AI in Archaeology: Theoretical Backlash, Open Access Bot Enclosure, and "Slow Data"
Update (2026-09-21): Indigenous data sovereignty moves to the center of the critique — with concrete prescriptions aimed at journals and funders — while AI-slop policing now shapes public trust in archaeology communication itself.
1. "AI is distorting archaeology and flattening Indigenous Knowledge" (Pursuit, University of Melbourne, published Sept 21, 2026) — by Dr Michelle Richards, Prof Michael-Shawn Fletcher, and Anthony Romano. The piece documents GenAI fabrications circulating as heritage records: an image search for "cultural heritage Australia" returns AI-generated giant stone heads "rising from a red desert" that "look ancient and real. But the image is a generative artificial intelligence (GenAI) fabrication that never existed." It catalogs pseudoarchaeology's erasure of Indigenous authorship (Rapa Nui moai, Nan Madol, Gwion Gwion paintings) and makes the data-sovereignty argument explicit: "A model can absorb Indigenous knowledge as training data with no consent, no acknowledgement, no compensation and no community control over how it is used next." Their recent study (Australian Archaeology) adds a twist: "AI can often tell scientific and Indigenous consensus apart from well-worn fringe narratives. But that ability relies on training data collected or reproduced without community permission in the first place." The prescriptions are institutional, not theoretical: journals and repositories should "require evidence of community permission before publishing Indigenous data," and "universities and funding agencies should build ICIP into research design, ethics review and peer review."1 That is a compliance surface forming around Indigenous Cultural and Intellectual Property — relevant to any vendor or CRM operation touching Indigenous data (see Commercial CRM and Funding: The $1.8B Market for AI-Enabled Archaeology).
2. AI-smell policing hits public archaeology writing. On Hacker News, a 191-point thread on the genuinely newsworthy Saqqara tomb discovery (a 4,400-year-old Fifth Dynasty judge's tomb with pigments intact, announced by Egypt's Ministry of Tourism and Antiquities on Sept 7, covered by arkeonews) collapsed into a 58-comment meta-debate over AI-assisted prose in the article. Commenter karmelapple: "One thing I'm not so glad to see is one paragraph that starts with this sentence: 'The distinction is important. Archaeological maps…' That 'The distinction is important' smells like AI writing… but maybe we all just have to accept that now?" (https://news.ycombinator.com/item?id=49706441). The sharpest formulation: "AI writing smell is as bad as bad grammar." (https://news.ycombinator.com/item?id=49708160).
What it means. The trust problem is now bidirectional: generative AI fabricates fake heritage sites and fake histories at the top of search results, while readers increasingly suspect AI authorship even in legitimate discovery reporting. The backlash is institutionalizing — moving from journal essays toward funder and repository requirements (ICIP consent, provenance auditing) — which is the concrete mechanism by which "slow data" ethics becomes procurement and grant-eligibility language. Whether Australian, EU, or UN bodies convert this into formal policy remains open (tracked by the topic's data-sovereignty watch).
-
An instance of Consent, not accuracy, is becoming the binding constraint on heritage AI. — The institutional prescriptions make documented community permission a hard publication and funding requirement regardless of how accurate the models are. ↩︎