Peter Railton

The past 15 years occasioned an extraordinary blossoming of research into the cognitive and affective mechanisms that support moral judgment and behavior. This growth in our understanding of moral mechanisms overshadowed a crucial and complementary question, however: How are they learned? As this special issue of the journal Cognition attests, a new crop of research into moral learning has now firmly taken root. This new literature draws on recent advances in formal methods developed in other domains, such as Bayesian inference, reinforcement learning and other machine learning techniques. Meanwhile, it also demonstrates how learning and deciding in a social domain—and especially in the moral domain—sometimes involves specialized cognitive systems. We review the contributions to this special issue and situate them within the broader contemporary literature. Our review focuses on how we learn moral values and moral rules, how we learn about personal moral character and relationships, and the philosophical implications of these emerging models.

David Hume famously criticized rationalist theories of practical reason, arguing that reason alone is incapable of yielding action, and that some passionate element must be supplied. Contemporary theories of Humean inspiration develop a causal-explanatory model of action in terms of the joint operation of two distinct mental states: beliefs and desires, one inert and representational, the other dynamic. Such neo-Humean theories claim that since desires, unlike beliefs, are not subject to direct rational evaluation, an act can be said to be rational only in the sense that it is instrumental to realizing the agent’s desires. The historical Hume appears to have embraced a “sceptical solution” involving a more dynamic conception of belief, while admitting a default sense in which both beliefs and actions can be deemed reasonable or unreasonable.

At least since Aristotle, practical skill has been thought to be a possible model for individual ethical development and action. Jonathan Birch’s ambitious proposal is that practical skill and tool-use might also have played a central role in the historical emergence and evolution of our very capacity for normative guidance. Birch argues that human acquisition of motor skill, for example in making and using tools, involves formation of an internal standard of correct performance, which serves as a basis for normative guidance in skilled thought and action, and in the social transfer of skills. I suggest that evaluativemodeling, guidance, and learning play a more basic role in motor skill than standards of correctness as such-indeed, such standards can provide effective normative guidance thanks to being embedded within evaluative modeling and guidance. This picture better fits the evidence Birch cites of the flexibility, adaptability, and creativity of skills, and can support a generalized version of Birch’s ‘skill hypothesis’.

Cushman seeks to explain rationalization in terms of fundamental mental processes, and he hypotheses a selected-for function: information exchange between “rational” and “non-rational” processes in the brain. While this is plausible, his account overlooks the importance – and information value – of rationalizing the emotions of ourselves and others. Incorporating such rationalization would help explain the effectiveness of rationalization and its connection with valuation, as well as raise a challenge to his way of bifurcating “rational” and “non-rational” processes.

Value and uncertainty are the critical components of decision and action. To think of the affective system as at the core of action is to draw attention to the role of affect in representing and combining these two dimensions, and orchestrating a wide range of mental capacities—attention, perception, memory, inference, motivation, and monitoring—in light of these evaluative representations. The commentators have helpfully enriched our appreciation of the various ways in which affect can contribute to the attunement, cuing, motivation, control, and ongoing assessment of decision and action.

Ruth Chang has argued convincingly that we must recognize that some choices will not involve strict, univocal comparison of options. How, then, can such choices be made well? Chang suggests that commitment is a fundamental way of ‘putting one's very self’ behind a normative consideration, thereby ‘endow[ing] that consideration with the normativity of a reason’. This view challenges what Chang deems to be three dogmas of normativity, and the current comment critically assesses the relation of her view to the first and third of these dogmas. I suggest some alternative ways of thinking about these dogmas, which enable us to see how her work might uncover a fourth dogma, and which can, I believe, lend strength to a position very close to Chang's.

In Susanna Siegel’s compelling presentation of the case for the rationality of perception, a “significant part of the constructive defense” is played by the idea that there are “inferential routes to perceptual experience” (Siegel 2017, p. 94). Inferences, after all, are epistemically evaluable and bear on the rational standing of their conclusions. She argues that an obstacle to accepting this idea is a “Reckoning Model” of inference, and shows by example that we recognize as inferences various familiar kinds of responses to information that do not fit this model. She offers a more general approach to the nature of inference that fits these examples and accommodates inferential routes to perceptual experience. I argue that Siegel needs to say more about the mental processes involved in such inferrings, and how it can be more than merely associative and yet still distinct from Reckoning. Fortunately, a psychologically‐ and conceptually‐grounded distinction between model‐free vs. model‐based learning and guidance processes can provide a characterization that plays the role Siegel needs.

Although the ‘naturalistic fallacy’ and the is/ought distinction have often been invoked as definitive grounds for rejecting any attempt to bring evolutionary thought to bear on ethics, they are better interpreted as warnings than as absolute barriers. Our moral concepts themselves -- e.g. the principle that ‘ought implies can’ -- require us to ask whether human psychology is capable of impartial empathetic thought and motivation characteristic of normative systems that could count as moral. As the essay by Flack and de Waal shows, evolutionary theory and evidence can help us answer the question whether the psychological ‘building blocks’ needed for morality are indeed likely to be present in a given species, including Homo sapiens. It is important, however, not to think that a positive answer to this question commits us to attributing identifiable moral concepts to actual members