Suggestion: ID Training Mode

[portioid]

I'm thinking it might be great to have a mode that lets interested users hone their ID skills, with a quiz-like approach.

For example like this:

- use only observations that are highly likely to be correctly ID'ed.

- present a subset to the user, for example only observations from two taxa that the user wants to learn to distinguish.

- the user has to determine each one. The correct ID is only shown after the user's suggestion.

This could be done on any levels:

- harvestman or spider?

- beetle or true bug?

- Argiope or Nephila?

- Plexippus paykulli or P. petersi?

It should be pretty straight-forward to program, and my personal feeling is this would really help me (and probably many others) to get a feeling for distinguishing certain groups.

[tony rebelo]

iSpot had a quiz (now that probably disqualifies iNat from considering one!).  The issue is that it is easy to design a quiz for beginners (select only the species with the most observations in the quiz area), and even for intermediate users (select less frequently observed species), but it is so easy to create an impossible quiz (species with very few observations): getting that balance right is a challenge.

Please also restrict any one species to only occurring once in a quiz - due to the Western Leopard Toad monitoring project 9 out of 10 quiz answers - every time!!! - in southern Africa for herps was this species.

But please, if a quiz is done, allow for feedback for misidentifications, also be prepared  for pictures with multiple species to generate ire when only one of the five species visible is an acceptable answer, or when the subject is microscopic and perhaps evident from the description, but impossible to discern from just the photograph..  But most importantly,  esp. at higher levels, find a way to sneak in unidentified observations for taxa that might  be in the focal group - so that the quizzes contribute to the identification of observations on the site and are not a stand-alone feature.. How to "score" quizzes of unidentified observations will be a challenge, and how to incorporate these IDs into iNaturalist will also require some thought.

[James Bailey]

Tony, just because the reputation system had negative views from iNat users, doesn't mean that iNat aims to avoid everything iSpot did.

On Monday, February 26, 2018 at 3:05:23 PM UTC-8, tony rebelo wrote:

iSpot had a quiz (now that probably disqualifies iNat from considering one!). 

I like the idea of quizzes, but they are quite difficult to do right. Reasons have already been given here. My general opinion is as follows: for a quiz to be accurate and reasonable, it cannot take absolutely random photos and observations across the site, and it must take observations with some bias (i.e. not allowing 100% of answers to be one species, when multiple occur). 

[Charlie Hohn]

a while back we had a blind study where you'd see the photo with no text or context. I found that I got a bunch of IDs wrong that I wouldn't have otherwise because I identified the wrong thing or missed details in the notes. So yeah it's really hard to do. I think the closest thing we may someday get is a reputation system based on actual IDs made onsite (rather than external qualifications)... which has been discussed, not sure if it would ever happen.

[tony rebelo]

I suppose the quiz could be part of the reputation system on iNat, but I was more thinking of it as a standalone.

[tony rebelo]

I think that the quiz should be not just of the "cover" or "random" photo, but should include all the photos in an observation.

However adding the description (and the locality!!!) often provides the identification, so including them can be a cheat.

On the other hand, a quiz with just one photo, but with costs/demerits/reduced score for  "see more photos" and "see description" and "see map" would also work.  (Obviously if description is NULL that option should not be presented, and tough if it is a quip like "Whole family under recycle bin" -  https://www.inaturalist.org/observations/10001252

[Chris Vynbos]

I often wish for a 'training mode' to help one improve one's ID skills. 

My idea of how it would work would be: you click the  'training mode' button, which changes iNat to a site where only research grade obs are seen, all identities are hidden but all other info remains, then you select your location and taxonomic area of focus (say 'Coleopetra' or 'Asteraceae' or a genus), you then can browse the obs and make IDs.  You can then make as many IDs as you like and at any point click a 'check IDs' button. iNat then shows you which ones you got right. You can then click 'end training' and you get a score sheet and are told which areas you were weakest with the option of returning to training mode again with those weaker areas the focus. Historical score sheets are kept for comparison so you can check back and see how you have improved over time.

Important: none of your IDs are 'real', it's just training. Nothing happens in the 'real' iNat while you do this.

On Tuesday, 13 February 2018 20:36:34 UTC+2, portioid wrote:

[tony rebelo]

you need to also select if you want an easy or hard test.

