Leon Assad gave a talk “What do you mean by ‘Likely’? How ambiguous testimony leads to polarization and wrong beliefs”.
Group deliberation has two aims: to lead to consensus, and improve beliefs. But deliberation can fail: we may polarize, or converge on wrong beliefs. Increasingly, philosophers put forward mechanisms that would lead to such failure even among rational individuals—or at least boundedly rational. This paper contributes a new mechanism to that literature: ambiguous testimony. Verbal expressions of uncertainty (”likely,” ”unlikely”) are compatible with a range of underlying credences, and decades of psychological research show that listeners resolve this ambiguity in different ways. My ”likely” is not your ”likely,” and this mismatch can produce misunderstandings between speakers and listeners. Using a new agent-based model of ambiguous testimony, we show that these misunder- standings alone are sufficient to generate both polarization and entirely wrong consensus at the group level. While we do not claim that our agents are perfectly rational, we argue that a realistic depar- ture from ideal deliberation—the use of ambiguous expressions—suffices to produce both outcomes among otherwise rational agents.