I understand that the imbalance between observers and identifiers is one of the major obstacles INaturalist faces. There are ten observers for every identifier, and a very small number of identifiers do the lion’s share of the identifying.
I finally decided to be a part of the solution rather than a part of the problem and got started on identifying.
I’ve found it to be almost as pleasant, interesting, relaxing, and intriguing as observing is. You check the observation against a menu of IDs suggested by the AI. It is easy and fascinating.
I suspect the slight clunkiness of the web design represents a disproportionately large roadblock for many possible observers. Setting up the “Identify” modality does not feel intuitive to me.
You have to go to identify, then set the filters to the time and place you want to focus on, then filter for unidentified observations, then hit “suggestions,” then change the “source” from “observations” to “visually similar.”
Once all that is done, and you start identifying, it’s awesome! Then, within about 20 observations, it says “an internal server error occurred.” So you have to stop and start again later.
If Any one of these steps could be eliminated–or if all of them could–it would be a lot easier to expand the pool of identifiers.
I love the wonky nerdiness quotient of Inaturalist, but a little commercial-style streamlining and promotion would go a long way to solving the observer-identifier imbalance, it seems to me.
I’m glad that you’ve found enjoyment in identifying, and also that you took time to share your experience of getting started with this.
I identify lots and this doesn’t happen for me. Even with searches that show thousands of observations across hundreds of pages. I would recommend fully closing your browser and seeing if the behavior stills occurs on subsequent occasions. If it does, please file a bug report via the forum.
Each person wants to identify different stuff in a different way, so choosing the time, taxon and place is always going to be necessary. I’d love to have an ability to store several sets of default options, so for example I could easily start a search for Iridaceae identified between family and genus in Mato Grosso, Brazil. One thing you can already do is set your own default location if you always/often want to ID things from a certain place.
You’ll probably find that you want to switch back and forth between “Visually Similar” (things within a parent taxon that the computer vision algorithm knows about) and “Observations” (things within a parent taxon that have already been recorded on iNat from this area).
Maybe. But the wonky nerdiness is a feature not a bug.
Added: I see you were referring mainly to how a user interfaces with the website when IDing. No objection there to streamlining some aspects if they are obstacles to people using the site.
Please don’t rely solely on the AI and comparing images when making ID suggestions for other people. This applies in particular to making fine-grained suggestions (genus or species rather than something general like “bug” or “spider”). It’s fine to use the computer suggestions as a starting point, but unless it is a taxon you already have some knowledge of, you should do at least a bit of checking off iNat to learn about field marks, whether there are similar species, etc.
Not all taxa are included in the CV model and it is pretty unreliable for certain groups. Some suggestions may look plausible for someone unfamiliar with the taxon in question, but still be completely wrong. We have a tendency to focus on things like color and patterns; this works well for some taxa, but for others – for example, many insects – color can be misleading because there are multiple unrelated species with similar patterns. Other species may be variable in color or display sexual dimorphism.
I note this because a large portion of my IDing time is spent just correcting wrong CV suggestions. It is also important to be aware that many users will agree to any ID suggested by another user, regardless of whether that user has any expertise in the taxon in question, so it is a good idea to be sure that you can support your ID with specific criteria about why it is that species and not some other.
If you are IDing unknowns, it is fine to add a broader suggestion of your own instead of something suggested by the AI.
If you want to do identifications based on mostly/entirely on CV suggestions, I would focus on using the suggests to get the observation to the family or genus level. This way, when the original identifiers inevitably agree with your ID it doesn’t become Research Grade. This will also make it easier to find by experts who can give a species level ID.
Thanks for being the change you want to see, happy IDing :)
It might help, but it wouldn’t solve the core issue of many observations not reaching RG even though they theoretically could (as in: there is enough information in the photo for a species-level ID).
From my experience of identifying both my own observations and those of others: There are, generally speaking, three types of IDs.
Broad IDs
Quick specific IDs
Slow specific IDs
The most time efficient is the quick specific ID, possible for taxa that are easily identifiable from a photo (with a few exceptions adult lady beetles of the Coccinellini tribe in Europe, for example). This allows a single identifier to work through tons of observations in a single session. Thanks to these identifiers a lot of observations reach RG status. Often quite quickly at that.
A bit less time efficient are broad IDs. These take more time because there is just simply more different stuff. So for every observation you need to focus on something else. (Additional problems are forgetting names of taxa you do know, often poorer photo quality in the observations, etc.)
