A predictive model for self-motivated decision-making behavior

Traditional models of decision-making make simplistic assumptions about the motivations of agents and consistently fail to predict realistic behavior patterns. Since most modern economic and social theories heavily use computational modeling and simulation to quantify the effects of various policy options, the failure of classical decision theories to adequately capture realistic agential behavior is a considerable problem. We present a computational model of decision-making that adapts the mathematical machinery of reinforcement learning to take the intrinsic motivations of agents into account. Our model is uniquely principled in theories of embodied cognition and bounded rationality and predicts previously unexplained 'irrational' behavior of human subjects on real experimental tasks.