That - Community taxon algorithm - thread is closed. We can search for our own Mavericks - but that is not very useful since the CID has moved on and discarded our Maverick ID.
If the issue is finding obs trapped at a (broader) ID. In particular 2 against 1, or 1 against 2, depending which side you are on. https://www.inaturalist.org/projects/pre-maverick contains over half a million (which is significant even across iNat) obs in need of a little TLC. Some are junk Reviewed, Next. Or Copyright Infringement.
We have that new Disagreements filter but somehow I donāt use it. PS and prompted by this thread to use the new filter - yes - it is useful !
To alleviate the issue a bitā¦ After a while you get a clear mental shortlist of the usual āunresponsive blind agreersā. From then on it is simply a matter of deleting your ID once the person has copied it. Downside: you wonāt make it as easily onto the leaderboards.
Orā¦ keep your own ID around, while requesting further independent review thanks to the DQA checkbox (ācan it still be improved? yesā).
I agree and have done both these things (removing ID or bumping back to a higher level or using āYes it can be improved), as well as leaving comments pointing users towards iNatās new-ish Help article which does a great job covering the issue:
One issue for me (and many high volume IDers), is that we have notifications for confirming IDs turned off (to keep notifications manageable), so we donāt see all of the confirming IDs that come in. And it would be a lot of work to check each individually to see if they are justified.
In 2020 there was a change which removed the Agree button on observations which are already RG, with the intention of disincentivizing āredundantā confirming IDs. That was quite unpopular and was reversed a couple days later. However I think it would be valuable to experiment with doing something similar but instead removing the Agree button on identifications (maybe just Leading IDs?) on your own observations.
sorry yes, there are no threads open for voting algorithm discussion. I think the current one is fundamentally inefficient (mathematically provable) algorithm for some cases of disagreements. As survey summary indicated, it needs lot more identifier efforts during disagreements consensus in multiple ways and not many new IDers even realise there is fixable shortcoming in algorithm for some of those cases.
yes one can simply tag another users always to switch to stronger consensus Observation ID, as is done by lot of prolific IDers but the onus should be more on consensus algorithm design and not identifiers effort and human patches to compensate for algorithmic weakness. yes I do love pre-maverick project
to give one concrete example where algorithm loses specificity:
the algorithm makes a hard assumption in scoring that I nor I assume many IDers are implicitly aware.
whereas identifiers always operate at different epistemic knowledge depths and the hard assumption is very unrealistic for all scenarios - that every IDer knows every node well between ID they are disagreeing to (which is what is indicated when clicking disagree option - you are only disagreeing to that ID) with broadest confident ID picked by them versus hard disagree of everything in between. This is very realistic on inat where expert IDers correct spurious CV locality IDs in another region for their taxa, but they donāt know the families/ any other nodes/ā¦ well enough for that locality so they recourse to picking higher confident IDs and how those well intentioned IDs block subsequent finer IDs compared to optimal voting algorithm. (yeah we have >2/3 eventual consensus, the point of algorithm fix is that there is still wasted identifier effort or genuine loss from IDs stuck at higher and so not reaching downstream expert groups from design per se) - although the assumption is also valid for most IDs with explicit intent on hard disagree over entire path and the only way to better decouple these is by using subsequent IDs temporal order weightage better in voting score and may in the future incorporate modeled identifier skill (who knows when).
This has unintended another causal effect - it loses specificity as tristan mentioned already - for spurious CV genus A ID ā a disagreeing disjoint (that is genus A is not here) Order X ID appears ā and other broad IDers now have to chain up lot to make Order X win ā then any Genus Y ID in Order X is lost in voting from all those well intentioned chain IDs whose goal was never to block Genus Y. it is better to surface Genus ID Y for observation ID (not community consensus 2/3 or RG) than letting it sit suboptimally at order X in current scoring design.
but anyway we can revisit this and math and simulations and various algorithm solutions for more edge cases in future thread after i or others think more :)
Thanks for the feedback and great discussion so far!
To be clear, all photos/sounds on iNat are stored on Amazon servers. Ones that have Creative Commons licenses are stored on https://inaturalist-open-data.s3.amazonaws.com/. Ones that are unlicensed (all rights reserved) are stored on https://static.inaturalist.org/, which is also on Amazon servers. Are you saying that photos stored on the former are loading more slowly for you?
Really? I thought the life stage annotation was attached to the observation, not an individual picture. When I add life stage to my own obs, I only do it once regardless of how many photos I have added.
Thank you for link (i only joined inat an year ago and its always funny there exists something on forum or somewhere for everything i feel suddenly lol).
I honestly donāt know that was the algorithmic choice anywhere since joining while i was disagreeing (only after reading code - although now i see journal posts and forum threads exists for those now)
they are implying unnecessarily clicking observation image aka opening each observation on identify page while they can instantly realise the lifestage and could have added it easier from thumbnails of identify page itself if such solution is developed. just conflated āimageā word usage for āobservationā.
That too is part of the iNat learning curve. I was shocked the first time I realise that - Whatās This ? at the CID - explains - for - each single obs - how the algorithm applies āhereā. The initial one size fits all wall of text is disconcerting. Some may never scroll thru or past that.
@prunhel in Identify. Work thru IDs for that page of 30. Then whizz thru a second time on the annotations tab using the keyboard shortcuts. Caterpillar is L and L for Life stage and Larva. etc.
Iām not sure itās possible to tease out those insights from this survey, but I think it would be great to try to collect relevant data in a future post. For example, my hunch is that people who focus on arthropod, annelid or fungus identifications would rate particularly highly for this response.
Too many Computer Vision identifications to correct
And maybe also for these ones:
Too many observations have poor quality evidence
There are too few identifiers on iNaturalist
Too many observations to identify
The same factors can be an annoyance for us plant identifiers, too, but we really have it so much easier!
Interesting to see the results; thanks for sharing!
I must admit it took me about two years of intensively using iNaturalist before I actually looked at the Identify page - I saw the tab but I just kept identifying through the Explore tab for the longest time. Little did I know!