Rational vigilance of intentions and incentives guides learning from advice

People vigilantly evaluate advice: they seek to learn from competent and well-intentioned informants, while ignoring incompetent or manipulative ones. Yet the mechanisms that support vigilance of informants' motivations remain poorly-understood. For instance, how do people combine informants' incentives and intentions when evaluating testimony---through rational inference, or simple heuristics? To answer such questions, here we develop a rational model of vigilance and test its predictions in two pre-registered online experiments with participants in the U.S (Ns = 626, 555). Across financial, real estate, and medical decision scenarios, participants' inferences closely tracked our model’s predictions: participants discounted advice when selfish speakers benefited from manipulating them, and this discounting was attenuated by perceived benevolence. When pay-offs were shared, incentives increased trust, as our rational model---but not simple heuristics---predict. Our rational analysis thus establishes a formal framework for vigilance of motivations that also sheds light on phenomena from persuasion to polarization.