Bounded Morality: The Psychophysics of Morality

Moral decisions are often cast as a conflict between utilitarianism (maximizing good) and deontology (following rules). We propose a unitary psychophysical model where moral decisions operate like other perceptual systems and track the subjective (information-theoretic) certainty that outcomes achieve the greater good. Due to psychophysical noise, decisions are governed by Weber’s law, with utilitarian and deontological responses as high- vs. low-uncertainty limit cases. Participants rated the acceptability of sacrificing fewer ‘victims’ to save more ‘beneficiaries.’ Consistent with this model, acceptability tracked the ratio of beneficiaries to victims (beneficiary:victim Weber ratios), but not utility (net survivors) or harm (victims). With decreased uncertainty (using decisions concerning eco- nomic goods that had clearer values than human lives), responses became more ‘utilitarian’; with increased attention to victims (via emotional manipulations), responses became more ‘deontological’, but always tracked the beneficiary:victim Weber ratios. Frontier large language models from three different jurisdictions (though not smaller models) partially reproduced human behavior, but used qualitatively different heuristics from humans and generally prioritized harm or utility over beneficiary:victim ratios. Beyond moral decision making, the same model provides the first mechanistic and parameter-free explanation of loss-aversion. People, but not machines, thus seek the greater good, but do so under perceptual noise.