Saturday, May 9, 2009

Fun With Feedback

When I was in high school I participated in an interesting exercise to demonstrate the importance of feedback. You might want to try this out as a party game. It is easy to laugh at others when they do poorly at this seemingly simple exercise, but it is in fact a rather difficult task.

Get a set of children's building blocks, such as a set of hardwood unit blocks, with a variety of shapes and sizes. From the original set of blocks, create two identical sets. Each set should have at least ten blocks in it with at least four different shapes. More variety is better. Having blocks within each set that are similar but not identical, such as two different kinds of triangles, makes it more difficult. Given these two identical sets, anything you can build with one set can be built with the other set.

Arrange two desks or tables, with their chairs, for two people such that there is a level work space in front of each person in which that person can build, and each person can hear the other, but neither can see the other person or the other blocks. On each of the two work spaces, place one of the two sets of blocks you created above. The two work spaces should be as close to identical as is practical to ensure that any structure built on one work space can be built on the other work space. One possible arrangement is to place a large vertical separator in the middle of a table such that it separates the two ends of the table, then have each person sit on an opposite end of the table facing each other. The separator must be large enough to prevent the two people from seeing each other (and stable enough not to fall over). This allows onlookers to stand on the sides of the table and easily see both of the builders and constructions.

Select two test subjects and sit them down in front of their respective work spaces and blocks. Designate one as the master and the other as the apprentice (or one as the speaker and the other as the listener, if you prefer). The master builds a structure with his blocks (any structure of his choice containing a least a designated minimum number of blocks), and instructs the apprentice as to how to build exactly the same structure. The apprentice is not allowed to ask questions or provide any other kind of feedback to the master. The apprentice must build the same structure as the master based solely on the verbal description given by the master. The master, in turn, must provide his verbal description of his structure without any knowledge of what the apprentice is doing.

It turns out that this is surprisingly difficult to do well. In most cases, the apprentice's construction will diverge from the master's construction after only a few blocks, and by the time the master is placing his tenth block and describing it to the apprentice, the apprentice is completely lost and unable to place the block in any location remotely resembling the master's description. At some point along the way, the apprentice becomes aware that something is wrong, but can't do anything about it - and the master, blithely unaware of the apprentice's difficulty, inexorably continues to build and describe something that the apprentice no longer understands. An alert master may be able to tell from the reactions of the viewers that the apprentice is having a problem, but without any feedback will not be able to know quite what the problem is, so will be unable to correct it and will have to forge on.

I watched a number of other pairs who attempted this exercise fail to build structures of any significant size, and I was determined to do a better job. When it was my turn to be master, I very carefully described each block I selected, how I oriented it, and how I placed it next to other previously placed blocks. I was describing everything at an excruciating level of detail, far more than I would in normal circumstances. And, as I learned at the end of my turn, for a while this was working quite well, and my apprentice was able to construct significantly more than had anyone else.

But there was a snag: at one point I selected a small triangular block, which happened to be a right equilateral triangle. When placing it, I instructed my apprentice to orient the triangle by pointing the corner which was a right angle in a certain direction. Unfortunately, math was not a strong point for my apprentice, and for a while she was unable to recall what a right angle was. As a consequence, she did not understand what I meant concerning the orientation of the block, but because I was unaware of her problem and continued my building and describing, she had to guess about how to place the block, and she guessed wrong. Of course, errors cascaded from there because further steps which relied on the placement of that block could not be done properly.

Had she been able to ask me a couple of simple questions, I could easily and quickly have clarified for her what I meant. Without any feedback at all, the failure to understand one small detail meant the game was over.

You can try the same exercise, but this time let the apprentice ask questions. It is amazing how much difference it makes. Without feedback, almost nobody can succeed on this task, and when they do it requires an extraordinary amount of redundancy in the master's descriptions. With verbal feedback, most people can do quite a good job, and much faster than without feedback, since the master has the liberty to be much sloppier in his initial description, confident in the knowledge that the apprentice will ask for more details on any part he does not understand. With visual feedback, virtually every apprentice can quite speedily reproduce the master's design. The speed and accuracy of communication improves dramatically as the feedback bandwidth increases.

I highly recommend this exercise for teaching the value of precise communication (especially when there is no or little feedback available), the power of a high-bandwidth feedback channel (especially when there is any complexity involved) and the recognition that some tasks are far harder than they appear (especially when you have not directly attempted the task yourself).

No comments: