Gains:
- Being able to distinguish where artificial intelligence saves time in the archaeological workflow (classification, measurement, summary, draft) and where decisions such as interpretation, dating and cultural attribution are left to the expert, depending on the level of risk.
- Ability to apply a discipline that verifies each output through the steps of linking it to the real source, confirming it with an independent source, and passing it through the filter of expert judgment.
- Understand why the confidentiality of sensitive sites, human remains and unpublished finds, and the risk of hallucination and authenticity must be taken into account from the very beginning.
You are in the middle of an excavation season. The drilling site yielded hundreds of ceramic shards (broken pottery), a handful of coins, bones and glass finds; Each one is waiting to be recorded with a context number (the layer, pit or place where the find was found — in archeology, the meaning of an object depends on where and with what it was found). Simultaneously, the survey team is sending drone photos, a graduate student is digitizing layer drawings, and the museum is requesting text for the next exhibit. The field of archeology (the science of studying past human societies through material remains—finds, structures, layers) and cultural heritage (tangible and intangible values that a community inherits and preserves from the past) is inherently data-intensive, multidisciplinary, slow, and meticulous. Here, artificial intelligence (AI - software that classifies images by extracting patterns from historical data, summarizes text, translates and produces drafts) accelerates you in this abundance of data.
But the very beginning of this module is clear: AI is a classification, measurement, summary and outline assistant; You are the competent expert who decides what a find means, how to interpret a layer, and what story to tell about the past. Archeology is a human science; The evidence is fragile, the context is unique, and the responsibility for interpretation lies with the human.
In this first unit, we will focus on the discipline, not the tool: you will learn where in the archaeological workflow AI saves real time, where it is dangerous, how to verify each output, how to protect precise location and find data, and how to stay alert to hallucination (AI producing something that does not exist as real).
Where does AI come in handy in the archaeological workflow?
Let's divide the archaeological work into two large clusters. First cluster: voluminous, repetitive, pattern-removable tasks. Preliminary sorting of thousands of ceramic sherds, visual grouping of similar finds, anomaly marking in satellite and LiDAR images, digitization of excavation notebooks and inventory cards, first summary of long literature, translation of multilingual publications, alignment of hundreds of photographs for photogrammetry. In these jobs, AI reduces days to hours and does not get tired.
Second cluster: tasks requiring interpretation, dating decision, cultural attribution and responsibility. What period and culture a find belongs to, stratigraphic interpretation of a layer, why a site was abandoned, the meaning of an inscription, how much of a reconstruction is based on evidence and how much is based on assumption, how to protect a heritage site. These decisions require expertise, knowledge of context, and scientific responsibility. Here, AI generates hypotheses and sketches — but the expert has the final say.
Let's clarify the distinction in one sentence: AI is strong on "what's in this pile, what does it look like, and what does the first draft look like" questions; The decision is yours when it comes to questions such as "what does this really mean and can I defend it scientifically?"
Tip: Before outsourcing a job to an AI, ask: “What do I lose if this output is wrong?” If the answer is "a few minutes of fixing", feel free to delegate. If the answer is "wrong dating, a fabricated source or a leaked site location", let the AI produce the draft and you make the decision and verification.
Hallucination and authenticity: the special risk of the human domain
Archeology is a science of text and hermeneutics; That's why AI's most dangerous mistake occurs here. The AI writes fluently and confidently, but frequently produces hallucinations: an excavation report that doesn't exist, a fabricated author-year attribution, a radiocarbon date that doesn't exist, a misattributed museum inventory number, or a comment that "sounds right" but has no source. A fake source pretending to be in an academic article or report destroys the credibility of the entire work.
In addition, there is the issue of originality: the text produced by the AI may resemble the sentences of others and carries the risk of plagiarism (appropriating someone else's idea/expression without citing the source). In a scientific study, the AI-generated manuscript is never published as is; It is rewritten with your ideas, your analysis, and references with verified sources.
Attention: "The AI said so" is not a scientific justification. If there is a fabricated source, incorrect dating, or a leaked site location, the responsibility belongs not to the AI, but to the researcher who uses that output without verifying it.
Verification discipline: three steps
Apply this three-step reflex to each AI output:
- Connect it to the source. Every claim must be based on a concrete, factual source (published report, inventory record, measurement). "In which publication, on which page, whose work is this information?" Ask and see the source for yourself. Always confirm the DOI, ISBN, or inventory number YZ provides with an actual catalogue.
- Confirm with independent source. Verify any significant dating, attribution, or comment with at least two independent expert sources.
- Expert judgment filter. Does the output match stratigraphy, context, and known typology? Is there an alternative interpretation? Your expert assessment is the final filter.
