One of the things that's rarely mentioned in these training attempts: Start by rewarding consistently, then become inconsistent, only rewarding, say, one time in 3. Or 4.<p>Why?<p>If you consistently reward every instance of good behavior, then when you stop, the good behavior will tail off remarkably quickly. You are caught in the trap of having to reward every good instance.<p>On the other hand, suppose the trainee has got used to being rewarded at random for good behavior. Then you stop entirely.<p>The first time the trainee says "Yes? Oh, no. Oh well, maybe next time." The next time the trainee says "Yes? Oh, no. Oh well, maybe next time."<p>And on it goes. Randomly reinforced good behavior persists for a very long time after the rewards stop. Very occasional rewards makes it even longer.