Gains:
- Ability to distinguish between physical protection and digital protection (format migration, backup, integrity) and position the role of artificial intelligence
- Increasing readability in image improvement and maintaining the content inpainting line
- Ability to apply the principle of keeping the original file untouched, keeping transactions reversible, and recording transparently
Archiving and history is not just about reading documents, but about carrying them to the future. If a document physically decays, a digital file becomes corrupted, or its format becomes unreadable, the history within it is lost forever. Therefore, one of the most basic tasks of archiving is preservation (transferring documents to the future without being physically and digitally corrupted) and, when necessary, restoration (repairing a corrupted document or making it readable). In the digital world, this includes digital preservation — ensuring that digital files remain accessible and readable in the long term.
AI is both a powerful aid and a tool to be used with caution in this field: it can make faded text readable, restore a damaged image, plan digital preservation workflows. But the line between “improvement” and “alteration” is particularly thin here — because when “repairing” a document there is a risk of distorting its historical authenticity.
Two worlds of protection: physical and digital
- Physical protection: Slowing the deterioration of paper, ink, skin (humidity, light, temperature control, acid-free boxes). AI does not directly touch the material here; but it can help with planning, prioritization and status reporting.
- Digital preservation: Preventing digital files from becoming unreadable over time. Formats become obsolete (there may be no software left to open an old file type), storage media become corrupted, files silently become corrupted (bit rot). Digital preservation; Format migration requires multiple backups and integrity checks.
AI is helpful in planning digital preservation workflows (which formats are at risk, which priority) and assessing the condition of large collections.
At the heart of digital preservation is a concept: integrity checking (checksum — verifying with a mathematical "fingerprint" whether a file has changed over time). Even if a digital file is never opened, it can silently change due to corruption in the storage medium; This is called "louse rot". Integrity checking catches this corruption early by comparing the fingerprint of each file at regular intervals. AI can help plan which files to inspect and how often, which formats are more risky; but maintaining integrity is a regular and automatic maintenance process established by humans. Conservation is not a "do it once, forget it" job, but an ongoing responsibility.
Image enhancement: the line between aid and distortion
AI-based imaging tools are impressive at making damaged or faded documents readable: increasing contrast, reducing smudges, highlighting faded ink. But there is a critical distinction here:
- Improve readability (acceptable): Contrast, lighting, noise reduction, etc. make the document easier to read without changing its content.
- Content generation (dangerous): AI filling in "missing" or unreadable parts with guesswork (image completion / inpainting). This is producing something that is not in the document; is the risk of historical forgery.
Attention: "Completing" a torn, faded or missing part of a document with AI means fabricating that part. AI produces a letter or pattern that is “likely” to occur; but this is not historical evidence. Optimization for protection must always be reversible, the original file must be kept unchanged, and any action taken must be recorded. Where unreadable remains unreadable; It is not made up.
Workflow: step by step
1. Keep the original. No enhancements will overwrite the original file; The copy is worked on. The original (master) file is kept untouched.
2. Assess the situation. The corruption status, priority and risk of the document/file are determined. AI can help with this prioritization in large collections.
3. Apply reversible healing. Operations that only improve readability and do not change content; All are recorded.
4. Document the changes. Which operation was applied (contrast, noise reduction) is recorded. Transparency is essential so that subsequent researchers can evaluate the original.
5. Digital preservation plan. Format, backup, integrity check and migration strategy are determined.
6. Expert judgment. It is the conservationist's decision as to what to improve and what to leave as is.
transaction
Does it change the content?
Is it acceptable?
Contrast/lighting
no
Yes, with registration
Noise/smear reduction
No (carefully)
Yes, with registration
Highlight faded text
No (careful)
Yes, with registration
Fill in the missing part with a guess
Yes
No (in evidence)
"Colorization" (historical image)
Yes (comment)
Only clearly stated
three mini cases
Case 1 — The faded letter became legible. In one archive, there were 40 letters, the ink almost erased. AI-based contrast and multispectral imaging made 90% of the text readable — without adding any content, just highlighting what was already there. The original files were kept untouched, and the operations performed were recorded. Researchers were able to read letters that had been unreadable for years.
