The article has a really interesting discussion of several issues with the quality of IDs on iNat that will be familiar to a lot of forum users. And it proposes some fixes!
1. Many observations don’t have photos with sufficient detail for an ID. The authors suggest providing standalone or interactive guides for identification and to assist in taking photos. Coincidentally (?) @bramb just added a great guide on How to make pictures of land snails that allow for species identification? Integrating this kind of guide into the observation and identification process, as many have already suggested in the forum, has the potential to really improve ID quality and observer knowledge.
2. The limited capabilities and coverage of computer vision and its bias towards species in the global north. The bias towards species with more observations can be effectively corrected by adding more observations and identifications for species in the global south. This has already been proved multiple times on iNat. This is where translated interfaces and partnering with local organizations can really pay off. When interested people take up the task of documenting and identifying organisms that are currently poorly covered, IDs improve and over time CV gets really good at any sufficiently frequently observed organism.
Other changes in the CV process have also helped, such as the inclusion of genera in the CV model where there are insufficient species-level observations. Where photos are not sufficient to distinguish species, CV faces a harder task (as do human identifiers).
3. The tendency of some iNat users to agree to IDs that they’re really not familiar with. These people are called “clickers” in the paper and their action are mistakenly conflated with “vandalism”. Interestingly, although the authors also describe a problem of “users who want to accumulate as many IDs as possible regardless of whether they are correct or not” (citing a 2021 paper on Italian mollusc IDs on iNat), they say it’s not something they have observed much themselves.
My experience is that competition for ID “high scores” happens a lot less than people expect, especially since iNat removed several gamification elements. The big factors that appear to drive these low-information IDs seem to be misunderstanding the meaning of the “Agree” button; confirming friends’ IDs within school projects; and the occasional tear of hundreds of dubious IDs from a young and well-meaning user getting started with iNat. In almost every case, users will respond to polite engagement so long as they’re still using iNat.
4. The difficulty in correcting wrong IDs when the opinion of inexperienced users is weighed the same as experienced academics. The authors have two suggestions: “Assigning greater weight to identifications made by curators and specialists, as well as providing tools so that qualified users can restart the identification process.” Valorizing expert IDs has been suggested a few times, but there are many reasons to shy away from this.
Certainly there are a few cases where observations languish with incorrect IDs or inappropriate “Research Grade” status because a few inexperienced users can outvote experts with 30 years of publication history. But these are a tiny minority. A more typical scenario is that an expert will add a better ID and other users will agree or will withdraw their IDs. Occasionally, someone will ask “Why do you say it’s this and not that?” and in the process they’ll improve their own ID skills. In my experience the main reasons people don’t correct their IDs are that they don’t use iNat anymore, don’t realize a new ID was added, or don’t understand how to update their earlier ID. Fixing those issues doesn’t require designating people as taxon experts.
And giving some users greater weight for ID determination has a couple of big downsides, too. First, any such process would require a whole infrastructure of assessing and delimiting expert qualifications, monitoring usage, etc. More fundamentally, it shifts iNat to a less egalitarian and cooperative structure. “Observers” would become dependent on “experts” and feel less able to contribute their own ID efforts; “experts” would be under pressure to validate IDs and using iNat would become more of a chore for them.
5. The scarcity of experts for many taxa in a country like Brazil with great biodiversity but limited science funding. The authors capture both the problem and potential solutions well here:
In this context, Brazilian reality imposes some difficulties, starting with the small number of experts, who are often too overloaded with activities to be able to interact with the platform. The contributions of experts from other Latin American countries and elsewhere were relevant to verify several observations analysed here, which also highlights potential benefits of the platform as a place for information exchange between professionals, networking and establishing collaborations.
In addition, if experts active on iNaturalist take the time to explain how a given species was identified, this can provide a valuable experience for community scientists. With time, community scientists will achieve a similar level to experts; iNaturalist already counts with many of these “expert observers”, who are community scientists with ample knowledge of particular taxa.
It can seem excessive to ask already overtaxed specialists to volunteer time to patiently correct the identifications of a legion of casual observers and citizen scientists. But those observers and identifiers include the next generation of professional scientists along with plenty of people interested to become amateur experts. Time spent communicating ID tips through iNat now can result in a much stronger base of knowledge for these taxa in the future.
I’ll be presumptuous and share the two things I feel would really improve ID quality on iNat:
- Improve the user interface so that inexperienced users have better guidance on what to photograph, what characters to look at for an ID, and what level of ID is appropriate.
- Better notifications so people know when other people have suggested different IDs and how they can respond.
I know a better notification system is under development (can’t wait!) and I really hope someone has put a guided user interface concept on a slide that will be seen by a funder. Not simple or quick but it could be hugely beneficial.