We cannot agree more with Levine et al. that human moral judgment is best described by a contractualist computational structure (André & Baumard, Reference André and Baumard2011; Baumard et al., Reference Baumard, André and Sperber2013; Baumard, Reference Baumard2016). In a recent paper, we have proposed an explicit evolutionary mechanism to explain why natural selection may have favored moral judgments with exactly this structure, and how the contractualist logic helps unify and explain the otherwise disparate patterns of human moral judgment (André et al., Reference André, Debove, Fitouchi and Baumard2023). We agree that contractualism has been understudied, that it better explains the patterns of moral cognition than utilitarianism or deontology, and that the key challenge for moral psychology to move forward is identifying the cognitive mechanisms that implement contractualist reasoning.

Where our accounts diverge is in the nature of the proximate cognitive mechanisms that generate contractualist judgments (André et al., Reference André, Debove, Fitouchi and Baumard2023). Levine et al. propose many distinct mechanisms, but the only one they consider to reach the contractualist ideal – which their other mechanisms simply approximate – is the mental representation of bargaining itself (either virtual or actual). People would determine what they owe to each other by either bargaining over the question at hand or by imagining what they would agree to if they were to bargain. Levine et al.’s argument is that if the function of moral cognition is to find social behaviors to which everyone would agree after a negotiation, then the proximate mechanism to realize this function is to mentally represent the bargaining process itself.

We challenge this assumption because it mischaracterizes the relationship between the ultimate function and proximate design of cognitive systems. Proximate mechanisms – whether physiological or cognitive – do not need to represent or recapitulate the logic that makes their function adaptive. Take vision. The ultimate cause of the evolution of the visual system’s structure lies in natural selection favoring organisms able to exploit physical regularities, such as optics and projective geometry, to reconstruct depth and shape in a way that allows them to navigate the world adaptively. But the proximate mechanisms implemented by the eye and brain do not mentally represent the physical regularities of optics and projective geometry; they simply exploit them by reconstructing depth and shape based on the constraints imposed by these physical regularities.

The same, we suggest, holds for moral cognition. Moral judgments track outcomes that would be selected through a balanced bargaining process – those that divide the surplus of cooperative interactions equally or more generally maximize the Nash product of partner payoffs. But to generate such judgments, the mind does not need to simulate a bargaining process (e.g., an alternate bargaining à la Rubinstein). Instead, it needs to solve an optimization problem: to (i) explore the space of possible social arrangements, (ii) estimate the marginal impact of each outcome on the utility of the social partners involved, and (iii) identify the one that maximizes the Nash product of these utilities (André et al., Reference André, Debove, Fitouchi and Baumard2023).

The work of some of the authors themselves, in fact, shows that people are readily able to produce precise estimates of the utilities that other people would derive from if a certain course of action was taken (e.g., Jara-Ettinger et al., Reference Jara-Ettinger, Gweon, Schulz and Tenenbaum2016). The task that remains to produce a contractualist moral judgment is to select the course of action that maximizes the product of these utilities. This is a demanding optimization task, but it is not qualitatively different from many other optimization problems the mind routinely solves – planning a route, preparing a meal, or assembling a piece of furniture. In each case, we evaluate multiple options, estimate their effects on intermediate goals (e.g., saving time, reducing effort), and select the one that best meets a given criterion. Moral judgment is often more complex quantitatively because it involves multiple individuals rather than just one – but not qualitatively.

This question of whether moral cognition proceeds by maximizing a quantity or by simulating a bargaining mirrors the old debate between the “simulation theory” and the “theory-theory” of mind reading (e.g., Apperly, Reference Apperly2008). Simulation theory maintained that we make sense of others’ mind and behavior by simulating what we would ourselves do or think in their situation. The “theory-theory,” by contrast, maintained that mindreading computations are made of a set of abstract concepts (belief, desire, etc.) and governing principles (e.g., expected utility maximization), without the need for simulation. By analogy with simulation theory, Levine et al. argue that moral judgment functions by simulating what people would agree to if they were to bargain. We suggest, by contrast, that the very same outputs can be produced by an abstract “theory of justice” – a cognitive model that takes as input the utilities of individuals and evaluates the moral value of actions by directly assessing their impact on the Nash product of these utilities.

This matters. If moral judgment required simulating bargaining, we should expect agents to routinely fall back on approximative, resource-rational heuristics, given the high cognitive cost. But if, as we argue, moral judgment relies on a simpler form of optimization, then heuristic shortcuts are needed less often. Agents can often compute the fair outcome directly by maximizing the Nash-product, unless the situation is highly complex or informationally impoverished. In such cases, default rules and rigid moral habits may arise, but they are context-specific workarounds, not fundamental features of moral cognition.

In sum, we welcome Levine et al.’s contribution for bringing renewed attention to the role of contractualist principles in moral judgment. But just as the visual system can reliably reconstruct depth and shape without representing the laws of optics, moral cognition can reach ideal contractualist targets without representing or simulating bargaining processes. We hope that future research will help us move forward in empirically discriminating between these alternative computational possibilities.