Sometimes, a species 1 is many times misidentified as a species 2, although species 1 and 2 are easily distinguishable.
For instance, reviewing the “Cultivated” occurences of the genus Senna, I find many Cassia leptophylla misidentified as Senna didymobotrya, although Senna didymobotrya shows a unique inflorescence of dark buttons and yellow flowers, that is in no way similar to the inflorescence of Cassia leptophylla:
Many Cassia leptophylla misidentified as Senna didymobotrya:
Many Cassia leptophylla misidentified as Senna multiglandulosa:
It might be due to people following a wrong AI automated suggestion.
I guess I might be inherent to a collaborative work: we are prone to be [too much] confident in other people opinions, without checking anything. And once it has started, it gets amplified, like an epidemia, because we get more and more confident when more and more people share the same… error.
In the example above, there were initially very few observations correctly identified as Cassia leptophylla. The recurring misidentification phenomenon may have been triggered by the fact that Cassia leptophylla may have not been observed at all, or may have been misidentified from the beginning.
Lack of reference (initial state) → first error (random event) → error amplification (deterministic process)
It’s not the first time I see such a recurring misidentification.
I think this is normal, and not really a problem, just how the system will work and improve.
You can see if the identification was one suggested by the AI - there’s a little symbol that looks like a down-pointing arrow with stars around it that appears when that’s the case. And it looks like you’re right, many of the misidentifications you found were AI suggestions.
The AI doesn’t learn constantly, it gets updated periodically. If one of your species didn’t have many identifications in the system when the AI was last trained, it will generate a lot of misidentifications. You are finding and correcting them - that’s great. It means the next time the AI gets trained, it will have more accurate information to use, and there will be fewer misidentifications in future.
In the meantime, keep in mind that without these misidentifications, the plant might not have been identified at all, or only as “Kingdom Plantae”. That’s a very big group to sort through. By misidentifying it to the correct family, it allows people who know the legumes to more quickly find it and correct it. Which is just what you’re doing.
People often complain about the AI misidentifying things, but it’s not intended to provide an authoritative id (at least one human has to agree before it gets research grade). It’s more of a first approximation that gets improved or corrected by humans.
@jeanphilippeb, here is how it looks on your first example:
Thanks a lot! Interesting feedback.
I agree with you. It is very good to have a first ID in the correct [sub]family, and the AI makes an important contribution.
I posted this for analysis and discussion, not for complaining or requesting a fix.
I didn’t know about the AI icon. It is worth noticing that.
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.