As discussed earlier (Feedback, Prices, and Sullen Spouses), feedback is an extremely valuable tool that is used extensively by engineers for all manner of applications to ensure accuracy. However, there’s a catch: feedback must be provided quickly, or it can provide the opposite of the desired result.
Let’s assume that you’re driving a car in the normal fashion, using its speedometer as feedback so that you can properly control your speed. Let’s also assume the speedometer is working properly, except that it’s very sluggish; i.e. there’s a time delay of two or three seconds until your car’s speed is displayed on the speedometer.
Let’s now assume that you’re zipping along and you notice that your speed is dropping, and you want to speed up. You press on the pedal and you feel the car surge a bit, but no, the speedometer (because of its delay) says that you haven’t picked up any speed (even though you have), so you push on the accelerator even harder. Therefore, because of the speedometer’s delay, you pushed the accelerator twice, whereas once would have been enough.
At the second push of the accelerometer, the speedometer finally registers the speed following your first push, indicating you are at the correct speed and everything is fine, but then the speedometer registers the speed from the second push, and you realize with horror that you are well over the speed limit. You immediately press on the brakes and you think you feel the car slow down, but no, the speedometer (because it hasn’t yet registered the slower speed) says you are still moving much too fast, so you press on the brakes some more. Therefore, because of the speedometer’s delay, you pressed the brakes twice, whereas once would have been enough.
After the second press of the brakes, the speedometer belatedly registers the speed following the first press, indicating that you are at the proper speed, but then the speedometer registers the second press, and you realize you have slowed down much too much, so you stomp on the accelerator…
…and your car continues to jerkily speed up and slow down, like a teenager first learning to drive an auto with a stick shift, until the police officer pulls you over and charges you with reckless driving.
Therefore, although the use of feedback achieves superior performance, feedback must be provided quickly. If there is too large of a delay then feedback will be interpreted incorrectly, which can cause a system to become wildly unstable and possibly even be damaged.
Feedback must be quick
In the earlier post we talked about how a lack of feedback from a sullen spouse could contribute to a poor relationship. In a similar manner, feedback that is supplied after a long delay can make things worse, rather than better:
He: Please pass the salt.
She: No. I don’t like the way you spoke to me.
He: What? When?
She: The last time we were at this restaurant.
He (becoming angry): That was two months ago! What does that have to do with tonight?
The above is an example of how feedback, if it had been delivered quickly, could have served a constructive purpose. However, delayed feedback loses its proper context, and instead of being corrective, easily becomes destructive.