A report on the experiments carried out with our robot in the Emotirob project is given in this paper, in which we show how we build emotion and personality in the robot. With children, the results of interaction with the robot are quite satisfactory in a short-term experiment. However, it was noted that during long-term interaction between the children and the robot, the relationship changes as a kind of lassitude sets up. Thus, the question addressed here is, how can we make a robot acceptable for long-term interaction? We propose to explain why empathy is a part of the solution and what the key points are for artificial intelligence to solve this new problem.