical classroom—there is no room in it for lectures or for rote learning. The appropriate role for a teacher based on this would be, first, to ensure that this sort of experience occurs, and, second, to be the coach when an expectation failure occurs.
Although this is a very simple model, it tells us a surprising amount about how an effective instructional design should look. If you believe that that learning works this way, even approximately, then the goal of training should be to drive learners through this cycles as efficiently as possible. To do this, we must:
- Give the learner a motivating mission (the goal).
- Put the learner in an appropriate context where he or she can act to achieve that goal in a realistic environment.
- Force a challenging decision that may lead to an expectation failure, then help the learner fix the expectation by explaining the failure.
- There are then two key things we can do to help the learner explain a failure:
- One is to show the consequences of the failure, which allows the learner to get feedback on what went wrong as they would in a natural environment.
- The other is to provide coaching, which helps the learner to understand expert theories and models that can help explain what is going on.
We can summarize the instructional design model that best fits our model of how learning works thusly:
MISSION -> CONTEXT -> CHALLENGE -> CONSEQUENCES & COACHING
If you agree with this model of how learning works, then this is the kind of training you should want to design. If you don’t agree… then it’s time to get to work on your own theory.