Here is an apparent common problem with CV, particularly among newbies: they think the CV list is of equally likely choices among which they should choose. The user then browses the options and picks the one they personally think looks best. Among the cicadas I’ve been identifying, if they don’t pick the first one in the list, they are usually wrong, meaning I have to override their ID. If the CV could somehow indicate confidence values for each item in the list, and let the user know that the first one is strongly preferred (if it is), they would be less likely to make wrong identifications, and it would improve the accuracy of identities considerably.
On a related note, it seems for cicada species with extremely obvious IDs, CV still just says it’s pretty sure the it’s the member of its genus, rather than saying is pretty sure that it’s the particular species. It would be very helpful if its more precise when it can be.
If the user has made any choice at all, it’s now more likely they will respond if you reach out to them. They may defend their choice, recognize it as incorrect, or ask for more information. They can learn something, and so can the identifier.
This is so much more preferable from my perspective as somebody who ID’s than someone who is in the habit of automatically picking the first suggestion. We shouldn’t encourage judgement free acceptance of the top suggestion by reinforcing that it’s often the right choice. We should encourage personal judgement on the part of the observer.
Learning is the objective, not choosing correctly.
in one of the posts in that other thread, @sessilefielder provides a link to a browser extension that will allow folks to have a visual indicator of the computer vision scores.
I think personal judgement is fine, but when they are picking something obviously wrong when they might not if they saw their choice has a 1% chance of being correct compared to 95% for the top pick, for example, there’s room for improvement in the system.
I also strongly believe iNaturalist should focus more on optimizing the identifier’s time rather than the user’s time, considering the platform is woefully short on identifiers. As a fairly heavy identifier, I wish my job was not so time consuming. Anything that will streamline that process will make me more likely to continue.
As a fairly heavy identifier myself, I sometimes get frustrated by the use of CV suggestions. However, I use it as an opportunity to teach, or if I’m frustrated by the same mistake again and again, I just move on. iNat’s main purpose is to get people to interact with the non-human world. The provision of a data base is a secondary goal. I think that identifications are important, but education is more important, no matter how much work it might be for us.