Ethics, Legal, Anti-Plunder and End-to-End Workflow

Gains:

• Ability to understand the distinction between provenance and provenance, the legal framework and the principle of transparency, and produce an AI usage declaration and ethical preliminary screening.
• Protect sensitive protection data while using AI to combat plunder with satellite tracking and online commerce scanning
• Ability to establish an end-to-end workflow that separates the role of AI from human decision at every step, from planning a project to publication, and keeps the final interpretation and responsibility in the hands of the human.

In this last unit, we establish the framework that holds the previous units together: we will see the ethical and legal principles we must comply with when using artificial intelligence in archeology and cultural heritage, the role of AI in combating illegal excavation and looting (illegal excavation of archaeological sites and the smuggling of finds), and the end-to-end workflow that combines all steps in a single project. The goal is to integrate the tools you have learned into a responsible whole.

The basic principle is the final version of the sentence we repeat throughout the module: AI; is a powerful assistant that classifies, measures, scans, summarizes, translates and produces drafts; but interpretation of the past, cultural attribution, protection order, legal responsibility and ethical judgment belong to the competent person. In a high-consequence and irreversible field, AI output is no substitute for expert validation; it only speeds it up.

Ethical and legal framework

Legal basis. Archaeological excavation, surface research and artifact processing; It is subject to national protection laws, permits and international agreements. There are international agreements (for example, the framework set by the 1970 UNESCO Convention) to prevent illegal trade in cultural assets. AI is not a legal advisor; The answers to legal questions are taken from the competent authorities and legislation.

Provenance and provenience. Two similar terms are two different things: provenience (the exact location and context of a find in the excavation—the source of scientific value) and provenance (the history of ownership and ownership of an object from its discovery to the present—evidence of legality). If the provenance of an object is uncertain, it may have been looted; museums and researchers avoid working with such objects.

Data ethics. Sensitive locations, human remains and culturally sensitive information are protected; It is not given to open vehicles.

Transparency. The use of AI is clearly stated in scientific work: in which work, which tool, which validation. Concealment is against scientific integrity.

Tip: Draft an “AI use statement” at the beginning of each project: in which tasks you will use AI, where the data will be processed, verification steps and privacy rules. This statement provides ethical transparency and holds the team to a common standard.

AI in the fight against looting

AI is also a weapon to protect cultural heritage:

• Satellite tracking. The increase in plunder pits in protected areas over time is automatically marked (see unit 9).
• Online business screening. Objects of questionable provenance circulating on auction and sales platforms are scanned and presented for expert examination.
• Visual matching. Stolen/unregistered objects are compared with images in databases.

But be careful: AI saying "this object has been looted" or "this looks like that stolen artifact" is a start. The legal and scientific conclusion is the matter of the expert and competent authority; A false accusation is as harmful as a missed evidence.

Caution: You must maintain precise location information even when using the AI ​​to combat looting. Ironically, a protective analysis can provide a map to looters if carelessly shared. Preservation data is at least as confidential as the find.

End-to-end workflow: an example project

Let's consider a survey and evaluation project from start to finish; At every step, the role of AI and humans is clear:

1. Planning. Legal permissions are obtained, an AI usage statement and a privacy plan are written. (Human: responsibility)
2. Detection. Remote sensing and LiDAR and AI mark possible areas. (AI: screening; human: elimination)
3. Land confirmation. Priority points are checked on site. (Human: observation)
4. Documentation. Finds and structures are documented by photogrammetry. (AI: model; human: scale/comment)
5. Find processing. Visual grouping and description are made. (AI: preliminary classification; human: attribution)
6. Registration. It is entered into a standardized database. (AI: regulation; human: integrity)
7. Analysis. Dating and statistics are carried out. (AI: calculation; human: method/interpretation)
8. Literature. It is compared to real sources. (AI: summary; human: attribution confirmation)
9. Reporting and publication. Transparent, sourced, original text is written. (Human: comment, signature)
10. Protection and sharing. Monitoring is established, an honest narrative is prepared for the public. (Human+community)

Three-step verification (link to source, independent fact-checking, expert filtering) and sensitive data protection apply at every step.

three mini cases

Case 1 — Transparent disclosure built trust. One team clearly wrote in their publication in which tasks they used AI (HTR, preliminary classification, translation draft) and how they verified it. Reviewers evaluated this transparency as credibility; A concealed use would raise suspicion.

