Levine, Chater, Tenenbaum, and Cushman propose resource-rational contractualism as a unified model of moral cognition. In this model, people engage in moral reasoning as if bargaining to reach mutually beneficial agreements, using cognitive shortcuts – such as mental simulation, welfare-based standards, or action-based rules – to approximate ideal negotiation outcomes. The authors argue that their model can better explain moral flexibility (how our moral rules and standards get created, updated, and overridden). They use COVID-19 to illustrate how communities initially adopted broad standards (e.g., strict lockdown) before developing nuanced exceptions through renegotiation (e.g., permitting social interactions with a small group). This narrative suggests moral change follows a predictable path: when circumstances change, individuals renegotiate toward new, mutually beneficial arrangements.

We argue that this model cannot capture the widespread persistence of moral disagreements when new situations arise. Moral flexibility does not always lead to a new consensus or improved mutual benefit; instead, it often leads to people diverging in their moral views. The COVID-19 pandemic precisely illustrates this limitation. Rather than converging via renegotiation, people polarized into different moral camps. Mask-wearing, vaccination, and social distancing became ideological battlegrounds where disagreement hardened instead of resolving. These divided views directly contradict the model’s core premise that moral cognition’s ultimate goal is to achieve mutually beneficial agreements. Here, we describe three factors that pose problems for the resource-rational model.

First, social identity and group dynamics can significantly influence one’s moral cognition (Chalik & Rhodes, Reference Chalik, Rhodes and Benson2020), including whether and how individuals change their moral values. For example, partisanship, as a crucial social identity, has impacted how people think and behave (Iyengar & Krupenkin, Reference Iyengar and Krupenkin2018; Mason, Reference Mason2018). Resource-rational contractualism assumes individuals primarily seek mutually beneficial arrangements through (actual or virtual) negotiations. However, research demonstrated that compliance with health guidelines during COVID-19 strongly correlated with political affiliation and social identity rather than with rational assessments of mutual benefits (Gollwitzer et al., Reference Gollwitzer, Martel, Brady, Pärnamets, Freedman, Knowles and Van Bavel2020; Kaplan et al., Reference Kaplan, Vaccaro, Henning and Christov-Moore2023; Young et al., Reference Young, Rasheed, Bleakley and Langbaum2022). Conservative communities might reject mask mandates not because they have calculated different cost-benefit ratios, but because compliance signals alignment with out-group values. Similarly, liberal communities might embrace restrictions partly to distinguish themselves from conservative groups (e.g., Kaplan et al., Reference Kaplan, Vaccaro, Henning and Christov-Moore2023). Adhering to partisanship is not synonymous with “renegotiating.” This is because renegotiation approximations are often absent: within the partisan group, “agreements” already exist; among different partisan groups, individuals merely adhere to or reinforce preexisting beliefs to reaffirm their social identity, rather than (re)negotiating. This identity-based cognitive process introduces values and commitments that resist integration into utility functions or bargaining scenarios.

Second, human reasoning is prone to biases such as confirmation bias (Klayman, Reference Klayman1995; Nickerson, Reference Nickerson1998). Preexisting beliefs often interfere with how evidence is weighted; confirmation bias tends to lead individuals to seek support for preexisting beliefs while ignoring evidence that challenges preexisting beliefs. As a result, confirmation bias creates asymmetric updating that prevents convergence. During COVID-19, individuals selectively attended to data confirming their prior positions – skeptics focused on economic costs and liberty restrictions, while supporters emphasized death rates and hospital capacity (Modgil et al., Reference Modgil, Singh, Gupta and Dennehy2024). For example, Malthouse (Reference Malthouse2023) demonstrates that in the case of vaccines, people made systematic errors when encountering evidence contradictory to their prior beliefs. Participants correctly evaluated evidence about vaccine efficacy 60.2% of the time. However, those with pro-vaccine beliefs were more accurate when evaluating evidence supporting vaccine effectiveness, whereas those with anti-vaccine beliefs were more accurate when evaluating evidence suggesting ineffectiveness (Malthouse, Reference Malthouse2023, pp. 526–527). This indicates that preexisting beliefs influence how people process and interpret evidence, leading them to reinforce their views despite possible mutual agreements. Such biased processing prevents negotiation approximations, as negotiation tends to require an ultimate convergence on shared facts.

Third, emotions and nondeliberative processes powerfully shape moral judgment, often bypassing deliberative processes (Haidt, Reference Haidt2001). During COVID-19, fear of disease, anger at authority, or disgust at perceived weakness can trigger immediate moral judgments without engaging the bargaining mechanisms described by the model. Fear might prompt support for restrictions regardless of actual risk levels, while anger at government overreach might fuel opposition independent of health consequences. For instance, someone opposing the lifting of lockdown may do so not because lockdown better promotes mutual benefits, but because they fear infection irrespective of new evidence, thereby refusing to comply with new rules (Harper et al., Reference Harper, Satchell, Fido and Latzman2021). These emotional drivers can distort utility calculations by overweighting certain emotional impacts that the unknown virus brought.

While Levine and colleagues acknowledge that people may use “the entire set of cognitive mechanisms” to evaluate norms, with different mechanisms producing different judgments, they crucially assume these ultimately aim for mutual benefit – an assumption we challenge. The three psychological factors aforementioned share a crucial feature: individually, none aim for mutual benefits, and each seems at odds with resource-rational decision making. Identity-based reasoning prioritizes group cohesion over optimal outcomes; cognitive biases maintain or reinforce existing beliefs regardless of consequences; and emotions respond to immediate stimuli rather than negotiated agreements. This suggests a fundamental limitation in the model’s scope: resource-rational contractualism may only explain moral convergence in low-stakes, emotionally neutral contexts where most individuals share the same social identity. This is highly limiting because in real-world scenarios, even within the same small group, individuals often hold different social identities, cognitive biases, and emotional responses (as demonstrated in COVID-19 cases).

Levine and colleagues might argue that disagreements are temporary. However, the persistence of moral conflicts across decades (e.g., abortion, gun control) or centuries (e.g., religious differences) challenges this interpretation. These are not failed negotiations but stable states where disagreement serves psychological and social functions beyond mutual benefits. When renegotiation fails due to these factors, we don’t simply maintain the status quo. Instead, we often see escalating polarization, the formation of conflicting moral subgroups, and the breakdown of shared moral frameworks – phenomena the model cannot adequately explain.