We view Levine et al.’s insightful analysis as the beginning of a monumental project: understanding morality through the lens of resource-rational analysis. Resource-rational analysis (Lieder & Griffiths, Reference Lieder and Griffiths2020) is an upward spiral wherein researchers iteratively refine hypotheses about a cognitive system’s function and constraints. Resource-rational contractualism explains an impressive range of phenomena, but its authors also acknowledged several phenomena that it cannot accommodate. Our commentary outlines how future cycles of resource-rational analysis might explain additional aspects of morality by considering alternative hypotheses about morality’s function(s).

Resource-rational analysis distinguishes the ultimate function of morality (computational level) from decision mechanisms approximating it (algorithmic level; see Figure 1). The computational-level theory usually draws on normative theories. For morality, the natural choice is normative ethics. There are three major theories of normative ethics: consequentialism, deontology, and contractualism (Parfit, Reference Parfit2011). Levine et al. chose contractualism.

Figure 1.

The ambiguous relationship between the mechanisms of morality and their ultimate function.

Their argument for contractualism included this thought experiment: Would a rich baker agree to give bread away for free to a poor customer? According to contractualism, the baker should receive at least something in exchange for the bread. We thought people might be more prosocial, especially outside market settings. To test this, we conducted a study (N= 100) with two conditions: participants in the “market” condition read the example above; participants in the “street” condition read about a rich bakery owner encountering a poor person on the street. Most participants in the “street” condition said the agreement would be to give away the bread for free (84%, 95% confidence interval (CI) [71%,93%]). In the “market” condition, participants were divided between the consequentialist intuition that the poor customer would pay nothing (43%, 95% CI [29%, 58%]) and the contractualist intuition that he would pay very little (33%, 95% CI [20%, 48%]) or market price (25%, 95% CI [14%, 39%]). Moreover, most participants said it would be most morally right to give away the bread for free (“market”: 78%, 95% CI [63%, 88%]; “street”: 96%, 95% CI [85%, 99%]). These findings suggest many participants do not follow contractualist principles – even in market contexts. Furthermore, participants’ intuitions differed by context: participants in the “street” condition were significantly less contractualist than those in the “market” condition. More participants in the “street” condition reported that the bread should (χ² (1,100)=6.03, p=.014) and would (χ²(1,100)=16.89, p<.001) be given away for free.

Participants’ intuitions in the “street” condition illustrate that resource-rational contractualism cannot explain certain phenomena, such as self-sacrificial prosocial behavior, as Levine et al. acknowledged. Future iterations of resource-rational analysis of morality may be able to explain those phenomena by refining the computational-level theory (see Figure 1).

Levine et al. showed that, in other situations, many moral mechanisms produce judgments/decisions approximating the prescriptions of contractualism. However, in most situations, many of these mechanisms also approximate the prescriptions of other ethical theories quite well. Various strategies to negotiate agreements could have been evolutionarily selected for their consequences. Although bargaining sometimes falls short of maximizing welfare (Le Pargneux et al., Reference Le Pargneux, Zeitoun, Konstantinidis and Chater2023; Levine et al., Reference Levine, Kleiman-Weiner, Schulz, Tenenbaum and Cushman2020, Reference Levine, Kleiman-Weiner, Chater, Cushman and Tenenbaum2024), its average real-world consequences might be better than those of alternative strategies. A negotiator’s proximal goal to reach an agreement could thus be an approximation to the ultimate goal of maximizing collective welfare.

Therefore, the ultimate function of morality could be a version of consequentialism (cf., Sunstein, Reference Sunstein2005). Ord (Reference Ord2009) proposed that rule-based, outcome-based, and virtue-based decision mechanisms can be derived from global consequentialism – the idea that people should employ whichever mechanism yields the best consequences. Parfit (Reference Parfit2011) argued that, under ideal conditions, the optimal application of contractualism yields the same moral principles as the optimal application of global consequentialism to rules. From this, we conclude that agreement-based approaches (e.g., virtual bargaining) can be used to approximate global consequentialism in some situations (e.g., when rule-based strategies are optimal). We propose that most mechanisms of moral judgment and decision-making can be understood as resource-rational approximations to global consequentialism. We call this approach resource-rational global consequentialism.

Resource-rational global consequentialism is appealing for three reasons. First, it might be able to explain more phenomena than resource-rational contractualism. Levine et al. acknowledge that their theory is limited to interdependent choice and cannot explain prosocial behavior toward powerless individuals, condemnation of consensual transgressions, universal rights, and many virtue-based judgments. According to resource-rational global consequentialism, these phenomena may arise from mechanisms generally producing good consequences in the real world. Second, empirical research suggests people learn to rely on the decision mechanism (e.g., rules or cost-benefit reasoning) that produced good consequences in similar past situations (Maier, Cheung, & Lieder, Reference Maier, Cheung and Lieder2025). Third, evolution selected moralities for their consequences on reproduction. Reproduction is not well-being, but it is a consequence nonetheless. Hence, the function of commonsense morality might be best understood as nonutilitarian global consequentialism.

Nevertheless, there probably are situations where human behavior deviates from resource-rational global consequentialism. In line with this, Levine and colleagues have provided empirical evidence for situations where people’s behavior appears most in line with contractualism (e.g., Le Pargneux et al., Reference Le Pargneux, Zeitoun, Konstantinidis and Chater2023; Levine et al., Reference Levine, Kleiman-Weiner, Schulz, Tenenbaum and Cushman2020, Reference Levine, Kleiman-Weiner, Chater, Cushman and Tenenbaum2024). Overall, people seem contractualist in some situations but consequentialist or deontologist in others (Caviola et al., Reference Caviola, Kahane, Everett, Teperman, Savulescu and Faber2021; Le Pargneux, Roberts-Gaal, & Cushman, Reference Le Pargneux, Roberts-Gaal and Cushman2025). This suggests that morality might not have one ultimate purpose, but multiple purposes. How many people rely on each of the corresponding mental functions might vary across situations. To examine this, methods for understanding individual differences (e.g., mixture models) should be used to classify who pursues which moral objectives when. Instead of switching between moral objectives, people might pursue multiple moral objectives simultaneously, as recommended by the expected choiceworthiness theory of decision-making under moral uncertainty (MacAskill & Ord, Reference MacAskill and Ord2020). If so, our moral mechanisms might be resource-rational for some tradeoff between multiple moral theories.

Whether morality’s ultimate function is consequentialist, contractualist, or something else remains unclear. To resolve this ambiguity, future work should mathematically formalize the objective functions for each potential purpose of morality and use automatic strategy discovery (Krueger et al., Reference Krueger, Callaway, Gul, Griffiths and Lieder2024) to derive their predictions about people’s strategies in different moral contexts.