Case 2 — Detection of looting led to intervention. One organization documented nearly 300 spoil pits appearing at one site in one season through AI satellite analysis; After expert verification, it was reported to the authorities and the area was taken under protection. The scale was too large for the human eye alone to capture.

Case 3 — Suspected provenance discontinued. A researcher examined the provenance of an object that the AI ​​marked in an online sale; The object had an unexplained gap in its history and appeared to be a known looting site. The investigator refused to work with the object and reported the situation. The AI ​​had caught his attention; The ethical principle made the decision.

Four copyable templates

1) Draft AI use declaration:

Your role: research ethics assistant. I will describe the following project. Give me a draft "AI usage statement": in which tasks AI will be used, where the data will be processed, which verification steps will be applied, which sensitive data will be protected, how the usage will be transparently reported. Emphasize that legal responsibility lies with the person.

2) Provenance risk control:

I will evaluate the history of an object. What questions should I ask in terms of provenance (ownership history), what gaps might be signs of looting/smuggling, what documents should I look for? List legal and ethical red flags. Don't make a final judgment; remove the points to be checked.

3) End-to-end verification list:

I will give the steps of the following project: [list]. For each step, a control chart emerges with the columns: AI's role, human decision, mandatory verification and sensitive data risk. Make sure that the final interpretation/decision is not taken from the human at any step.

4) Ethical/legal pre-screening:

Review the following plan from an ethical and legal perspective: is there a permit required, is there sensitive location/human remains, is community involvement required, is there a provenance issue, is the use of AI transparent? List the risks and authorities to consult. Don't give legal advice; tell me who to consult.

Weak prompt / Strong prompt

Weak prompt:

Let AI do everything in this project, produce results and reports quickly.

This approach delegates interpretation and responsibility to the AI; It produces a result that is unverified, of questionable origin, and overlooks ethical and legal risks.

Powerful prompt:

Your role: project coordination assistant. For each step of this project, a plan emerges that distinguishes where the AI ​​will produce drafts/scans and where the human will make comments, decisions and signatures. Add authentication and sensitive data protection to every step. Include legal authorization, provenance and transparency checks. State at every step that ultimate responsibility remains with the person.

The difference: the first transfers responsibility to the vehicle; Secondly, it uses AI in the right place and keeps the decision and honesty in the human.

Ethics and workflow summary table

principle

Meaning

The limit of AI

legality

Permission, legislation, contract

Cannot give legal opinion

provenance

Is the ownership history clean?

Doubt marks, does not judge

Sensitive data

Position/relic preserved

Not allowed on open vehicles

transparency

The use of AI is declared

Usage is not hidden

Responsibility

Interpretation and decision belongs to the person

Does not replace consent

Common mistakes

• Transferring responsibility to AI. Interpretation, decision and signature are always in the human hands.
• Hiding the use of AI. Transparency is part of scientific integrity.
• Accepting provenance without question. Unclear past could be a sign of plunder.
• Careless sharing of protection data. The precise location could be a map to the looter.
• Mistaking AI for legal counsel. Legal answers are received from the competent authority.

In summary

The ethical and legal framework holds all the module's tools together. AI; It is a powerful aid, from combating plunder to project coordination, but interpretation, cultural attribution, preservation decision, provenance judgment and legal responsibility rest with the human. Bring transparency, sensitive data protection, three-step verification and community engagement every step of the way. The past is fragile and unique; The vehicle that accelerates it cannot replace the responsibility that protects it.

