We present a model of learning in which agents learn from errors. If an action turns out to be an error, the agent … an acceptable solution is asymptotically reached. Moreover, one can take advantage of big errors for a faster learning. …