Can AI claims about ancient writing be tested before people believe them?

Seattle, Washington
Computer ScienceAnthropology
$0
Pledged
0%
Funded
$4,900
Goal
10
Days Left
  • $0
    pledged
  • 0%
    funded
  • 10
    days left

About This Project

AI can produce confident interpretations of ancient writing before the evidence trail is reviewed. This project tests the hypothesis that a fixed Falsifiability Sheet will make AI-assisted claims clearer, more consistent, and easier to check. Five Egyptian control claims will be reviewed with public sources, fixed questions, and reviewer feedback where possible. Claims will receive Green/Yellow/Red outcomes, and the project will produce five public sheets, a small dataset, and a short report.

Ask the Scientists

Join The Discussion

What is the context of this research?

AI can produce confident interpretations of ancient writing before evidence is checked. This project responds to concerns raised in the NIST AI Risk Management Framework about evaluating and managing AI risk. It asks whether a simple traffic-light evidence sheet can help readers review claims across linguistics, iconography, and epigraphy.

Egyptian hieroglyphics are used as a control case because the script is already understood; the British Museum explains that the Rosetta Stone helped decode hieroglyphs. Five claims will be tested using fixed evaluation questions, public evidence, and reviewer checks. The hypothesis is that the sheet adds a practical layer to the scientific method by turning disputed interpretations into testable claim units with a preserved evidence trail. Success means the sheet clearly separates Green supported, Yellow uncertain, and Red unsupported claims. Outputs: five public sheets, a small dataset, and a short report.

What is the significance of this project?

This project matters because ordinary fact-checking often checks a final answer, while this framework checks the claim before it becomes accepted as true. In ancient writing, a polished interpretation can hide weak evidence, missing context, or AI overconfidence.

The gap this project addresses is claim-level review. Existing source checks and AI evaluations often ask whether an answer is accurate overall. This pilot asks a narrower question: which parts of an interpretation claim are supported, uncertain, or unsupported?

The new insight will come from testing the framework on five Egyptian control cases. The pilot will show where claims hold, weaken, remain unresolved, or fail, and whether a fixed evidence sheet makes disputed interpretations easier to compare, review, and correct.

What are the goals of the project?

Phase 1 will test five short interpretation claims about Egyptian hieroglyphic signs, words, or passages. The goal is not to re-decipher Egyptian, but to see whether disputed interpretation claims can be checked before they are accepted as true.
For each claim, the project will record what the claim says, what evidence supports it, what evidence weakens it, and what remains uncertain. Each claim will be checked through fixed questions, public sources, and reviewer feedback where possible.
Success will be measured by completing five public evidence sheets, assigning each claim a Green, Yellow, or Red result, and publishing a small dataset plus a short report. Green means supported, Yellow means uncertain, and Red means unsupported. The report will show which claims held, weakened, remained unresolved, or failed.

Budget

Please wait...

This budget funds the smallest serious version of the project: five Egyptian control-case tests, a public dataset, and a short final report. Project lead time is estimated at 50–60 hours across five cases. This includes claim selection, source review, evidence packet preparation, AI-restricted review, Falsifiability Sheet completion, documentation, dataset cleanup, and final reporting. Evidence collection funds support public sources, image/object documentation, comparison material, and organized evidence packets. Reviewer funds provide small honoraria for independent R1/R2 checks where possible. Dataset and formatting funds support the public archive, methods page, graphics, and final report. Platform fees and contingency cover Experiment/payment deductions and a small cushion so the project can still be completed.

Endorsed by

I am really excited about this project. It would really answer so many questions of mine and I believe this is going to be the best researcher for this mission.

Project Timeline

After funding, this will be a three-month Phase 1 pilot. Month 1: finalize the Green/Yellow/Red criteria, select five Egyptian control-case claims, and prepare evidence packets. Month 2: complete the five Falsifiability Sheets and request reviewer checks on evidence and outcome labels. Month 3: incorporate reviewer feedback where available, finalize outcomes, and publish the dataset, methods page, and short final report.

Jul 13, 2026

Project Launched

Jul 31, 2026

Select five Egyptian control-case claims and prepare evidence packets.

Aug 31, 2026

Complete five Falsifiability Sheets with Green/Yellow/Red outcomes.

Sep 30, 2026

Publish public dataset, methods page, and short final report.

Meet the Team

Michael Grasa
Michael Grasa
Independent Researcher, AI Verification and Ancient Scripts

Affiliates

Echoes of the Script
View Profile

This Phase 1 project is led by Michael Grasa as an independent researcher. Outside reviewer checks may be added where possible, especially for evidence review and traffic-light outcomes. No collaborator is required to complete the five Egyptian control-case tests.

Michael Grasa

Michael Grasa is an independent researcher and founder of Echoes of the Script, a publicproject developing falsifiability-first tools for testing AI-assisted ancient-script interpretation claims. His related work has received Emergent Ventures support and was presented at the Gyan Bharatam International Conference on “Reclaiming India’s Knowledge Legacy through Manuscript Heritage (PG 24),” held September 11–13, 2025, at Vigyan Bhawan, New Delhi, where his Indus script research appeared in the detailed conference schedule and Book of Abstracts. His work is also publicly documented through Fractured Atlas, GitHub/OpenLab materials, and Zenodo-linked worksheets. This Phase 1 project applies that same claim-record method to Egyptian control cases: one claim, evidence trail, uncertainty label, reviewer checks where possible, and a traffic-light outcome.

Lab Notes

Nothing posted yet.

Additional Information

This is Phase 1 of a larger public falsifiability project. AI is not treated as the authority. AI comments are logged as input only. The evidence trail, reviewer notes where possible, and final traffic-light outcome determine the result.

If this pilot succeeds, a later phase may expand to more Egyptian cases and eventually to more uncertain scripts such as Meroitic.

This version is stronger because the traffic-light logic is not hidden. It tells people immediately how the results will be judged.


Project Backers

  • 0Backers
  • 0%Funded
  • $0Total Donations
  • $0Average Donation
Please wait...

See Your Scientific Impact

You can help a unique discovery by joining 0 other backers.
Fund This Project