In the mathematical models for network formation by Bala and Goyal(2000), it is shown that a star network is the strict Nash equilibrium. However, the result of the experiments in a laboratory using human subjects by Berninghaus et al.(2007) basing on the model of Bala and Goyal indicates that players reach a strict Nash equilibrium and deviate it. In this paper, an agent-based simulation model in which artificial adaptive agents have mechanisms of decision making and learning based on nueral networks and genetic algorithms is constructed, and we provide that one of the reason of the deviation from the strict Nash equilibrium in the experiments by Berninghaus et al. is that the players have mechanisms of decision making by trial and error and with a long-term view.