Project summary:

AI can now generate polished interpretations of ancient inscriptions and artifacts in seconds. But a second reviewer often cannot tell what evidence was used, what uncertainty remains, or whether the claim can actually be checked.

The bet:

AI is getting better at giving answers that sound confident. But confidence is not proof.

This project tests a simple but important problem: can AI-generated claims be checked against real evidence before people trust them?

I am starting with Egyptian hieroglyphics because Egyptian is already understood. That means AI can be tested against known answers instead of guesses.

If AI cannot reliably handle known Egyptian claims with evidence, then it should not be trusted with harder ancient-script claims without a verification layer.

This project builds and tests that verification layer.

Example: one Egyptian claim tested through evidence, reviewer checks, AI comments, traffic-light labels, and one recorded result.

That is the gap this project addresses.

I created an original one-page prototype called the Falsifiability Sheet. Think of it as a claim-record scorecard for ancient-script interpretation. Each sheet takes one interpretation claim and records the evidence, reading decisions, AI comments, reviewer feedback, and final traffic-light outcome: Green = supported, Yellow = uncertain, Red = unsupported.

This pilot focuses on Egyptian hieroglyphics because Egyptian is already deciphered. That makes it the right control case. I am not trying to re-decipher Egyptian. I am using Egyptian as a stable testing ground to see whether the Falsifiability Sheet can evaluate claims clearly, repeatably, and publicly.

Example: one Egyptian claim tested through evidence, R1/R2 review, restrained AI analysis, traffic-light logic, and a recorded outcome.

The pilot will produce completed Egyptian test sheets, R1/R2 reviewer feedback where possible, AI traffic-light labels (Green = supported, Yellow = uncertain, Red = unsupported), a public dataset, and a short report showing which claims held, weakened, remained unresolved, or failed.

The goal is simple: Do not just trust an interpretation. Test it.

Large AI organizations often work on policy, polling, or model safety.

• This project works at the claim level:

• One claim.

• One evidence trail.

• Two reviewers when possible.

• One AI record.

• One public result.

  
  What are this project's goals? How will you achieve them?

Goals:

• Test whether the Falsifiability Sheet can evaluate interpretation claims using Egyptian hieroglyphics as the control case.

• Produce 10 completed Egyptian test sheets.

• Publish a public dataset showing which claims held, weakened, remained unresolved, or failed.

How:

Select 10 Egyptian hieroglyphic cases using already-deciphered material.

Complete one Falsifiability Sheet for each case: claim, evidence, reading decisions, reviewer notes, and outcome.

Upload the completed sheet/photo for AI analysis.

Keep AI restrained to the fixed evidence packet: image, claim, evidence, notes, and uncertainty labels.

Apply traffic-light labels: Green = supported, Yellow = uncertain, Red = unsupported or overconfident.

Run R1/R2 peer checks where possible.

Publish the completed records as a public dataset and report.

How will this funding be used?

$7,000 – Project lead (analysis, case preparation, documentation).

$3,000–$4,000 – Independent reviewers (R1/R2).

$1,000–$2,000 – Publishing, dataset preparation, and materials.

Minimum ($12,000): Complete 10-case pilot with full dataset and report.

Goal ($15,000): Expand to 12–15 cases, increase reviewer coverage, and strengthen dataset quality across a four month project timeline.
Who is on your team? What's your track record on similar projects?

Project led by independent researcher: Michael Grasa.

Previous work and public outputs:

Zenodo (DOI-linked research archive): https://zenodo.org/records/18518231

LinkedIn (public-facing updates and engagement): https://www.linkedin.com/in/michaelgrasa

Github ( AI ): https://github.com/mlge9900-crypto/echoes-of-the-script-openlab

This work builds on prior support from Emergent Ventures (Mercatus Center, George Mason University) and was presented in an ASI-linked conference environment, with proceedings forthcoming.

The falsifiability framework has been developed through multiple iterations (V1–V5) and includes structured peer review (R1/R2) and AI accountability tracking.

Who is leading this project?

This project is led by Michael Grasa, an independent researcher working at the intersection of ancient-script interpretation, AI verification, and public falsifiability tools.

Michael presented original ancient-script research at a 2025 conference in New Delhi, India, with proceedings forthcoming. That conference-stage work began as Version 1 of the method and has since developed into the current Version 5 Falsifiability Sheet.

He is an Emergent Ventures fellow, with prior support from the Mercatus Center at George Mason University. He has also been nominated for the Mozilla AI Fellowship.

The project is built around a simple public method: one claim, one evidence trail, reviewer checks where possible, AI comments, and one recorded traffic-light result.

This is not a large organization. That is part of the advantage. A small independent pilot can move faster, cost less, and produce a clear public result.

Nonprofit/fiscal support is also available through Echoes of the Script for partners or donors who prefer a formal route.

What are the most likely causes and outcomes if this project fails?

Risks:

Independent reviewers may disagree.

Some cases may remain unresolved.

The method may need revision before wider use.

Outcomes if unsuccessful

The project will still produce a public record of what worked, what failed, and where the method needs revision.

The goal is not to force success, but to produce transparent, testable outcomes.

How much money have you raised in the last 12 months, and from where?

Approximately $12,000 received through Emergent Ventures (Mercatus Center, George Mason University).