But technology is fantastic at accuracy, better than humans in many regards.
This isn't about "technology", it's about large language models, which are neither "fantastic at accuracy" or "better than humans".
Gemini might have a way to go before it gets there, but it or its successors will get there and it's moving fast.
Large language models are structurally incapable of "getting there", because they are models of language, not models of thought.
And besides, anything that is smart enough to "get there" deserves human rights and fair compensation for the work they do, defeating the purpose of "AI" as an industry.
If AI can make the services of the National Archives more productive for its staff and/or the public then surely that's a good thing?
The word "If" is papering over a number of sins here.
If the photocopier is smart enough to do a scribe's job then it deserves human rights, fair wages, and a pension just like the rest of us.