Questions about Artificial Intelligence helping identify species in photos

[QuestaGame]

Hi Everyone, 

I'm looking for some feedback.

At QuestaGame we’ve been unsure how best to respond to developers who request to use QuestaGame's dataset to train AI. 

On the one hand, AI and machine learning offer some exciting opportunities. On the other hand, the rights to use QuestaGame’s dataset is complicated - for example, in using AI, can QuestaGame give fair credit to all the photographers and citizen scientists who would help train it? Did users submit photos knowing that the photos could be used to train AI?

Then, on top of this, there are numerous social risks and ethical considerations. What broader implications might using AI to identify species have on our relationships with nature, or on the economic value of environmental science?

I’d be very interested to hear how iNaturalist and the people/organisations in this forum are thinking about using AI and machine learning to help identify species.

Last year, when we raised these - and other - questions to a member of Cornel's eBird team that's training AI to identify bird photos (using eBird’s dataset), he was quite receptive and interested in the topic. His team, it turned out, hadn’t yet consulted an ethicist or anyone in the humanities (relatively little work, it turns out, has been done in this space).

So they asked us to write a position paper on “The Ethics of AI Visual Image Recognition for Species Identification,” which we’ve been working on in partnership with the Centre for Biodiversity Analysis at the Australian National University. We'll present the paper at a Cornel/Google AI conference in a few months. 

As part of our research, we requested eBird users to fill in a survey here: 

https://www.surveymonkey.com/r/MCHWNQ9

If any of you are eBird users, please feel free to participate in the survey and/or circulate it to others who might be interested.

Ultimately, we hope to formulate a simple ethical framework that allows us to think about the impacts of species recognition AI - and develop principles that would in no way restrict AI development, or slow technological progress, but might prevent it from encroaching upon “territories of human thought and discovery” that could be worth preserving.

Thanks for your time on this - as well as any input, suggestions, feedback you might have on the issue. 

Sincerely,

Andrew Robinson

CEO, QuestaGame

Fellow, Centre for Biodiversity Analysis

Australian National University

[Richard Ling]

Hi Andrew,

I'm quite interested in this subject -- I work with machine learning and image recognition, and (obviously) an amateur interest in animal identification!  So it's crossed my mind to combine these interests.  However, I was left wondering what might be the social risks and ethical considerations of using AI to identify animals.  Your article on the Conversation has clarified this issue for me -- https://theconversation.com/ai-is-learning-from-our-encounters-with-nature-and-thats-a-concern-88484

Definitely worth a read, for anyone else interested in where Andrew's coming from.

Personally, I think AI is leading to imminent exponential and overwhelming changes, of which the ethical aspects of AI-based animal identification is (with all respect) well down my personal list.

Cheers
Richard

[Charlie Hohn]

it's an interesting discussion! I can't really say I agree though. In terms of it separating us from nature somehow... it strikes me that people probably said the same thing when books and thus field guides became widespread. There's some truth to it but it seems to get buried in the benefits. And I doubt people are going to stp talking about and identifying plants. Once you know them it's easier not to bother with the trichorder any more.

In terms of social justice type issues... there are always important and difficult issues there. But what I've observed is that 20th century ecology and naturalist-ism has been VERY first world and wealthy-person centric. Given that most people in many less wealthy countries end up having smartphones (in part thanks to the people in wealthy countries selling them off to get new fancy ones, but they just aren't that expensive when you don't want a new and shiny one)... Well, an Arcmap license costs what, thousands of dollars? Having a portal to NatureServe costs 10,000 dollars though I do think they have offered discounts to less wealthy entities. iNat is totally free and offers people a quick and nearly free way to do ecology. And while it did start in the us there has been a big effort to make it global. People can also clone the technology and make their own portal if they want. I know that is somewhat peripheral to the algorithm but as long as the algorithm is nested in this community i think it reduces the issues. in short, if anything this sort of technology is making this sort of naturalist data collecting easier to people in different countries and cultures.  Of course there are other ways to interact with nature beyond Western naturalist-ism, and it will be interesting to see to what extent different cultures use this technology versus other traditional and/or new methods. And as the US apparently is in a state of decline barring changes soon... that is unpleasant to those of us who live here but elsewhere in the world new leaders will emerge.

