ID Tool and Persistent Errors

The appears to be an issue, concerning at least one species, with the website identification suggestion tool consistently suggesting a common, variable, plain-looking species for a large variety of similarly dull but often quite unrelated species. I’m posting to see whether there is a technological fix to this issue.

I have over the last several months corrected easily 200 misidentified Hofmannophila pseudospretella from eastern North America. I have a hard time believing that at least a few dozen contributors all happen to have misidentified moths in several families, many of which look very different from Hofmannophila, as this species, which is evidently restricted to the west coast. The most likely explanation seems to be that iNaturalist often provides this species as an ID suggestion for many poor photos of brown moths. Despite my best efforts to “teach” the ID suggestion tool by correcting all its errors in this case, at least several more images masquerading as Hofmannophila show up every week. This is especially frustrating when such misidentifications make it seem as if the species is widely distributed throughout North America unless errors are corrected immediately. Has anyone noticed a similar problem with other species?

1 Like

You highlight the problem yourself: ‘similarly dull’. The computer vision does not (and cannot) account for whether species are closely related, it’s purely based on morphology. So if two things look similar, the computer vision will probably get them mixed up. This is a common problem to varying degrees across different taxa. For example, the computer vision struggles with taxa like flies or molluscs.


Yes, and it’s a constant educational process to help users understand the limitations of the “Computer Vision” suggestions. In a nutshell,

  • Not every species in iNaturalist is “known” by Computer Vision.
  • Only species above a certain threshhold for number of Research Grade observations are included in the Computer Vision data set. (This biases its effectiveness toward areas of the world where iNaturalist is in greatest use, and toward more commonly observed species and groups).
  • Computer Vision will always offer suggestions based on similarity, even if the actual species you have is not in its data set.
  • Computer Vision is not (yet) geographically “aware” enough to exclude suggestions from different continents, much less from the wrong side of the same continent. This is being worked on though…

That said, it is amazingly effective for certain taxa and regions, and getting better with every observation we add and identify.

But it’s use should always include an independent reality-check and a grain of salt.


Thanks. It sounds like this problem will continue until some software update fixes it. I just find it strange that the program happened to fixate on this one species despite an error rate of at least 30-40%.

1 Like

Probably because observations of that species dominate the available iNaturalist data set.