However everyone will be able to contribute here in some way, and every bit helps a lot, so the pool of potential identifiers for this category is large.
The most time consuming and difficult IDs to make are the slow specific IDs. These usually require a good identification key which takes time to go through. Often they also require pre-existing knowledge of the taxa, as these keys usually use technical terms. Also, even just finding good keys can be a challenge. For these reasons, many of the people providing these types of IDs are taxon experts (often actual working scientists). Their time and energy is limited. Additionally, they probably get tagged in a lot of observations and have to correct mis-IDs. With the time it takes to make a single ID, it is no surprise that they cannot work through that many in a day.
I feel like, in this “category” of identification, the Identifier-Observer-Imbalance is most pronounced and where additional identifiers are needed most. Due to the amount of work, I cannot see a UI-change to contribute much here, though.
I second dianastuder’s suggestion. I work through the Explore tab instead of Identify, but I save a bookmark with my preferred filter options. If you have a list of favorite taxa you like to check, it might pay to save a bookmark for each one of them, rather than trying to remember that list every time you sit down to work.
I put mine in my profile (see bottom of screenshot below). It helps me and it also let’s people know what I’m interested in /check for already. Granted, I don’t have a lot of different queries I use.
I’ve had more and more “internal server errors” lately, so you’re not alone there. One consistent one is if the person has marked their observation “life” and you try to click for “visually similar” taxa, it will freeze up and push that message every time. I guess I should submit a bug report.
Speaking of bug reports, (or reporting on bugs, insects, ha ha) I’m so glad you’re identifying! It really can be quite a relaxing groove (or a fun puzzle). For the more mindless identifications, you can at least fish completely unidentified photos by making some broad classification. It feels like being back in preschool. I feel proud of pointing at things and saying, “Plant! Spider! Butterfly!” and it’s nice because giving it a broad classification like spider or snake can push the observation onto the radar of a spider or snake expert.
Another great way to start is to pick some very specific but commonly confused taxa that you learn how to differentiate. For instance, the computer vision straight up cannot tell the difference between the Eastern Comma Butterfly and the Question Mark Butterfly, but I bet you can if you take a couple minutes to learn the minor markings differences both on the outside and inside wing surfaces. Sometimes photos are blurry and you can’t narrow it down either–that’s when you want to pick the genus Polygonia which includes both species. That’s just an example–you can pick whatever species you’re interested in the most :)
I don’t like the implied gatekeeping sometimes when people want to constantly defer to a very small group of scientist/expert identifiers. Anyone with eyes (or ears, for audio recordings) can become an identifier. Like sure. I am a scientist. And I have actually helped museum collections sort out misidentified specimens in a species complex I know well. But that doesn’t make me ANY MORE QUALIFIED than a person who spent five minutes reading about, for instance, the difference between Eastern Comma Butterflies and Question Mark Butterflies, at identifying those correctly.
It is always up to you, as the identifier, it is your name on the ID, not iNat’s CV.
If you are unsure of the species, go up taxon levels to where you are absolutely confident. Easier for taxon specialists to narrow down a broad ID (but please not too broad), than to fight back against a wrong ID with the CID’s more than 2 thirds rule.
What interface are you using when IDing? (There are some minor differences depending on whether you are pulling up the observation pages from “Explore”, accessing observations through one of the apps, or using the “Identify” module on the website.)
The CV suggestions on the observation page should never include a species as the first suggestion. By design, the top suggestion is always at least genus or higher.
You are not required to use the CV suggestions. You can type your own ID in the box instead (Latin or common name in your language) and select the item you want from the list of name matches in the taxonomy database.
Yes. I’ve got at least two dozen URLs saved in my browser: some I check frequently, some daily, some that I started working on but then my energy petered out but I mean to go back to.
I’m constantly re-arranging my priorities (via bookmark subfolders) based on my interest for a time.
Thanks so much! I do use the Identify modality, though it does seem to suggest species rather than genus some fo the time. And yes, I do not always follow the suggested ID. It tried to tell me a frog was a butterfly the other day.
Yes, by [the current] design, but I wouldn’t mind about a correct top suggestion at species. For instance, this observation has a correct suggestion at species, with a vision score 99 (over 100):
At the other end of the spectrum, for example, there is this observation, with all suggestions wrong, with vision scores < 3 (the likely correct ID is a taxon not yet in the c.v. model):
