All on its own, the system had absorbed gender bias from the reams of data. Similar issues arose with race and cultural data. And what about the issue of intellectual property and copyright? Generative AI programs like Midjourney and DALL · E have trained themselves using existing artwork and copyrighted materials. Is that fair to the artists? In the UK, a government-funded research project into these very questions resulted in the creation of an“ AI Museum Planning Toolkit”, available for download. Its intent is to provide a guide in the development of ethically robust project concepts when using AI in your museum.
4
Can AI be trusted?
I recently attended a course on the ethics of AI, and my group was tasked with creating an HR application that introduced teachers to the integration of AI in the classroom. While using Claude to look for examples related to science curriculum, my teammate had this exchange with the chatbot.( right)
An AI hallucination is an incorrect, nonsensical, or entirely fabricated output caused by limitations in an LLM’ s training or understanding. A research study in April 2025 by OpenAI found that its 03 and 04-mini models hallucinated 33 % and 48 % of the time respectively( AI Hallucinates More Frequently the More Advanced It Gets. Is There Any Way of Stopping It? | Live Science, n. d.). Hallucination rates vary amongst the different LLMs, but the propensity to make stuff up is an ongoing problem. As a museum, that means being proactive in both staff education and creating
MATAHEF ISSUE 01 41