I click in the name field as the first step of supplying an initial ID, either on the mobile app or on the website.
Before I start typing, the computer vision results appear. I scan them quickly to see if I can select the correct ID without any further typing, but alas, the species or subspecies I have in mind isn’t in the list, though some related species and possibly higher-level taxa do appear.
I start typing, and the almost-right computer vision results disappear, replaced by a list of choices that match what I’ve typed so far. Most or all of these are organisms of completely unrelated types.
I either have to type the entire name of the taxon I have in mind, or type a few characters, then pause and wait for the suggestions list to update, then perhaps type a few more characters, then pause and wait again, etc.
If the autocomplete list took the knowledge from the computer vision into account, then step 3 and/or 4 could be greatly reduced.
Here’s a real example I just encountered.
My photo is of Uma inornata, the Coachella Valley Fringe-toed Lizard. I click in the “Species name” field and get the following list from computer vision:
One moment ago, the system was almost certain that I was looking at a lizard in the Phrynosomatinae subfamily – true! Now that I have given it more information, the system forgot all that, and is offering me a bunch of friggin’ plants, wasps, and slugs! Slugs, I tell you!
The autocomplete list is not in an obvious order (e.g., not alphabetical order), so I guess it’s sorted by likelihood-of-match based on number of observations or some such. It would be a much more useful list if the computer vision guesses strongly influenced the sort order.
“Uma ino” should be enough to shortlist your species. First three letters of each part of name is usually enough. To my thinking, AI suggestions are for those that don’t know the name or for those that want to not wear out their keyboards too fast… Typing in and accessing the auto-complete is for those that know the name or in the process of learning them (reinforcement through use).
i can only tell you what happens to me fairly often:
1. I click in the name field as the first step of supplying an initial ID, on the website.
2. Then i wait 10 to 15 minutes until the computer vision results appear. And nothing that i type in during this time is accepted. I have slightly more than 1008 observation, this times 10 minutes wait ... that makes 7 days of waiting until now.
@mreith Then you have a happy internet connection! Or not, as the case may be… while you wait for AI/CV list, starting to type a name should immediately change to fetching that list for auto completion. Bang in the 3+3 letters, wait for your happy internet connection to do its thing, and then pick your taxa from the short list provided… I can’t see any happier way than that… Making the list building more complicated would surely only make your happy internet connection more happy, no?
I understand intellectually that the computer vision list and the autocomplete list are two different things that are constructed through entirely different code paths. But as a user, it is both weird and inconvenient that the presented list gets worse when I supply more information. Of course I wouldn’t want to make the autocomplete list significantly slower to appear, but I have no reason to believe that incorporating the computer vision results (assuming they are already computed) would make it significantly slower. Maybe it would, maybe it wouldn’t. I don’t want to rule out the idea based on fear that it might.
The “Uma ino” case wouldn’t require too many typed characters, in this particular case. Most genera are more than three letters though, and I have run into this on many occasions. I imagine others have also.
for Hemiphaga novaeseelandiae, typing “hem nov” puts it at top of list. Typing “he no” gets it to 6th on the list. For MOST species just 3+3 is enough, but 4+4 will make it top for nearly all species. It works on vernacular names too, so I am not excited when I see bulk additions of vernacular names that all seem quite similar, as it just muddies the search results.
And before anyone whacks this for being off-topic, it is an alternative and/or workaround for the feature request… and therefore directly speaks to the level of need.
It works on project names too… I had a project “In My Back Yard”, and for a while I was able to add obs to that project just by entering “I M B”.
Sometimes if you type it too fast, or if you type too slow (ie I don’t really know why!) it seems to not send for the shortlist. In those cases (where it doesn’t come back in the expected timeframe) you can give it a g-up by adding a space to the end of it, which triggers the request.
I still think the current behavior to a non-technical/non-knowledgable observer is confusing and mysterious and non-helpful, but since there is a nice workaround for knowledgable users, it is less important. I guess in the name of not overloading the system with feature requests that are unlikely to be addressed, it would be fine to close this one.
Maybe we can keep the suggestion open but come from another angle: while you are requesting that the auto complete to take the CV into account, what do you reckon about going the other way round and having the CV take the auto complete into account? So when it offers “Phrynosomatinae” and you click on it, the CV now limits its comparisons to all the spp that fall under that Subfamily? Like wise if you type in Uma the CV looks at the best match from the Uma spp and ignores everything (even what it thinks are better matches) that is not an Uma sp.