But I dont see why your identifications should be wasted.  Correct IDs and incorrect IDs are valuable data telling us how likely the IDs are to be correct.  If there was some way to use these to validate IDs would that not be useful.

Similarly, why train (or allow someone to train themselves) and then stop there?   At the most difficult levels start throwing them unidentified (or poorly identified) taxa, to feed into the ID system. 

And why not an ID mode?: when a certain level of proficiency is achieved use the quiz to actually post IDs.

My only worry about going this route, sooner of later someone is going to suggest that you should HAVE TO take the quiz to determine your level of proficiency to make an ID in a group. 

Oh:  and please a "revision" level as well.  I find one gets rusty in the groups that almost never come in for ID (both on Citizen Science sites and the herbarium).  Would be cool to keep up to speed with a quiz that focusses on the rarest (least observed) species.

[Chris Vynbos]

Cos it aint 'training mode' if your IDs are real, is it? At least, it should be optional whether the training IDs are real or not. To get ppl fully into the training mode, ppl should be able to relax and not worry about mistakes being public. I notice many people are so scared of making wrong IDs that they 'follow' an ob and then leap in an agree as soon as someone else they consider to be authoritative has made the ID for them. Some people don't like their mistakes to be public and it means they never really learn. 

[Charlie Hohn]

Yeah I learn a lot by looking through stuff and just pressing 'R' if i am not sure what it is. I personally probably wouldn't use a quiz beyond that, but maybe others would like it. Just make sure not to add an ID if you aren't sure.

In terms of a reputation system i meant that it itself would be part of a quiz, not that you'd need a quiz to use it.

[Chris Vynbos]

I think it important that the 'training mode' IDs arent' real IDs. For example, say I want to test myself on the genera of Asteraceae, then I'll load up asteraceae and test my genera level ID ability. This will let me know where I'm falling short, what needs attention etc etc. But if my IDs are real, that may cause all sorts of havoc with the CIDs, I might even be re-IDing obs that i've already taken the time to ID to species level. 

I think we must distinguish between the request for a true "training mode", which does not impact iNat and is invisible to all other users, and 'blind IDing mode' which, if I understand Charlie and Tony's comments, is what they are referring to. 

[tony rebelo]

I hate data being wasted.  A quiz (or gaming ID if you prefer) generates huge amounts of data.  These are useful to identify easily made mistakes (and with enough iterations, will sort out mistakes made in spite, or frustration or by accident from those that are genuine mistakes, impossible to resolve or mental slips), and to use them in in say training the identotron, or to refine lists of easily confused species, or alert taxonomists to issues with identification.  I still think that the quiz could inform identifications: whereas an observation may have 4 IDs, if the quiz is popular observations may have hundreds or thousands of IDs  each with a profile of mistakes for other species (or not!), potentially with corresponding information on users and their "test-age" and scores.  Not to use this in some way at some stage to inform the IDs would be criminal.  It does not have to impinge directly on the observation ID, but could be a parallel or ghost system:  or it could be used and flag up observations that are consistently identified in the quiz as different to its real ID.

The only difference between Vynbos's "training mode" and the "blind IDing mode" and a "gaming (competitive)  mode" would be the feedback to the user while gaming and the level of feedback to the observation.

[paloma]

My two cents' worth is that "a mode that lets interested users hone their ID skills, with a quiz-like approach" should be implemented ASAP. I think iNaturalist should definitely focus on catering to people who want to learn. I would use something like this every day. Please, everyone, just agree on something in a spirit of compromise, and get it started. I think it will attract more users, improve the photos people take, improve how careful people are with their IDs, and make more people want to stay on iNaturalist.

[Tony Iwane]

I agree a "training mode" could be fun and helpful, but keep in mind it would take a substantial amount of work (coding, design, testing, etc), and we're a small team. Not to bring down the energy of the thread but realistically this would take some time to build, if it's something we go forward with. 

Paloma, glad to hear from you. We are definitely thinking of improving iNaturalist in ways that would (we hope!) help people learn, although the core (I believe) should always be based on the first-hand experience of making observations and then discussing them with others.

Tony Iwane

[paloma]

Thanks, Tony. It's good to know upfront that something being discussed here would be far down the road. Observations that have ID discussions are great, but there doesn't seem to be a way to search for them when one wants to review a particular taxon in a particular region, without bringing up the vast majority of observations that don't have such discussions in them.