I understand your point and I think it’s an excellent one to bring to discussion. It’s important to take it into account.
I personally would say we should never leave that for AI to do (the supposed alt text, because AI could definitely hallucinate.
On the other hand, if a person could provide textual descriptions, and if other users could confirm or deny these textual descriptions (much like we can agree or disagree with a taxon), then the evidence that the petals are actually petals would be strenghtened. Perhaps the alt text wouldn’t also have to be simply “alt text”, but maybe iNaturalist could add a section to add notes about what it looks like:
I would say that most importantly, the first thing to have alt text needs to be the page of specific species and taxons. That is immediately useful for every user. Plus with a little glossary of anatomy of the animal/flower/fungus for anyone who doesn’t know. This helps immediately with the step about comparing my flower with the flower the ID is suggesting.
If alt text seems to not be the right step for now for user-created photos, then at least we should add a field for “Visual, textural and «fragranceal» description of what I saw”, encouraging users to describe what they saw. With a guide to help on it.
(Edit: Helpful guide about how to take notes about a birds’ behaviour to help identify it: How can I identify birds when they’re flying? )
Note also that these days, screen readers can use AI create their own alt text of a picture for pictures that don’t have it and aren’t marked as decorative, which is immediately worse than a real person doing the description because AI hallucinates. Therefore, there is a bigger need for this to be parsed by users now that a blind person would really have to rely on AI (which, like you say, could think the petals are a butterfly).
This makes me agree with you that if a flower is graded, it doesn’t mean it’s the flower, so automatically adding AI alt text presuming it is a poppy with crumpled red petals is overriding the ability of humans to confirm whether it is or not. So I don’t think anymore that this specific suggestion I made is a good suggestion.
I also reiterate my suggestion of other users being able to confirm the description is correct, and add to it.