Questions / open one at a time, or several
Beside-the-Label Questions
The useful questions are often modest: what changes, who decides, and can I say no?
Does adaptation require generative AI?
No. Reviewed branching content, language choices and visitor-controlled depth can adapt interpretation without generating new prose. Start with the smallest mechanism that meets a real visitor need. Explore the interpretive layer.
What should a system personalise?
Prefer explicit, changeable choices such as language, format and depth. Avoid inferring identity, disability or knowledge from behaviour. A choice made for this object does not have to become a permanent profile.
Can an AI answer be trusted if it cites a source?
A citation helps only if the source supports the specific claim. Retrieval and generation can still introduce errors, so review and testing remain necessary. Check that a citation belongs to the object and supports the qualification as well as the headline claim.
How should a system handle a question it cannot answer?
State the limit clearly, avoid invention and offer a relevant source, a related question or a route to museum staff. “The record does not establish that” is a useful answer when it accurately describes the evidence.
Can visitors decline personalisation?
They should have a stable default route and a useful non-digital alternative without being required to create a profile. Declining optional data collection should not remove access to the essential object account.
Does longer screen time mean better learning?
No. Time alone cannot distinguish learning from confusion. Combine focused evaluation with voluntary feedback and attention to the object encounter. Someone who leaves the screen quickly may have found exactly the prompt they needed.
Is automatic translation enough for access?
No. Review important terminology and meaning, and provide other access routes. Translation, plain language and Easy Read are distinct services. Read about opening more ways to engage.
What changes when children use an installation?
Use age-appropriate explanations, minimise collection and assess safeguarding and data protection obligations. Do not treat every child as an independent account holder. Plan for families sharing a screen and avoid prompts that invite unnecessary personal details.
Who should approve the answers?
An accountable museum editorial process should involve relevant curatorial, education, access and community expertise. A software supplier should not silently determine institutional interpretation. Read about curatorial responsibility.
Does this website run a museum AI assistant?
No. Wafflerocketznd is an independent information site. Its contact tool prepares a local plain-text note and does not send a message. You can inspect, copy or download the note before deciding whether to share it.
A question missing from the label? Write to the reading room.