Flag for Taxon: Māhoe (Melicytus ramiflorus)
| Flagger | Content Author | Content | Reason | Flag Created | Resolved by | Resolution |
|---|---|---|---|---|---|---|
| treegrow | sea-kangaroo | Māhoe (Melicytus ramiflorus) |
Endemic to New Zealand but lots of out of range observations due to iNat autosuggestions |
Oct. 19, 2019 04:43:19 +0000 | cwarneke |
See comments on this flag. |
Comments
Thanks for the heads up on this. There are a few other NZ plants that I routinely sweep across the globe and clean up iNat autosuggestion mis-identifications. It looks like this will be another one.
Another good reason for having a country filter on the autosuggestions!
Yes, most of the time "seen nearby" is on by default (to an area smaller than NZ), but it sure doesn't work that way all the time.
Does that apply to the website, the app, or both? It works hardly any of the time, creating a load of ongoing unnecessary work to clean up after the AI!
It works for me on the web all the time. By default it only lists nearby species followed by "Include suggestions not seen nearby". One trick is that once that option is pushed, it will do what it says it will do, and problems can follow.
It seems to work OK for me on both website and app, though I'd want to test it for cases where there are no nearby suggestions (does it then go global by default?)
At any rate, the large number of far away autosuggestion mis-identifications coming through, somehow, poses an ongoing problem. Maybe users are too glibly opting for "Include suggestions not seen nearby"? I'd suggest a country filter that cannot be overridden. New country records should not be made by autosuggestion, it is just more trouble than it is worth and there will be very few genuine new country records, but lots of misidentifications.
If no visually similar taxa have been seen nearby, the inat suggestions will switch to visually similar taxa regardless of where they have been observed (see help doc here). So areas where there isn't a lot of iNat activity are vulnerable to having autosuggestions for out of range species. And as soon as one out of range misidentification goes to research grade, the species will become "seen nearby" for the area leading to more misidentifications. Cleaning up species maps and doing some maintenance regularly really does help cut down on the number of misidentifications. I have seen this work for several species that used to be overrun with out of range observations but now only get the occasional misidentification which is then quickly dispatched by a dedicated "maintenance crew".
@treegrow @jon_sullivan Yes, but the need for a "dedicated maintenance crew" is an unnecessary bad consequence of an AI that could quite easily and IMHO should have a country filter on it. My "dedication" to cleaning up after it for N.Z. insects only stretches so far and I won't be around forever. With few other N.Z. entomologists on iNat, as soon as I stop cleaning them up, they will quickly spiral out of control with nobody to stop them. The number of genuine new to N.Z. records on iNat is very small (maybe 6 or so per year) and mostly obscure insects or fungi which the AI has no data about anyway. A genuine new to N.Z. record should not be added by AI suggestion. It can still be added manually, so a country filter on the AI wouldn't prevent genuine new to N.Z. records being added, it just prevents the continual stream of bogus records generated by the AI. Out of range misidentifications should not, of course, get to Research Grade! If they do, then the species will be included in "seen nearby" (or my proposed country filter) until the misidentification is corrected (and that species returns to zero observations for the country in question), but we hope the misidentification does get corrected fairly quickly and it at least requires two errors to get into that scenario (initial misidentification followed by agreement to RG), so the number of cases should be manageably small compared to the larger number of initial misidentifications due to the AI. The algorithm for a country filter (e.g. N.Z.) couldn't be simpler: if there are zero observations from N.Z. for a species, then the AI doesn't suggest that species. Very simple and would solve a lot of problems, without any significant side effects. As I said, new to N.Z. records should not be the result of AI suggestions, but added manually instead. They might still be misidentifications, of course, but the point here is to stop a constant stream of automatically generated misidentifications that raise false biosecurity alarms as new to N.Z. records of exotic species.
This would leave countries that are not yet well covered by iNat observers with little or no autosuggestions. If you're somewhere in Africa or South America, it's useful to have visually similar suggestions for out of range species, because they can at least get you in the general ballpark of what you may be looking at. The problem is that there's no guidance for people on how to use and interpret these out of range suggestions.
The solution to that would be for the AI to go up a level if there are zero observations from the country and then repeat until there are observations. As a hypothetical example, the actual visual match might be for Aus bus (zero obs), so go up to genus Aus (zero obs), then up to family Aidae (nonzero observations) and so suggest Aidae.
An alternative solution would be to do it on a case by case basis. So, after consultation, staff could permanently disable zero observation autosuggestions for N.Z., but leave it enabled for Mongolia or wherever.
Ok, I went through and cleaned up all the observations of this species outside of its native range. I also added a common name in Spanish ("Mahoe de Nueva Zelanda") which will hopefully reduce the number of mis-IDs in the Americas (most of which were in Spanish-speaking countries. Mahoe is an accepted common name in Spanish and adding the bit about NZ will let people know that it is from there, rather than from the Americas.
I'm going to close this flag. Hopefully someone can periodically clean up mis-IDs of this species. As for discussions about how the AI should or shouldn't work, [the iNat forum] is a better place for those discussions, as the people with the power to change the AI are more likely to see them there.
Not sure if this is cultivated outside of New Zealand, but most observations elsewhere are probably misidentifications.