Everything will depend on the implementation details, and I think this community could have a very constructive role in suggesting those details, which the announcement makes clear have not yet been developed.
For example, I could envision an implementation where a LLM run will initially populate an editable tab on the taxon pages with identification hints derived from any existing user-supplied comments for each taxon. Initially the raw generated text would be visible only on this tab, and be prominently marked with a disclaimer “raw unreviewed AI-generated content” or some such.
Before it could be used anywhere else on the site (or even be visible off-site?), it would need a vetting process similar to that for observations - at least two humans, comprising more than 2/3 of the total votes, up-voting the accuracy of the content on the taxon tab. If implemented well, hopefully there would be many cases where the text could be up-voted as-is. But otherwise, humans would review and edit the generated text, and each edit would notify those with existing votes that (further) edits had occurred. Edit history would be preserved and reviewable, and revertible by curators.
Taxa with blank tabs (no relevant information yet found by the LLM) could also be populated directly by humans without waiting for the LLM.
Taxa with existing human edits/votes would be unavailable for further updates from the LLM, but those whose text has no votes or edits (or is blank) would be replaced by subsequent periodic LLM update runs (and still only be visible on the taxon tab with the disclaimer).
Taxa with insufficient expertise and/or attention to garner the required threshold of upvotes for their ID text to become usable with CV or other areas of the platform would continue to have their text visible on the taxon page, but only there.
There could be a button on the tab to run the LLM on-demand for just that taxon, which would either replace existing raw or blank text, or display the results in a separate window to compare with existing edited/voted text.
That’s just off the top of my head - there are probably more elegant and/or useful implementations to be dreamed up. But it’s an example that keeps the results user-based, with AI used only as a tool to assist users in creating content helpful to the community.
In the end, having been here for over 10 years now, I trust that the original creators and founders of iNaturalist (who still lead the organization) know as well as any of us the unique and wonderful attributes of the platform that are valued by themselves and the whole community here, and that they would have no intention or tolerance for compromising those attributes. I think everyone involved in the heated reaction to this announcement needs to take a step back, seek some perspective, see what evolves, and look for constructive ways to help it enhance the things we value here.