Could UK spider recorders be encouraged to select more realistic species?

You need to clear old records first, so cv won’t suggest foreign species as seen nearby, there’s a request as I remember to propose higher levels more often, you can look it up if you want to vote, for now you don’t have to write anything, but that way nobody will knew their mistakes, you can avoid it with old records of non-established users and stick to new ones, but as ider you will always have a lot of potential work even if system will never propose species from other continent, people will still click on those species they prefer as an id.

6 Likes

Do you mean that algorithms should be developed to do this? Who will develop the algorithms to accomplish those tasks? I can’t imagine that writing an algorithm capable of “pointing out where species-level IDs are unlikely” would be an easy task.

3 Likes

The number of UK spider species that can generally be identified to species-level from a decent photo is fairly low. I don’t see why it should be so difficult to code this in somehow.

2 Likes

For every country in the world? Or should UK be given special attention for some reason?

2 Likes

Are you just trying to be difficult here? I only look at UK records. I didn’t think it was an unreasonable suggestion.

2 Likes

You ask for something unreal, there’re thousands of species worldwide, you need to code somehow for the system to know for sure it’s not idable for all of them, but by clearing all the wrong ids you will reduce the amount of new wrong ones, they won’t show up as seen nearby, as well as iding e.g. Enoplognatha complex, so it is suggested more often and species observations will be lower than now.

2 Likes

Not trying to be difficult at all. Just imagining the challenges to implementation! iNat is a global platform, and I suspect that the rest of the global population might be upset if a problem is solved for only one country.

iNat is a platform primarily for citizen scientists. The best way to nurture citizen scientists is to communicate with them as if they were a group of friends or colleagues. Providing resources to aid identification, links to resources, and developing a following is the best strategy.

Education is key here…engagement. Correct IDs on every observation would be ideal, but having some incorrect IDs doesn’t ruin things. Scientists shouldn’t be using unverified records. If they want to use iNat data, they can verify IDs themselves (I’m a scientist who uses iNat data, and that’s what I do).

I think it’s great that you’re helping with IDs! Spider experts are low in number and highly valued. But education and engagement with the community is key to how, I think, this particular platform was intended to function best.

10 Likes

The answer? Impossible. Every week I clean up the European records of Niebla - lichen genus that is found only in western North America. And the records are coming not only from newbies. I am starting to recognize some names who repeat and repeat the same mistake.

2 Likes

UK diptera were a total mess a year or two ago, and I’ve expressed similar frustrations on the forum in the past. But after a lot of input from both UK observers and identifiers, and then in turn the latest model update, there has been a radical shift in accuracy of CV IDs and the accuracy of initial IDs by the average user. There is some light at the end of the tunnel I think, so don’t despair! :wink:

There has been a recent shift to taking geography into account in CV suggestions, but I think @tiwane said its still being ironed out in Seek and Android(?). There might also be older observations which factor into this atm - either as they need fixing so the system doesn’t think this species is nearby, or if you are going through observations from before the date of implementation.

I agree there are interventions which could be implemented to limit how frustrating this can be for identifiers. Always offering genus or family level IDs when autosuggest is unsure would make a significant difference I think - this - is the feature request @marina_gorbunova referred to maybe.

Regular problem taxa can also be flagged up on the CV clean-up wiki - so other users can help keep tabs on them. One success in this regard is with Sarcophaga carnaria which was massively oversuggested and in constant need of fixing. But the diptera identifiers have kept it at bay and its now no longer being autosuggested outside of Seek.

10 Likes

There are various design interventions I can imagine which could change how users interact with this aspect of the system. Broad scale algorithmic approaches to resolving these issues sound complex, but smaller tweaks to the user interface could also help resolve the problem.

e.g.
Couldn’t an annotation system be crowdsourced somehow ?
For example by having an option on the taxon page where users can add a flag to specific taxa to beware of X or Y :-

Screenshot 2021-07-25 at 16.47.27

…and this in turn triggers a warning symbol on that taxon in the future :

Screenshot 2021-07-25 at 16.51.32


Yes I agree, it’s not just new users. Indeed, I might be one of the observers in question as I am totally clueless on spiders - apologies if so - the elephant path of the autosuggest is too attractive sometimes.

In general @mhiggott, I think you are one of the first active identifiers I’ve seen on UK spiders, so it will likely appear an uphill struggle at first but get easier as time goes on. If a clueless set of hands is useful somehow in clearing the backlog though, feel free to rope me in :)

7 Likes

Doubt you can be a user who’d be mentioned in such context! Though I also know users with 20k+ obs that blindly click on cv suggestions even when it’s obvious they’re completely wrong or with common species they observed so many times it’d be logical to learn them already (and our flora, for example, is not tropical, so common species are mostly “one of a kind” and are learned pretty easily).

2 Likes

Well… I take care in insects as I know roughly where the traps are.
I try to take care to leave coarse IDs in lichens due to @jurga_li’s previous thread discussing similar concerns. But in spiders I have very little idea tbh, so I am certainly guilty of being more careless at times.

3 Likes

In a different topic I had formulated the suggestion of a ‘self-reflecting’ AI - meaning it takes into account if the CV suggestions have subsequently been disagreed with. If for a certain taxon the proportion of mis-IDs surpassed a certain threshold, then the CV will be more reluctant and rather suggest higher level taxa.
Haven’t heard from the developers if something like this is possible to code into the existing model

7 Likes

Do you comment that the genus isn’t found in the country? I found after a couple of times (for me it was Condylostylus, an American genus of flies not found in Australia) individuals stopped repeatedly IDing it (although it didn’t stop happening from new IDers until the iNat algorithm changed to stop recommending it). My experience may just differ to yours, though.

5 Likes

I wish there were a way to stop some of the common errors in plant identification, too. I don’t think there is. I do maintain a list of common explanations to paste into my corrections, and sometimes the observers indicate that they help – when I can summon the patience to insert the explanation.

3 Likes

Naturally. Standard comment: Niebla does not occur in Europe.

4 Likes

Note @mhiggott - I added this as a feature request - now active here

2 Likes

For reference, with regard to more automated interventions, here is the feature request mentioned above by @carnifex I think also

1 Like

Thanks everyone who has replied - it’s clearly not an issue only with the records I’m looking at. I limited my post to UK arachnids as they’re the only records I am looking at a the moment (I’d love to do more invert groups but I don’t have the time). I’ll continue correcting and adding explanations and see if it starts improving. I submit my own records via iRecord, which flags records (rather too precisely) if outside the known range of the species, and in the interface we have locally, species are RAG rated according to difficulty of identification, which acts as a prompt to recorders to consider whether their ID is reasonable. I’m not suggesting iNaturalist should necessarily do the same, but it suggests that improvement is possible.

4 Likes

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.