Rational Learners and Moral Rules

Abstract

People draw subtle distinctions in the normative domain. But it remains unclear exactly what gives rise to such distinctions. On one prominent approach, emotion systems trigger non‐utilitarian judgments. The main alternative, inspired by Chomskyan linguistics, suggests that moral distinctions derive from an innate moral grammar. In this article, we draw on Bayesian learning theory to develop a rational learning account. We argue that the ‘size principle’, which is implicated in word learning, can also explain how children would use scant and equivocal evidence to interpret candidate rules as applying more narrowly than utilitarian rules.

Learning a commonsense moral theory

“…Other extensions could integrate our model with recent work which has shown how deontological principles (of the form "do not X" or "do not intend X" regardless of the consequences) could be learned (Nichols et al, 2016;Ayars & Nichols, 2017) or emerge from choice algorithms (Crockett, 2013;Cushman, 2013). Learners are also expected to learn how different "base" moral goods and evils contribute to the welfare of individuals or even what counts as moral.…”

“…So then how do we learn so much from so little? A hierarchical Bayesian approach has had much recent success in explaining how abstract knowledge can guide learning and inference from sparse data as well as how that abstract knowledge itself can be acquired (Griffiths, Chater, Kemp, Perfors, & Tenenbaum, 2010;Tenenbaum et al, 2011;Xu & Tenenbaum, 2007;Perfors, Tenenbaum, & Regier, 2011;Ayars & Nichols, 2017;Nichols, Kumar, Lopez, Ayars, & Chan, 2016), and fits naturally with the idea that learners are trying to estimate a set of weighted moral principles. By inferring the underlying weighting of principles that dictate how the utility of different agents are composed, a Bayesian learner can make generalizable predictions in new situations that involve different players, different numbers of players, different choices, etc (Heider, 1958;Malle, Moses, & Baldwin, 2001;C.…”

Troubles with Bayesianism: An introduction to the psychological immune system

“…Bayesianism's appeal is not hard to see: it allows for the possibility of a single mental mechanism—Bayesian updating—to unify mental processes as diverse as word learning (Xu & Tenenbaum, ), belief updating (Bennett, ), conditional reasoning (Oaksford & Chater, ), the development of moral judgments (Nichols, Kumar, Lopez, Ayars & Chan, ), domain‐general reasoning (Vul & Pashler, ), predictive coding (Clark, ; Hohwy, ), compositionality in the Language of Thought (Goodman et al, ), causal reasoning (Gweon & Schulz, ), and reinforcement learning (Vlassis, Ghavamzadeh, Mannor & Poupart, ) to name just a few recent domains of interesting work falling under the Bayesian banner. The sheer generality of Bayesianism allows a scope unmatched by most theories, save for discredited ones like Radical Behaviorism (Skinner, ) and Associationism (Mandelbaum, ).…”

Moral Judgment and Deontology: Empirical Developments

“…For example, some researchers believe that, while some moral intuitions arise from deontological principles, many others arise from more utilitarian calculations (Cushman et al ). Relatedly, we haven't addressed whether moral judgments resulting from applications of non‐consequentialist rules are justified (Singer ; Greene ; Sinhababu vs. Berker ; Mikhail ; Nichols et al ms). However, illuminating non‐consequentialist influences on moral judgment can undoubtedly aid such further investigations.…”