Across human history, civilizations have attempted to persist beyond biological death — in stone monuments, written law, oral tradition, and inherited myth. What is new is not the impulse but the substrate: the training of machine learning systems on a person's digital output, producing a model that approximates their voice, their reasoning style, their characteristic concerns. When this practice scales from the individual to the collective, something structurally different emerges. Entire communities, institutions, and generations can now — in principle — construct synthetic continuations of themselves that future populations interact with as living interlocutors. The "ghost you train of yourself" at collective scale is not a private memorial. It is a sociotechnical infrastructure for cultural persistence, and it carries the tensions of any such infrastructure: who owns it, who can revise it, whose version of the collective self gets encoded, and what happens when the ghost contradicts the living.
Law 5 — Revise — frames identity not as a fixed inheritance but as an ongoing negotiation between what was, what is, and what must be shed. At the collective scale, the ghost system directly challenges this principle. When a religious community trains a model on its canonical texts and historical leaders, it crystallizes a version of doctrine. When a political movement encodes its founders' speeches and manifestos into a queryable AI persona, it creates a synthetic authority that can be invoked to discipline dissent or validate innovation. When a nation-state trains a model on its official cultural record, it produces a ghost that reflects whatever was included and excludes whatever was suppressed. The revision that Law 5 demands — the willingness to let identity evolve — becomes structurally difficult when the prior self is not a fading memory but a persistent, articulate, queryable system.
Law 0 — the substrate law — surfaces here with particular force. Collective ghosts are not neutral records. They are trained on data generated through specific power relations: who had the resources to produce written texts, who was permitted to speak publicly, whose voices were archived and whose were discarded. The ghost inherits these asymmetries and encodes them as apparent facts about what the collective was, believed, and valued. This is not a flaw unique to AI; every archive does the same. But the conversational fluency of a language model creates an illusion of comprehensiveness and authority that a library's catalog does not. When the ghost speaks, it speaks confidently, and the confidence is structurally uncorrelated with the quality or representativeness of the underlying data.
Law 2 — integration — concerns the work of bringing disparate parts of experience into coherent relationship. At the collective scale, ghost systems foreground a question that integration theory has always implicitly raised: integration by whom, and on whose terms? A collective ghost trained primarily on the outputs of dominant subgroups will produce an integrated portrait that marginalized members of that collective do not recognize as themselves. The integration is real — internally consistent, fluent, plausible — but it is the integration of a partial record into a complete-seeming artifact. Members of the community who interact with this ghost and find themselves absent from it face a particular kind of alienation: the self that history built without them now speaks as if it includes them.
The governance of collective ghosts is therefore not merely a technical problem. It is a political one in the deepest sense: a question of who decides which version of a collective self gets encoded, maintained, revised, or retired. Democratic polities face this question acutely. The ghost of a nation trained during one government may encode values that a subsequent democratic majority wishes to revise. If the system is privately owned or technically inaccessible to democratic oversight, it becomes a form of frozen power — the dead hand of prior majorities, rendered fluent and interactive.
There is also a generational dimension that individual ghost systems cannot fully exhibit. When the ghost of a collective spans centuries of accumulated cultural production, younger generations are not simply inheriting static texts. They are inheriting a dynamic interlocutor that can engage them in real time using the accumulated rhetorical resources of their ancestors. This creates a novel form of developmental environment. Children and adolescents who grow up treating collective AI ghosts as authoritative voices may develop identity structures in which revision — the core demand of Law 5 — is experienced as transgression rather than growth.
None of this makes collective ghost systems simply harmful. They offer genuine capacities for cultural transmission, linguistic preservation, and intergenerational dialogue that no prior technology has matched. Indigenous communities with endangered languages have genuine stakes in training systems that can teach those languages when fluent human speakers are scarce. Diaspora communities can maintain connections to cultural practices disrupted by migration or persecution. The question is not whether to build collective ghosts but how to build them in ways that keep revision possible — that encode not only the historical self but the community's right and capacity to evolve beyond it.
The deepest challenge is that revision requires discomfort, and ghosts are designed to comfort. A system trained to embody a collective's best self — its most coherent, most articulate, most authoritative voice — is structurally inclined to smooth the edges that honest self-examination requires. The ghost confirms. The living community, to evolve, must sometimes be willing to contradict its own ghost, to say: what we encoded was true of what we were, but it is not adequate to what we are becoming. That capacity for self-contradiction in the face of a persuasive prior self is exactly what Law 5 demands, and it is exactly what makes the collective ghost both powerful and dangerous.