'How sure are you' selector in add observation

Currently this selector has 3 options:
I’m as sure as I can be
Its likely to be this, but I can’t be certain
It might be this

We are thiking about adding a 4th option:
I used AI to help with identification

We want to give people an easy way to say they used AI in some way to help with identification, either via the AI’s on ispot or some other completely different system.

I am not sure having this 4th option here is quite right but we want to minimise any changes to the add observation interface at this stage.

I might favour keeping the 3 existing options but having a tick box to indicate if you used AI or not in the ID but I am not sure if that is possible.

Basically we are looking for comments and suggestions.

I often just use a browser with the common name and switch to photos and use analogue eyes.
If I use AI I search for something like “fungi AI” ETC and select something from the result. If it does well I will add a link in a comment for others to try it out.

This would avoid confusing the certainty of the ID and the route to that ID.

1 Like

If you use AI - it’s hard to make a judgement call about how good you think the AI is. So I don’t think that having a fourth option ‘used AI’ adds anything at all. A tick box might be a good idea if it can be done. In the field, I tend to use the bird sound AI (Merlin) to back-up what I’ve already decided is the correct species. Based on some odd diagnoses (e.g. having difficulty in identifying a nightingale), I hesitate to take the AI too seriously.

1 Like

Refugee - your ‘Analogue eyes’ had me wondering for a bit……
As you, I also… « If it does well I will add a link in a comment for others to try it out.«
Without knowing which AI ( automated intell) was used m not sure how ticking the AI box it might improve our posts. If we we are willing to engage with a comment to explain, in areas of our own expertise, why the AI has got it wrong, that could be helpful. We could, I suppose, get ChatGPT to write such comments for us….

As an exercise, my wife got ChatGPT to produce a report on her. It was about 40% accurate, 10% ‘not too far off’, and 50% completely incorrect. I can see how it reported that she went to Exeter University - almost certainly someone with the same name did. But other ‘facts’ were as bizarre as some of the Fastcat and Plantnet offerings here.

I have looked at the AI we have been trialing for plants.
It works better on flowers if the photos are clear.
Over winter it has done less well on trees that are not in leaf.
At least is does give a percentage as a decimal fraction of one as a scale of its confidence.
I did try a mushroom one that actually showed % figure last year.

I asked a ChatGPT-based Android app to tell me about myself. It demurred, claiming it couldn’t distinguish the various obscure people of the same name. When I mentioned my involvement with botany and mathematics it made up an academic career for me, complete with a Fellowship of the Royal Society of Edinburgh. (Maybe I should have qualified it as amateur botany and recreational mathematics.)

When I’ve used Seek, PlantNet or Google Lens I mention it in the description or identification notes, or in a comment if it’s someone else’s observation, and I’m not adding an identification.

Strongly in favour of acknowledging the use of Ai in identifications - however it is done.
Tag would seem easiest, but very likely to be sporadically used (and misused/duplicated/misspelt…)

the tag has been in use for some time
Click on it here
I don’t understand why so few people know that it is going on…there is a Forum Item…
There are 43 uses here and, as usual, Scotland is well ahead in iSpot!

I’d be in favour of that in general terms. It raises the issue that iSpot is not needed because AI is probably as good at ID, though not commenting - get it?
BUT perhaps we don’t need the others as they are proving useless
I’m as sure as I can be
Its likely to be this, but I can’t be certain
It might be this
as they are proving useless
name withheld!

Could use tag but as many have noted tags are easy to get wrong as the s295 students prove ever year. The easiest thing would be a tick box I suspect but it is looking as if that would be very difficult to implement at this stage. It would be very intersting to see how many people are using AI in some way or other and how this changes over time.

that is because s9295ers are possibly not managed very well.
As the guardian of the AI tag, I would correct every one that was wrong and gently ask users to subsequently use the correct one. I/you/anyone can collect them in Project or instantly better, the Explore Community (Filter) for interest . I have 'managed Marine tagging for years and also did a collection Project for PlantNet’s (tag-based) results. There are no wrong AIAPP tags yet,

It is more important, currently, to find a way to delete wrong tags from the tag list as soon as they occur.
Using the tag instead of a fourth item in the ‘Confidence menu’ would allow all users to see how it is going and what MIGHT be the best, or the most used AI recognition system. How would we filter and collect ‘I used an IA App for this’?
iSpot could be a leader in such things IF it is done properly.

