Continued from Artificial Intelligence
“Whether we like it or not, it is here to stay and unless we take a LITTLE control, will destroy iSpot as we know it” is from here
https://www.ispotnature.org/communities/uk-and-ireland/view/project/865652/artificial-intelligence-ai-id-in-ispot
We really need to discuss AI and share experiences
My experience is very limited. Current thoughts/opinions:
- Plant ID apps have (independent research) around a 69% ‘success rate’ (i.e. the first ID offered is correct).
- Most ID apps do not give any rationale.
- iSpot users are in general (a) more reliable, (b) more likely to provide ID notes, and (c) often add other useful/diverting/funny content.
- iSpot users in general don’t seem to have ‘engaged’ with the trials.
- The ‘un-evidenced’ contributions can cause doubt about IDs added by human users: this may turn out to be a Bad Thing in the long run.
- Unlike some others, I don’t find the offerings intrusive, for the most part. Occasionally, though, they are annoying.
Edit: and I forgot to add this, in regard to point 2 above:
The likely reason ID apps don’t, and won’t give any explanation of their choice is that it is their ‘stock in trade’. If they tell you why it’s a certain species, you might remember it and not use the app next time you see it. Keep that to themselves, you will have to use the app again.
Good points, Amadan.
I don’t have a smartphone and hadn’t seen GoogleLens until Dejay mentioned it once. That’s now what I use to get an idea of where to begin when:
a. with plants when I need but don’t know the name of the “bits” to begin to describe it. For goodness sake, petals aren’t even always petals!
b. with invertebrates of any type that look interesting but I need a taxon of some sort to get started at all.
I would never rely just on GoogleLens - it’s only pattern-matching after all - because I’ve learnt from iSpot that there’s always more than one of a kind and I like to try and get close by researching further.
My thoughts aren’t as well formulated as the above two contributions.
I think that, like most technology, AI is a good servant and a bad master.
I wouldn’t dream of using it as a first resort on taxa such as ladybirds, birds, Odonata, hoverflies and others where I have a fair amount of field experience. But if I want a plant ID, usually because it’s relevant to IDing a bug or a moth, then I have no scruples about going to AI. I wouldn’t take the answer as gospel, but I might post it on iSpot to take advantage of the considerable expertise in this group.
At the moment, AI is still in its infancy as far as insects are concerned. If I have an aphid or inconspicuous ladybird I’m stuck on I’d probably email an acknowledged expert. For hoverflies and Hemiptera, Hymenoptera there are also excellent Facebook groups. So, I suppose that what I’m saying is that I don’t have time or inclination to try and work out every ID for myself and I find AI can be a big help - as well as a colossal red herring.
I think that iSpot will probably die for other reasons if it ever does - I think its main vulnerability is that it is very dependent on about 20 - 30 very active users. And I don’t know but I suspect that far more experts have left than have joined.
This is still a LIVE thread despite it being deeply asleep
Browse this perhaps
https://www.ispotnature.org/communities/uk-and-ireland/view/project/865652/ (82 comments)
and then perhaps wake it up.
Should we have the facility to click for AI help?
Should the current Trials in AI ID be stopped?
Should we encourage the use of AI for ID?
.
I think my main worry about AI is that it may discourage people from learning about identification the hard way - i.e. by learning how to pick out the diagnostic features and interpret them. The temptation, especially for busy people, is to just run it through AI and take what it says as Gospel. (Something to which I plead guilty on many occasions!).
Why worry. A few of us ID things to show that we KNOW what they are and why we do. Most users only want an ID or to show us their photos. Why would they need to remember?
Name withheld!
If we surrender human learning to machines, we really will be in Orwellian territory.
Not at all, we can use that portion of our brain for more important tHingze
I don’t follow. If we don’t use the learning portion of our brain for learning, what will we use it for!
Oh, speaking languages fluently and memorising C++ programming protocols for ID Apps
Not much point learning languages - just get AI to translate!
But seriously, if no one has any expertise in, say, insect ID, who will evaluate how good the AI is
I can just about handle botanical literature in Latin, French and Spanish. (Italian and Portuguese are a bit of a stretch). Google Translate is good enough to give access to the German literature, and there might be other languages were it’s also good enough. For other languages it might not have enough of a corpus to do an acceptable job.
There are apps currently been sold for phones which offer real time translation of audio. (And yet the transcription of English audio for subtitles by Amazon is quite ropey.)
I’ve used MS speech recognition from interviews I’ve recorded. The results are sometimes hilarious!
Yes, we did appreciate that when we added the feature - and that’s why the AI IDs are only entered as comments rather than the same kind of ID a human user would add.
I hadn’t noticed, but the vertebrate-ID one (Fastcat?) seems to be MIA. Disconnected, or resting?
It’s still doing some of the birds. It agreed with Roger’s reed warbler a day or two ago.
Remember Fastcat due to be switched off later in the summer due to lack of funding, Think funding ran out a year or two ago.
Interesting discussion here
Beware: AI Images on inat - General - iNaturalist Community Forum which is linked in https://www.ispotnature.org/communities/uk-and-ireland/view/project/865652/
Source 49 Strange Animal Hybrids Bred In Photoshop
Thanks for keeping this going DJ.
I don’t think I spotted this before but agree with the above- Good points, Amadan.
.
I’m not a botanist so would never want to have my identifications used without review.
Scary though has been my experience with various contacts who contribute images to SANBI Red List of South African Plants (Threatened Species Programme | SANBI Red List of South African Plants)
The UCT images and locations of BOLUS Herbarium specimens has become a great asset and suddenly I have become aware that not all the Redlist entries are correct (infallible) - so would AI know this?
Commenting to one of my contacts, who also is very active on iNat - the reply was a brush-off - “going on holiday, -also my botanical knowledge is rather sparse, and so I could not comment intelligently” - WOW there was no response from the forwarded contact (who has some very iffy pictures on SANBI and no refs).
I’m sure they don’t admit this when agreeing on INat.
So how will AI sort this out - one really needs to check and double check and compare - @JoC sets a good example