Privacy and precise location
In archaeology, one type of data requires special protection: site locations. If the coordinates of an unexcavated mound, a cave painting, or a shipwreck are revealed, they become a target map for illegal diggers and treasure seekers. So don't enter publicly available AI tools that haven't been validated with precise coordinates, unpublished finds, and sensitive information about human remains. In addition, information deemed culturally sensitive by indigenous and local communities is protected with the same care.
three mini cases
Case 1 — Pre-sorting saved time. An excavation team encountered approximately 12,000 ceramic sherds in a single season. AI-assisted visual grouping pre-sorted pieces into piles based on form and dough similarity, pre-sorting the pile for the expert to inspect by hand in half a day instead of a day. But the expert himself made the period reference for each cluster.
Case 2 — Verification caught a fake source. A student had the AI extract the literature of a region; AI gave a convincing list of "author, year, journal, page". When the student searched the library catalog, he found that two of the three articles listed did not exist at all—the AI had combined real author names with made-up titles. The linking step prevented fictitious citations from entering the thesis bibliography.
Case 3 — Return from location leak. A new assistant pasted the GPS coordinates of an as-yet-unpublished rock shelter into a public tool to draw a quick map. The head of the excavation realized: this was a sensitive location that was not protected. The work was only redone in the in-house secure system and the coordinates were not recorded.
Four copyable templates
1) Job suitability assessment:
Your role: senior archaeologist consultant.I will describe the job below. Tell me (1) whether this work is a classification/measurement/summary/sketching job that can be delegated to AI, or a critical decision requiring interpretation, dating, or cultural attribution, (2) the scientific cost of incorrect output, (3) the verification I need to do before and after delegation. Job: [insert job here]
2) Obligation to link to source:
I will give you archaeological texts. For each claim, MUST indicate the actual source: publication name, author, year, page or inventory number. Do not make any claims that have no source, do not fabricate citations or DOIs. If you are unsure, mark it as "must be verified".
3) Precise data pre-check:
Evaluate the following text BEFORE processing: does it contain precise site coordinates, unpublished artifacts, human remains, or culturally sensitive information? If so, which sections? Warn me if it shouldn't be processed in a public vehicle as it is. Text: [write text here]
4) Interpretation-evidence distinction:
When interpreting the find/layer data below, make a distinction between the exact FACT (measured/observed) and INTERPRETATION (inference/hypothesis). For each interpretation, write down what evidence it is based on and what an alternative interpretation might be. Do not make precise dating or cultural attribution; present options.
Weak prompt / Strong prompt
Weak prompt:
To which period do these findings belong? Tell us the history of the site.
This prompt is context-free and asks the AI to make a scientific decision. AI predicts, invents, gives confident but unsubstantiated history.
Powerful prompt:
Your role: ceramics assistant. Below I will paste the size, dough color, form and surface treatment data of 15 parts. Task: group the parts according to their similarities, based only on these data I have given, and write the POSSIBLE form type and justification for each group. DO NOT make a precise period reference; Tell me what additional analysis (reference catalogue, archive) is required instead. Mark if you are not sure.
The difference is clear: the data, the "only based on that data" constraint, and the "strict attribution" constraint make the output safe and useful.
Role/task comparison chart
business
Role of AI
man's role
verification
Preliminary classification of finds
Visual grouping outline
Period/culture reference
Reference catalog
Site detection (remote sensing)
Anomaly marking
Field confirmation, comment
terrain control
3D documentation
Model/measurement production
Comment, survey decision
Scale verification
Literature review
First summary draft
argument, choice
Citation confirmation
Report/exhibition text
skeleton sketch
Content, signature
Source mapping
Common mistakes
- Mistaking AI output for fact. The output is always a draft to be verified; It does not enter the report without connecting to the context and source.
- Relying on fabricated attribution. AI produces realistic but fake resources; Each reference is confirmed in the actual catalogue.
- Giving the precise location/find to the open vehicle. If it leaks, it will invite illegal excavation; It is the most dangerous mistake.
- Asking for precise dating/attribution. The decision belongs to the expert; AI generates options and hypotheses.
- Publishing the AI text as is. Originality and risk of plagiarism; The text is rewritten with your analysis.
In summary
AI is a powerful assistant in archaeology: it visually classifies, measures, summarizes, translates, produces sketches. But interpretation, dating, cultural attribution, and ultimate responsibility rest with humans. Carry two-set separation (voluminous-pattern works vs. interpretation-liability decisions), three-step verification (link to source, independent verification, expert filtering), sensitive data awareness and hallucination/authenticity alertness as the backbone of this module.
Application task
List 6 tasks from your own (or imaginary) digging/research desk. Classify each as “AI-transferable pattern/summary/outline” or “expert commentary/attribution decision.” For one of the transferable ones, use the “Job suitability assessment” template above and get a response from the AI; then apply three-step verification and sensitive data precheck.
checklist
- [ ] I divided the work into two clusters (pattern / comment-responsibility).
- [ ] I implemented three-step verification (source, independent fact-checker, expert).
- [ ] I did not give the sensitive site location/find to the open vehicle.
- [ ] I verified every reference given by YZ in the actual catalogue.
- [ ] I checked the distinction between fact and interpretation and originality in the output.