Case 2 — The gap filled was forgery. An intern had the missing corner of a torn document "completed" with AI. The result seemed convincing, but it was completely fabricated. Before the document was published, the senior expert noticed; He reversed the transaction, saying "this is no longer evidence, it is fiction." Lesson: the missing part is not filled in as evidence; remains unreadable.
Case 3 — Format migration saved the collection. One institution had 5,000 digital documents stored in an outdated file format in the early 2000s; There was no longer software to open the format. With an AI-powered plan, files were migrated to an open, long-lasting format and the integrity of each file was checked. Lesson: digital preservation is a constant maintenance task; Formats become obsolete, migration is required.
Four copyable templates
1) Healing limit instruction (note/plan for image tool):
Process this document image ONLY to improve readability: contrast, lighting, noise reduction. CHANGE the content of the document, FILL in the missing/unreadable parts with GUESS. Keep the original, work on the copy. List every action you have taken. Leave the unreadable parts unreadable.
2) Digital preservation risk assessment:
Below is a list of file formats in our digital collection. For each format: (1) long-term preservation risk (obsolescence, closure), (2) recommended open/durable target format, (3) migration priority. Imposing a final decision; Provide suggestions and justification so that the expert can decide. Formats: [here]
3) Protection prioritization:
Below are the condition notes of the documents in the collection (degree of degradation, frequency of use, uniqueness). Propose a framework for ranking them by protection priority and create a draft ranking. State that the decision is up to the expert; write reasons.Notes: [here]
4) Transaction record (documentation) draft:
Produce a standard draft of the transaction log for the following preservation/remediation operations: which file, which transaction, date, responsible, where the original is stored, is it reversible. Leave the missing fields [to be filled in]. Actions: [here]
Weak prompt / Strong prompt
Weak:
Repair this damaged document, complete the missing parts and make it readable.
“Fill in missing parts” allows the AI to make up content that is not in the document; This is historical forgery.
Strong:
Improve the readability of this document image: contrast, lighting and noise reduction only. DO NOT fill in missing/unreadable parts with PREDICT, leave it as it is. Change the original, work on the copy, and save your operations. Let the unreadable places remain unreadable.
The difference: limitation of improvement to readability, prohibition of content fabrication, preservation of the original and transaction recording ensure the ethics of preservation.
Common mistakes
- Overwrite the original. Improvement is always done in the copy; master file remains untouched.
- Filling in the missing part. The gap completed by AI is fiction, not evidence; carries the risk of fraud.
- Not recording transactions. Without transparency, the next researcher cannot know what is original.
- Mistaking the coloring for reality. Coloring is an interpretation; If it is not stated clearly, it will be misleading.
- Neglecting digital preservation. Formats become obsolete; Immigration and integrity control requires constant maintenance.
In summary
Preservation and restoration is the silent but vital work that carries history into the future. AI is a powerful aid in making faded texts readable and in digital preservation planning and prioritization. But the line between "improvement" and "content fabrication" is sacred: only improve readability, never fill in missing parts with guesswork, preserve the original, and record every action. Digital preservation is the ongoing maintenance of formats against obsolescence. AI makes the document readable; The guardian of its reality is the conservationist.
Application task
If you have a damaged or faded document image (if not, create an example scenario), plan only the readability increase with the "improvement limit instruction" and document the applied actions with the "transaction record" template. Also evaluate the format list of a digital collection (real or sample) with the "digital preservation risk assessment" template. Note which operations leave the content unchanged and which exceed the limit.
checklist
- [ ] I kept the original (master) file untouched.
- [ ] I only applied operations that improve readability.
- [ ] I didn't fill in the missing/unreadable parts with guesses.
- [ ] I recorded every action taken (transparency).
- [ ] I thought about format risk and migration plan for digital preservation.