AI usage declaration

[juliajbeine]

We may discuss how to deal with AI usage when writing the encoding guidelines. If we use AI to revise and add to the encoding guidelines, such as this passage related to #59, I would suggest adding an AI usage section at the end of the guidelines that gives the AI model used and the sections for which it was used (for inspiration, see the GAIDeT taxonomy). This is the recommended standard for scholarly documents, which I would consider the encoding guidelines to be. I have seen that the information is given in the commit for the aforementioned passage, but I think declaring the AI usage in the document itself would be good since some colleagues only use the front-end version.

In any case, I think linking to the AI model in the commit would be a good standard. Following a pragmatic approach, we could also add a sentence to the document itself, such as
“AI usage
For writing and revising this document, generative AI was used. For details, see the git commits on GitHub.”.

In that case, we would not need to continuously revise the AI usage section in the document, but follow the standard to declare AI usage in the commits.

What do you think, @lehkost, @cmil, @ingoboerner?

[ingoboerner]

because I am tagged and this is dear to my heart: IMHO Code being written by AI has become the standard, not an exception in the last months. Same is true for documentation. I would not tag that on each and every single incident. A disclaimer on the level of the ODD/Readme might be a way to go in this transitional phase. Regarding "colleagues only use the front-end version": I personally think this wild hold true for a couple of months, but if we switch to agentic encoding workflows, the primarily consumers of the encoding guidelines will be the agents anyways...

[cmil]

I'm almost completely on the same page as @ingoboerner on this. I don't think declaring AI usage in detail is very practical or even useful. Will it matter in a year which model helped in creating a certain git commit? Nearly none of my commits in the last weeks was not assisted by some AI tool, mostly Claude. The change I think @juliajbeine was originally referring to, was one of the rare occasions where I let Claude directly do the commit and it shamelessly used the opportunity to advertise itself. Usually, while often letting Claude draft those technical texts, I push the buttons myself reserving the chance for last touches to keep the typical AI verbosity at bay.

I do think, however, the primary target audience for the Encoding Guidelines should be human readers. Because, only if we understand our choices and decisions in designing the schema, we can catch and correct the mistakes AI will definitely make for the foreseeable future.