Behavioral experiments have shown that if you reward an action consistently, every time, and then stop rewarding it, animals will learn the behavior, repeat it, and then stop shortly after the rewards do.

However, if you reward an action inconsistently, and gradually decrease the frequency of rewards, many animals will continue long after the rewards have stopped.

I have neither a citation nor an excellent memory of the methodology of the experiment, but IIRC it was done with apes, levers, poker chips and grapes sometime in the late 90's.