Most AI ethics writing works from three lenses: utilitarian reasoning, which judges a system by its net outcomes; deontological reasoning, which judges it by duties and rights regardless of outcome; and virtue ethics, which asks what an organisation of good character would do. All three are useful. All three also share a starting assumption that can quietly narrow what a finding actually sees: that the unit of moral concern is the individual, considered more or less on their own.
A family of relational traditions starts somewhere else.
Ubuntu, the Southern African tradition holding that a person is a person through other persons, asks what a system does to a community’s capacity to trust itself, not only whether any one person consented to it.
Confucian role ethics and Buddhist accounts of interdependence make a related point from a different direction: a model is not a discrete artefact sitting apart from the world, it sits inside a network of the people who labelled its training data, the people who use it, and the people who live with what it decides, and its effects on that network do not reduce cleanly to individual harms.
Dharmic traditions add that what is right can be role- and context-sensitive rather than fixed, which is a genuinely different question than asking whether a rule was universally applied.
None of this replaces utilitarian, deontological, or virtue reasoning. It sits alongside them, and it tends to surface a different category of finding: not “did this harm an identifiable person” but “what did this do to the trust a community had in each other, or in the institution deploying it,” which is a harm a purely individual-outcomes lens can miss entirely, even when it’s doing careful, honest work.
We read against all of these traditions for the same reason we read against multiple regulatory frameworks rather than one: a finding built on a single lens tells you it is right within that lens, which is a narrower claim than most people reading it assume it to be…