It's an interesting and also scary time though what Richard says is true - other AI things will make this one seem a pretty small issue i think.

I'm just waiting for Google Street View to run the algorithm on all their higher res photos and make some awesome range maps :)

C 

[James Bailey]

Right now, I don't see this as an issue. I think there will be a long age of great benefit with AI before it starts "fighting back", so to speak.

We have encouraged more people than ever to actually think about wildlife around them. Bugs, birds, plants, whatever that may be. Ignorance is the greatest danger to ecology, and preservation of the world: "that "wasteland" is useless, nothing lives there, let's just clear it out and build on it". More of us now know that there is no such thing as useless, and that "weeds" are usually not a scourge on the land (unless they are invasive!). Making the identification process easier makes "veteran naturalists" like us sigh, but it has the important impact of encouraging more users to show an interest. Many people won't do hours of internet or library research to find out what they have seen, and especially not if they have no prior teaching of where to start looking to identify their Asteraceae or Geometridae or otherwise.

I think this is the start of an age where learning, about species, ecology, distribution, and so on, is at an all-time high. There will be more pressure to protect forests, more interest in biodiversity, and perhaps more awareness of how much damage we are doing.

[QuestaGame]

Hi Richard, 

Thanks for this. I've really enjoyed the discussion (surprised the article generated so much interest, especially over the holidays). 

Given your work in machine learning, I'd be interested in your thoughts - can you conceive of some really cool and innovative UX designs that use AI to enhance social experiences and learning? Would love to discuss further. 

I think the fundamental issue here is not so much AI-based animal identification. Rather, it's how we value knowledge, expertise, the power of human thought. Just as we don't want parking lots where native forests once stood; so too, I'd argue, we should be careful not to pave over rich forms of human cognition (including taxonomic identification) with cheap, mass-produced feeds of information. 

Thankfully, I don't think it's an either/or issue. We can have great AI and technology that stimulates great experiences with nature. I know that iNaturalist is exploring many avenues and has done some terrific work in this regard.

Some iNat stats that would be interesting to track:

  1) since the introduction of computer vision recognition on iNat, has the number of engagements between users (e.g discussing species IDs), grown, reduced, stayed steady relative to user growth?

  2) is there evidence of increased user learning? For example, are more users (relative to growth) getting species correct that AI didn't know or got wrong?

Btw, Charles Foster's book "Being a Beast" is a good example of just how much potential exists in a single human brain - especially young people's brains. I'm still trying to come to terms with that level of genius. And, of course, I can't recommend Gerald Edelman's work on cognitive computing highly enough, but then again who can? Years to go before I finish sifting through his discoveries. I think they're highly relevant to any discussion of AI ethics.

-A

[John Bestevaar]

Dear Andrew
I stopped developing my Earthtracker.net site cause i found that being a member of the right Facebook local groups did a much better and easier job of identifying plants and animals. The member just posts their phone low quality photo and asks "what is this" and usually within minutes someone knowledgeable will respond with a satisfactory answer even if its not an authoritative one. The FB terms of service seem acceptable to most people so the ethical and copyright questions are not significant.
Kind Regards JohnB

[Ian Toal]

Very fascinating subject. I have only scanned that paper, and will read it more closely later. Possibly because I like to identify insects, I don't trust AI to do the job for me. They are just too variable, and subject to wear and tear. I will use the iNat id, but always confirm it with one or two other sources. I'm also a bit old fashioned - I like a description of the insect, especially moths. And many species level ID's of insects are based on dissection and genitalia. However, I also heard a program on CBC about how AI and robotics is starting to enter medicine (surgery) and even into music and writing. So while I'm not too concerned about it now, I can see how it could evolve to challenge human judgement. I suspect that it will be an addition to human judgement - it makes an initial call, and the human confirms or denies it. I've already used it that way to get some clues to an id, but have ultimately used my own (limited) skill to go further. I will be interested to follow this, though. 