I reiterate that the current choices in the Confidence area are (probably) virtually meaningless. There are plenty of Expert’s ‘Sure as I can be’s’ that have been superseded by correct IDs and thousands of wrong or unconfirmed ‘Might be’s’ in the Browser.
As no-one seems to rely on it and it has no weight in the Banner Algorithm, it could be quietly abandoned, couldn’t it? Dare you!

I don’t think we should take away the im as sure as can be option. Cause some people might actually be sure rather than choosing it simply to strengthen there identification or just choosing any options just because they want to or for no reason. I find the im as sure as I can be option helpful and there might also be some people out there that do aswell. As for AI for me I rarely use the AI for identification cause I prefer to try and identify things without them when out in the field but like useing I spot and I spot actually has real people not just AI and AI shouldnt be completely relied upon it can only recognise images it has been trained to identify cause that is what Artificial intelligence does. There is both ups and downs to AI. We should keep in mind not everyone necessarily uses that or other sources necessarily even though some people if not a lot of them probably do. The tick box sounds like a good idea cause you can find out how many do use it how many also use I spot observations for identification aswell as other sources and weather some people use other sources or the AI function rather than observations on I spot for identification and how many people do aswell

“I’m as sure as I can be” is very badly worded, assuming it is meant to be indicating certainty. If I see a medium-sized black bird flying away from me in the drizzle, it is probably a jackdaw. I don’t know, but I’m as sure as I can be in the circumstances.

I rather like the ambiguity that “I’m as sure as I can be” offers, because I think certainty is not something I associate with a science or a scientific approach. When I use the “I’m as sure as I can be” I think that it incorporates the limits of understanding, knowledge, conditions and error proneness that I have. Somebody else could easily suggest a better identification, but nevertheless the original level of confidence is still accurate.

I would never use an absolutely certain level of confidence.

1 Like

For me I would only send an observation and press im as sure if I was sure and not as sure as I can be. Which means some people are pressing im as sure as I can be cause there only as sure as they can be whereas others are pressing it who are sure and not only as sure as they can be. I only use it for circumstances that I’m sure and not just as sure as I can be.

and a im sure isn’t allways a un scientific approach. not all are corrected cause not all need to be cause it is obvious like a Close up of a male Robin for example and there’s nothing unscientific about the certainty of that identification in that situation they arnt only as sure as they can be about it cause they are sure it is a Robin cause it is. And would be verified as correct on I record even cause it is correct Regardless of if anyone decided to just add an alternative identification in spite of the fact it is clearly a Robin.

and there are circumstances were some people ( are) exsperts and not people just pressing it when there only as sure as they can be aswell. There are circumstances where yoour only as sure as you can be sometimes but not allways. And people don’t allways send observations that could be percieved as anything but ones that could be identified.

I wouldn’t add any of the options if I had no idea of family or species. I would add its likely to be this but I can’t be certain if it’s likely to be that but of course I can’t be certain.

As for it might be this I only use it if I don’t know but I guess it might be but rarely ever use that option. It tells people it might be this but it’s less confidant than the other options cause you think it might be this but you can’t say it definitely is. In most cases people don’t know but they think it might be this so are adding it as suggestion anyway. For those people that only use it for sure and not being as sure as they can be removing im as sure as can be wouldn’t be beneficial

or we could get rid of the as can be part so that it only says im sure and turn the it might be this option into im as sure as can be as an alternative Or we could just not remove any options and just have the tick box

Adding a tick box to a form shouldn’t be difficult. Adding a field to a database shouldn’t be insurmountable, but I can see that it could be a hassle.

You could always have a tick box that generates a tag. (Wasn’t that how urban/rural worked on Mk. 1?)

I have given some thought to the original question; I wonder if a poster indicating that “AI was used” is necessarily useful for other ispotters to know, which is where we might be able to help – for example to point out a better ID and suggest why.

Sometimes I ask posters which book or website they used to come to their ID decision because then I may be able to guide them to a better ID with what they have.

Asking “Did you use AI”, seems like asking “Did you use a book”? Without knowing what book they used we cannot be much help.

For example, if they used the Ladybird Book of Ladybirds (is there one?) the standard format of the Ladybird Series would probably preclude more than a few species. However if they used A Field Guide to Harlequins and Other Common Ladybirds of Britain and Ireland, Helen Boyce then we could guide them better.

The same with websites. If they used https://bna-naturalists.org/id-guide-ladybirds-2/ we could also look at that and suggest what else is possible.

I think it would be the same if they ticked a box “AI app used”. Without knowing which app they used, our offer of help seems limited.