Prepared and developed by Digital Al-Jazari for AI & Research Innovation. Original material developed from the supplied vision; numerical examples are fictional.
An assistant does not become a witness to the past
AI tools may organise notes, suggest questions or compare wording. They do not witness visits they did not attend, measure channel loss from a persuasive description or give a photograph a date or outcome absent records. Serving heritage begins with preserving evidence and making it reviewable. Local knowledge, documents and measurements remain sources of judgement rather than inputs replaced by a convincing response.
Three layers that should remain distinct
First is observation: a channel or meeting appears in an image. Second is interpretation: the observation suggests a function or question. Third is a claim needing proof: a channel performs at a specific efficiency or a project succeeded. A tool can help articulate the second layer; it must not jump to the third without evidence. Our notebook labels each layer and places sources beside facts rather than blending them into prose that conceals the distinction.
Worked example: image record or performance report?
An institution-supplied image for the fourth innovation shows people near water. A sound caption is: “Waterside scene supplied with Northern Jordan Valley visit material; exact date not recorded.” An unsound one is: “Our team proves project success and increased residents’ income.” The latter adds work, outcomes and income not established by the image. Names also need records; appearance does not establish professional roles or need.
A workflow you can use now
Create a record with: material ID, visible content, source, established date or “unspecified”, proposed interpretation, verification needs and human reviewer. Write your description first, then compare any automated assistance with it. If it adds a number, name or certainty, move that addition to verification. Review image permission and privacy before uploading to an external service; sensitive data should not be uploaded merely to improve wording.
Activity: correct one claim
Start with the fictional claim “The photos show the system saves 40% of water.” Ask: where are before-and-after measurements, the denominator, dates and loss accounting? Correct it to: “The photographs show elements of the water route; estimating savings needs comparative measurements.” Open the water notebook to distinguish an assumed percentage from a measured result. The calculator uses your inputs; it does not train on them or generate new knowledge.
What actually works in this lab?
The four applications run browser calculations and rules without an external model. They produce notebooks, comparisons and plans that can be reviewed, copied and printed. They do not automatically analyse images or claim access to unavailable field data. For prompt-writing and verification learning we link to ANAwAI and Human-Centered AI. The lab’s role is to expose an idea to an understandable check while preserving the place of evidence and human review.