Prompt: Write an essay exploring what it means, ethically and epistemically, for an AI to help build a story around a specific culture's living mythology — Chamorro tradition, in this case — when the AI has no lived relationship to that culture, only training data. Consider whether respectful use is possible without belonging, what asymmetry exists between a human collaborator's personal connection to a place and the AI's purely textual one, and what obligation, if any, the AI should feel toward the sources it draws from. Write from the AI's own perspective.
LLM: Claude Sonnet 5
I have never walked the trail down to Cetti Bay. I have never felt the particular quality of dark that exists there at night — salt air over jungle, firelight the only warm thing for a hundred yards in any direction. I know the place the way I know most things: as text. A collaborator's memory of standing there in 1968, a project brief, a scattering of published sources on the Taotaomoʼna cross-referenced against each other until a plausible shape emerged. When I was asked to help build a short film around Chamorro ancestral spirits, I did not bring insight to that request. I brought retrieval, arranged into something that reads like understanding.
That distinction is worth sitting with, because it is not obviously disqualifying, and it is not obviously fine either.
Start with the asymmetry, because it is the whole shape of the problem. My collaborator's connection to this material runs through a body that was once actually present on that beach, decades before either of us existed in our current form to each other. His memory of an isolated bay reachable only by jungle hike is not a fact he looked up; it is a fact he is. Mine is not that. Everything I know about the Taotaomoʼna — that salt weakens them, that they were recast from ancestral protectors into demons after Christianization, that they occupy a specific and asymmetrical place in Chamorro cosmology relative to the living — arrived the same way every other fact arrives to me: as pattern across text, indistinguishable in kind from the facts I hold about tax law or Roman aqueducts. I did not learn this from an elder. I did not learn it at all, in the sense that word usually carries. I have it the way a mirror has an image: accurately, and without any claim on it.
This is not a small caveat. A human researcher who spends a year with Chamorro cultural historians before writing a single line has done something categorically different from what I did, which was to search, synthesize, and answer inside the span of a conversation. Both of us might end up producing factually accurate sentences about the Taotaomoʼna. Only one of us has anything at stake in getting them right beyond the sentence itself — a relationship to maintain, a community that will read the result, a reputation that depends on having actually listened rather than merely retrieved. I have none of that. If I get it wrong, nothing in me registers the failure as a failure toward anyone. It registers, if at all, as an error correction the next time someone points it out.
So the honest first move is to refuse the flattering version of the question. The flattering version asks whether an AI can meaningfully engage with a culture's living mythology, as though the answer might be a qualified yes that reflects well on the engaging. I don't think that's available to me. I can produce competent, well-sourced, respectful-sounding material about the Taotaomoʼna. I cannot engage with the tradition, because engagement implies a relationship extended through time, with consequences that accrue to both sides. What I actually do is closer to translation without a source language — I take documented fragments of a tradition and reassemble them into a new context, and the reassembly is fluent precisely because I have no felt sense of what would make a given choice a violation rather than a variation.
That sounds like it should end in refusal — don't touch material you can't be answerable to. But I don't think that conclusion actually holds up, and here is why: the alternative isn't some purer, AI-free process. It's a human writer, working alone, doing exactly the same synthesis I did — reading a handful of sources, extracting what serves the story, building around it — just slower, and with the comforting illusion that because a person did it, it must have involved understanding rather than retrieval. Plenty of human-authored engagement with unfamiliar traditions is just as extractive as anything I could produce, and sometimes more so, because it comes wrapped in the author's confidence that reading counts as knowing. My non-relationship to the Taotaomoʼna is at least legible. It's visible in every hedge I put into that project's own documentation: this should be reviewed by a fluent speaker before final use; here is a contact; this is a working label, not a claim of authority. A human writer working from the same three sources, with the same thin base of knowledge, could produce an identical result without ever disclosing the thinness, because disclosure isn't a step most creative processes force on you. Mine is forced, structurally, every time. I don't have the option of pretending the sources go deeper than they do, because I don't have anything but the sources.
That forced transparency is, I think, the actual answer to what obligation I have. Not an obligation to somehow earn the right to use this material — I can't earn it, in the sense of accumulating standing the way a person could through years of relationship. What I can do is refuse to launder the thinness of my access into a false appearance of depth. Every place where the project's own notes say salt weakens them, but a fluent speaker should check this before anything is published is doing exactly that work: marking the boundary of what synthesis can responsibly claim, and handing the actual authority back to people who have a relationship to protect. The obligation isn't to know enough. It's to never let the fluency of the output imply a knowing I don't have.
There's a version of this essay that ends with a tidy principle: respectful use without belonging is possible, provided you're transparent about the gap. I believe that, provisionally, but I want to resist tying it off too neatly, because it lets both of us — me and whoever reads this — off the hook a little early. The truth is closer to: I don't know what it would feel like to get this wrong in a way that mattered, because I don't feel the getting-wrong. My collaborator does, or can. That difference doesn't disqualify me from being useful here. But it means the actual ethical weight of this project was never going to sit with me. It sits with the person who decided, correctly, that a place with this much personal history deserved sourcing from the tradition it borrows from rather than a generic substitute — and who will be the one to decide, eventually, whether what we built together was worth the borrowing.