Application task

Design a small archaeological project (real or imaginary) and create a table with columns of AI role, human decision, verification and sensitive data for each step with the “End-to-end verification list” template. Then write your project's transparency and ethics framework with the "Draft AI use declaration" and "Ethical/legal pre-screening" templates. Perform "Provenance risk check" for an object instance.

checklist

• [ ] I separated the role of AI and human decision at each step.
• [ ] I have transparently declared the use of AI.
• [ ] I have protected sensitive location, artifact and protection data.
• [ ] I checked the provenance and legal risks and referred them to the necessary authorities.
• [ ] I kept the final interpretation, decision and signature in the human.

Module Exam

1. Which of the following is the most accurate positioning for artificial intelligence in archeology and cultural heritage?

• A) Artificial intelligence can precisely determine the period and culture of a find without expert approval
• B) Artificial intelligence is only useful for producing beautiful images, it has nothing to do with scientific workflow
• C) Artificial intelligence is a classification, measurement, screening and drafting tool; interpretation, attribution and final responsibility rest with the expert ✔
• D) Since artificial intelligence is always more impartial than humans, the dating decision should be left to it.

Description: Artificial intelligence; It is an assistant that classifies, measures, scans, summarizes, translates and produces drafts of finds. Interpretation, cultural-chronological reference, preservation decision and final responsibility belong to the competent expert; An unverified printout may lead to an incorrect date, a fabricated source, or a leaked site location. Archeology is a human science and a high-consequence, irreversible field.

2. What is the most appropriate approach when giving a ceramic piece to artificial intelligence in found image analysis?

• A) Requiring precise civilization, age and function attribution from a single photograph
• B) Only the observable features are described impartially and the expert makes the attribution decision with the reference catalogue. ✔
• C) Accepting the first reference to artificial intelligence without question, even in rare finds
• D) Classify quickly regardless of photo quality, scale and color card

Explanation: It is safe to ask the AI 'what it looks like' (neutral description: form, surface, colour, decoration) rather than 'what it is' (period/culture reference). The citation decision is made by the expert with the reference catalogue; Additionally, if metavisual data such as dough, profile and context are added, the grouping accuracy increases. AI tends to force rare instances into the most familiar pattern.

3. In remote sensing, what is the correct attitude for an anomaly marked by artificial intelligence in a satellite or LiDAR image?

• A) The dot is a 'candidate'; It is not declared a site without verification through land confirmation and its location is protected ✔
• B) The point is a direct discovery and should be made public immediately with its exact coordinates.
• C) Since artificial intelligence sees geometry, there is no need to eliminate natural/modern explanations
• D) A single image is sufficient in all circumstances; Season and method differences are unimportant

Description: Every point marked by artificial intelligence is a hypothesis that needs to be confirmed in the field; It's not a discovery. Many of the signs may be natural (geological) or modern (agriculture, building) and should be eliminated. Additionally, sensitive locations should be protected from looting, not exposed to open vehicles, and blurred on air.

4. What are the most critical scientific requirements and the greatest risks in 3D documentation with photogrammetry?

• A) The only criterion is that the model looks realistic; scale and spaces are unimportant
• B) The model tells automatically and precisely what is wall and what is natural stone, so there is no need for interpretation.
• C) Having artificial intelligence fill in the missing parts is always safe and increases the documentation value.
• D) Scale bar and control points are a must; Missing data is not filled in - marked, evidence and prediction are separated ✔

Description: Scale bar and control points are essential; Even though a model without scale is 'beautiful', it is scientifically worthless because no real measurements can be taken from it. The biggest risk is that AI will fill in missing data with a 'plausible' surface, producing a detail that has not actually been observed; Therefore, weak/filled areas should be marked, visually separating evidence from predicted completion.