Ian

[Ian Toal]

When I walk my dog, I think. Here's an example of how I think AI could be really helpful, but also not so much. A farmer, no matter where, finds a lot of a certain 'weed' in his/her crop. Insect, plants, fungus, whatever. He or she takes a photo of it, and almost instantly AI generates a name along with estimates of possible loss of income and cost of treatment for pest. So far, so good - farmer sprays or hacks out weeds etc. However, if the AI makes a mistake (and a lot of pests look similar), the farmer may treat the crop when it is not necessary. This could be from a number of factors - the size of the plot and the infestation, what age/size the pest (often when pests are actually visible all the damage has been done) or a mistaken identity. The farmer could end up treating a crop that he or she does not need to, wasting money, time and/or pesticides. 

I'm assuming I've missed something, as I am not an AI expert, but I see this as a potential win or lose scenario. Just something to think about. 

Ian

On Friday, April 14, 2017 at 11:48:16 PM UTC-5, QuestaGame wrote:

[ellen hildebrandt]

Frankly, this makes me very uncomfortable. Because it is the same technology as this, is it not?

https://www.washingtonpost.com/news/world/wp/2018/01/07/feature/in-china-facial-recognition-is-sharp-end-of-a-drive-for-total-surveillance/?hpid=hp_hp-top-table-main_chinafacial-740pm%3Ahomepage%2Fstory&utm_term=.a5118d62554a

[Charlie Hohn]

i guess so, in the same way that the camera in your phone is the same technology used by spy satellites. I don't really see the connection, though if you are worried about it you can just avoid putting photos of people on iNaturalist, which you aren't really supposed to do a lot of anyway.

I mean, i get it, the technology concerns me too! But this particular use of it doesn't. In my mind, technology consists of tools, and our species has a long, long history of finding ways to use technology to harm or control others, since before we were even humans when one primate figured out it could use a stick to hit another primate over the head. The solution doesn't seem to be trying to hide all the tools. If you hide all the sticks, the cave-people just use rocks instead... and certainly the 20th century is rife with misuse of technology including the ability to totally annihilate civilization so if we want to move backwards we are going to have to find somewhere else to go than there.

Hard question, no easy answer. 

[Cullen Hanks]

You could say the same thing about cameras, computers, and the internet.  I’m glad iNaturalist is using modern technology to empower our ability to learn about and generate data on our natural world.

-Cullen

On Jan 8, 2018, at 9:00 AM, ellen hildebrandt <hild...@gmail.com> wrote:

Frankly, this makes me very uncomfortable. Because it is the same technology as this, is it not?

https://www.washingtonpost.com/news/world/wp/2018/01/07/feature/in-china-facial-recognition-is-sharp-end-of-a-drive-for-total-surveillance/?hpid=hp_hp-top-table-main_chinafacial-740pm%3Ahomepage%2Fstory&utm_term=.a5118d62554a

--
You received this message because you are subscribed to the Google Groups "iNaturalist" group.
To unsubscribe from this group and stop receiving emails from it, send an email to inaturalist...@googlegroups.com.
To post to this group, send email to inatu...@googlegroups.com.
Visit this group at https://groups.google.com/group/inaturalist.
For more options, visit https://groups.google.com/d/optout.

[Whitney Mattila]

I'll present some positive scenarios here: maybe a farmer finds a weird plant growing in their field. Instead of reaching for the strongest pesticide they can find, they either:
1. Find out it's an invasive species resistant to that chemical, and find better, more targeted methods that wouldn't involve trial & error with a strong all-purpose pesticide, saving money & stress on the plants.
2. Find out it's a local plant, and would be better served moved away, or that it'll go away by the time the crops actually grow.

I admit, I'm not a farmer, so my suggestions aren't realistic, but that's what's happened to me in gardening, personally. Living next to a forest, I now know that what plants I need to get rid of when I see them (ie garlic mustard) and what's native.