PlantNet, iNaturalist’s main competitor in the automatic plant ID published all its observations - 6 million with automatic IDs and 0.6 million verified records.
Thereby, there are 6 million records in GBIF which potentially were not checked by humans at all. Of course, there is a disclaimer in the description of the dataset and it says that only high-confidence data were used.
This is neither good nor bad. Just today’s reality. Hope, iNaturalist won’t do like this.
There will be more such datasets coming up. If there is no easy way to filter data on automated identification from GBIF searches, then it can really become a problem. But there are of course a lot of similar problems already from vegetation surveys, bird sightings, etc. where there is limited verification of records. I think, a similar search option to the “basis of record”, giving more details on the identification, could help.
There are also some other issues, like coordinate precision and not proviing the photos. I hope they will improve on these things. (I actually like using Pl@ntNet for IDs - it is often very helpful.)
The iNat dataset in GBIF is in many ways a positive example, with good documentation, only providing research grade records and easy access to the photos for verification by the user.