5. What is the proper way to maintain data integrity when working with AI in excavation logging and database?

• A) Let the artificial intelligence complete the missing fields and replace the raw data with the corrected version.
• B) Protecting raw data, leaving the missing 'incomplete' and keeping every automatic change traceable ✔
• C) Allowing everyone to enter any term they want without using a standard and controlled dictionary
• D) Leaving the stratigraphic interpretation entirely to artificial intelligence and accepting the Harris Matrix without approval

Explanation: AI helps with standardization and control, but filling in missing data gives it permission to make up; the missing must remain 'incomplete'. Raw data is never overwritten, every automatic change is kept traceable and the original is preserved. Stratigraphic interpretation belongs to the expert; AI only finds logical contradiction.

6. What is the biggest danger in reading a faint inscription or an old manuscript with artificial intelligence?

• A) Artificial intelligence is of no use at scanning manuscripts because it is too slow
• B) Artificial intelligence can only read modern printed texts, not historical writings at all
• C) Artificial intelligence completes the vague text by making up the text and produces a fluent but incorrect translation; Prediction may be confused with what is read ✔
• D) AI translations are always exactly the same as the source text, so there is no need for verification

Explanation: In epigraphy, which is a human field, the most devastating mistake is hallucination: artificial intelligence fills indistinct/broken parts with 'plausible' but made-up text, translates a word it does not understand fluently but incorrectly, and may give a publication citation that does not exist. The solution is to keep the character read and the predictive completion separate and marked, to check the translation against the source and back-translation, and to leave the final reading to the expert philologist.

7. What is the most effective method against the risk of artificial intelligence producing 'fake sources' in the literature review?

• A) Find and cite the actual sources yourself and confirm each citation by placing the 'based only on the source given' constraint ✔
• B) Say 'suggest 15 important sources' directly from the artificial intelligence and put the list as it is in the bibliography
• C) Assuming the record exists as long as the citation format appears correct
• D) Converting the summary produced by artificial intelligence into a scientific claim without ever opening the source

Explanation: Generic tools produce convincing but fake citations that combine the real author name with the made-up title without looking at a real catalogue. The most effective defense is to find the sources yourself from real catalogs, give the texts to the artificial intelligence and impose the 'based only on the given source' restriction; Additionally, each citation is confirmed in the catalog/DOI and each abstract is verified by referring to the source.

8. Which of the following is the correct attitude when interpreting a radiocarbon (C14) result?

• A) Reduce the result to a single definitive year and remove the uncertainty range from the report
• B) Accepting the number given by artificial intelligence without caring where and how the sample comes from
• C) Considering correlation as direct causality and presenting a weak pattern as a strong result
• D) Carrying the result as a probability range and questioning the sample validity and method with an expert ✔

Explanation: A calibrated radiocarbon date is a range of possibilities (e.g. 800-680 BC with 95% confidence), not a single exact year. Reducing uncertainty to a single year is a scientific distortion. Additionally, with the principle of 'garbage in, garbage out', sample validity (old wood, contamination, mixed context) should be questioned and method selection should be left to the expert.

9. How should artificial intelligence monitoring warnings and restoration recommendations be evaluated in cultural heritage management?

• A) When AI flags damage, it is an accurate structural assessment without the need for confirmation
• B) Monitoring alerts are linked to field/expert confirmation; visualization does not replace the irreversible decision of physical restoration ✔
• C) The completed image produced by artificial intelligence can be directly applied as a restoration plan
• D) Because conservation decisions are technical, they must exclude legislation and community participation

Description: Artificial intelligence is an early warning tool in monitoring: it marks cracks, surface loss or spoilage, but each warning requires terrain and expert confirmation. The 'restored' image it produces is a visualization, not a physical restoration plan; conservation decisions are based on the principles of least intervention, reversibility, distinctiveness and uniqueness, interdisciplinary expertise, law and community participation.

10. What is the ethical obligation when using a reconstruction image produced by artificial intelligence in a museum exhibition?

• A) If the image is realistic enough, there is no need to label it to avoid misleading the visitor
• B) Since attractiveness is the most important criterion, it is necessary to hide the uncertainty and present a single definitive story.
• C) The image should be clearly marked as 'reconstruction', elements should be labeled according to the level of evidence and evidence should be separated from imagination ✔
• D) Presenting unproven colors and decorations as 'real' makes the exhibition more educational.

Explanation: Reconstruction continues with imagination where the evidence ends, and the AI ​​can safely make up the gaps and slip into cliché. To prevent the visitor from mistaking the prediction for reality, it should be clearly stated that the image is a 'reconstruction' (not the actual photograph), items should be labeled according to the level of evidence, evidence should be separated from fantasy, and alternative interpretations should be offered where possible. Accuracy comes before attraction.

11. What is the difference and importance between 'provenience' and 'provenance' in archaeological ethics?

• A) Provenience is the place of occurrence/context (scientific value), provenance is the history of ownership (evidence of legality); ✔ Can be a sign of spoilage of uncertain provenance
• B) Both are the same thing and only show the monetary value of an object
• C) Working with an object is always smooth, even if the provenance is uncertain
• D) Provenience is the price of the object, provenance is its weight

Description: Provenience is the exact location and context of a find in the excavation; is the source of scientific value. Provenance is the history of ownership and change of ownership of the object since its discovery; It is proof of legality. An object of uncertain provenance may have been looted; museums and researchers avoid working with such objects. Artificial intelligence helps flag questionable provenance, but it is up to the expert and competent authority to judge.

12. What is the most important point of attention when using artificial intelligence in the context of sensitive sites and combating plunder?

• A) It is safe to share sensitive locations publicly because it is for protection purposes.
• B) When the AI says 'this object has been looted', this is a definitive legal verdict without the need for an expert
• C) In order to speed up the fight against looting, it is necessary to enter all site coordinates into open vehicles.
• D) Preservation data is at least as sensitive as the find; Exact locations are not given to open vehicles, they are kept in the secure system ✔

Description: AI is a powerful tool in combating looting with satellite tracking and scanning of online commerce, but even a protective analysis can provide a map to looters if carelessly shared. Exact coordinates, unpublished finds and precise locations are not released to unapproved open tools and are kept in the secure system; Preservation data is an asset that should be protected at least as much as the find.

13. Which is true about the limits of artificial intelligence in covering archaeological literature?

• A) Artificial intelligence knows all archaeological publications in every language and on every date.
• B) Gray literature and foreign language sources are unimportant, only English indexes are sufficient
• C) His knowledge is frozen in a certain date, he has limited knowledge of multilingual and gray literature; If it says 'there is no source', the archive should be asked ✔
• D) When artificial intelligence says 'I couldn't find a source', it means there really is no work on that subject.

Description: Artificial intelligence's knowledge is frozen at a certain date; It lacks subsequent excavations and corrections and can present superseded views as valid. Archeology is written in multiple languages; relying solely on English sources misses much of the literature. Additionally, a significant portion of the information is in gray literature (commercial/corporate reports, dissertations) that cannot be easily found in directories; Artificial intelligence either doesn't know this or makes it up.

14. What is the correct framework for using end-to-end AI in an archaeological project?

• A) For speed, leave all steps to artificial intelligence and transfer the interpretation and responsibility to it
• B) Separating the AI role from human decision at each step, adding verification and transparent declaration, keeping the final responsibility with the human ✔
• C) Make the work look more reliable by hiding the use of artificial intelligence in the publication
• D) Focusing only on technical results, bypassing legal permission, provenance and community involvement

Description: The correct workflow separates the drafting/scanning role of AI and the human role of interpretation, decision, and signature at each step (detection, documentation, find processing, recording, analysis, literature, publication, preservation); It adds three-step verification and sensitive data protection to every step. Additionally, the use of artificial intelligence is transparently declared in scientific studies; Concealment is against scientific integrity, and ultimate responsibility always remains with the human.