AI and rare birds (1 Viewer)

[Mysticete]

Earlier today on Facebook a post was shared to a group I belong to, of some horses on a beach in the outer banks next to a giant sea turtle. It was an awesome image and I can get why it was shared, especially if your knowledge of sea turtles was limited. However, most people quickly came to the conclusion that it was a AI generated image. For one, that species of turtle doesn't get that big, and secondly there was some weirdness that you could pick out if you looked carefully, like a horse with only three legs. And of course, earlier this year a AI-generated image of a ridiculously colorful "Baby Peacock" made the rounds fooling folks.

Anyway, AI image generation has already gotten ridiculously good. So it occurred to me, probably in a year or two, if someone wanted to prank someone or just be nefarious, it would be pretty straight forward to create images of rare birds in unusual locations that would be much more difficult to detect.

So do you folks think this might pose a future challenge when trying to figure out if a rarity is real? Will photographs become less useful by themselves? Anyway just some idle thoughts I had at lunch today.

 

[Owene]

People already look at photos of real birds and invent cage damage based on whether they saw the bird or not. I’m sure some of them will now mention AI online more than they used to.

I’m sure it will make the hoaxes that already happen harder to detect

 

[jurek]

This type of hoaxes happened for decades, usually people who posted photos taken abroad.

It would be more useful if AI could sensibly predict rarities!

 

[Mysticete]

This type of hoaxes happened for decades, usually people who posted photos taken abroad.

It would be more useful if AI could sensibly predict rarities!

The issue I could see would be that now someone could generate an image of a bird in a familiar location or with native plants that wouldn't be in their normal home range. I've seen examples of what you first mention, but usually the bird sleuths figure things out from comparing the habitat/plant species with the background and plants in the submitted photo

Its probably a bigger issue with the general public, where people just accept anything they are shown. I am just imagining alligators generated into local lakes or fake game trail footage of various out of place predators and so forth.

 

[opisska]

To be fair, this is probably one of the least worries when it comes to AI generated fake news ...

 

[Andy Adcock]

Earlier today on Facebook a post was shared to a group I belong to, of some horses on a beach in the outer banks next to a giant sea turtle. It was an awesome image and I can get why it was shared, especially if your knowledge of sea turtles was limited. However, most people quickly came to the conclusion that it was a AI generated image. For one, that species of turtle doesn't get that big, and secondly there was some weirdness that you could pick out if you looked carefully, like a horse with only three legs. And of course, earlier this year a AI-generated image of a ridiculously colorful "Baby Peacock" made the rounds fooling folks.

Anyway, AI image generation has already gotten ridiculously good. So it occurred to me, probably in a year or two, if someone wanted to prank someone or just be nefarious, it would be pretty straight forward to create images of rare birds in unusual locations that would be much more difficult to detect.

So do you folks think this might pose a future challenge when trying to figure out if a rarity is real? Will photographs become less useful by themselves? Anyway just some idle thoughts I had at lunch today.

Most AI images can still be detected by the human eye in the main, the so called 'deep fakes' where real images are merged, are more problematic I think.

 

[Mysticete]

To be fair, this is probably one of the least worries when it comes to AI generated fake news ...

Yeah but that is a lot more depressing to think about...

 

[lockbreeze926]

Bring on the Loch Ness Algorithm.

 

[jurek]

I guess together with fake images, people will develop better ways of spotting fakes, and generally stop unquestionably believing images as a proof. Like today, people no longer believe news unquestionably, but ask whether is it likely, does it fit other events etc.

I am more worried that AI will make people able to recognize reality but indifferent to nature. Since access to real nature is increasingly more difficult or prohibited, peoples interests will shift to making better AI technology and away than protecting the reality. People will consume AI films and generally not care that eagles are endangered, dinosaurs are extinct and sandworms are imaginary. People will watch excess of AI images and not care that real wildlife goes extinct.

 

[THE_FERN]

People will consume AI films and generally not care that eagles are endangered, dinosaurs are extinct and sandworms are imaginary. People will watch excess of AI images and not care that real wildlife goes extinct.

Already in train. Ironically I think associated with nature doc spectaculars (99% of which are about the megafauna). Watch a few and you come away with the impression that the world's carpeted in tigers (yes I know every one says "we're trashing the place"). So just a continuation of the trend

 

[noujun]

Earlier today on Facebook a post was shared to a group I belong to, of some horses on a beach in the outer banks next to a giant sea turtle. It was an awesome image and I can get why it was shared, especially if your knowledge of sea turtles was limited. However, most people quickly came to the conclusion that it was a AI generated image. For one, that species of turtle doesn't get that big, and secondly there was some weirdness that you could pick out if you looked carefully, like a horse with only three legs. And of course, earlier this year a AI-generated image of a ridiculously colorful "Baby Peacock" made the rounds fooling folks.

Anyway, AI image generation has already gotten ridiculously good. So it occurred to me, probably in a year or two, if someone wanted to prank someone or just be nefarious, it would be pretty straight forward to create images of rare birds in unusual locations that would be much more difficult to detect.

So do you folks think this might pose a future challenge when trying to figure out if a rarity is real? Will photographs become less useful by themselves? Anyway just some idle thoughts I had at lunch today.

I've generated an image of a Black Robin using AI. Since I've never actually seen this bird, experts might easily spot it as a fake. However, I could potentially lower the resolution to make it less obvious that it's a fake. While there are AI tools being developed to detect such fakes, it's likely to become a never-ending cat-and-mouse game. Bird identification is already challenging for humans, and there's a scarcity of accurately identified images, making bird identification apps unreliable. Yet, most people have limited knowledge about birds, making it easy to deceive them. This isn't just about birds; the more popular an animal, the easier it is to fool people with fake images.

※To clarify the copyright of the image, I will replace it with a version that includes a watermark.

 

[Andy Adcock]

I've generated an image of a Black Robin using AI. Since I've never actually seen this bird, experts might easily spot it as a fake. However, I could potentially lower the resolution to make it less obvious that it's a fake. While there are AI tools being developed to detect such fakes, it's likely to become a never-ending cat-and-mouse game. Bird identification is already challenging for humans, and there's a scarcity of accurately identified images, making bird identification apps unreliable. Yet, most people have limited knowledge about birds, making it easy to deceive them. This isn't just about birds; the more popular an animal, the easier it is to fool people with fake images.

Can you explain what you mean by AI?

 

[noujun]

Can you explain what you mean by AI?

I'm a former data scientist, and I'd like to explain the following as a general overview.

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think, reason, learn, and solve problems. AI technologies include machine learning, neural networks, natural language processing, and robotics, among others.

In the context of my previous comment, I'm specifically referring to a machine learning model that has been trained to generate images resembling real-life birds.

 

[Andy Adcock]

I'm a former data scientist, and I'd like to explain the following as a general overview.

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think, reason, learn, and solve problems. AI technologies include machine learning, neural networks, natural language processing, and robotics, among others.

In the context of my previous comment, I'm specifically referring to a machine learning model that has been trained to generate images resembling real-life birds.

This looks like a copy and paste or other digital manipulation which isn't AI?

 

[qwerty5]

I recall hearing somewhere that there was an idea to "watermark" AI generated text so it could be recognized by computers. The same thing would be possible (easier?) with photos.

 

[noujun]

This looks like a copy and paste or other digital manipulation which isn't AI?

Do you have doubts about my image?
This image was not generated by an AI that I programmed myself, but rather by an image-generating AI called Midjourney, for which I hold the license. Feel free to do a Google image search; you won't find the same image anywhere else.

 

[Mysticete]

This looks like a copy and paste or other digital manipulation which isn't AI?

I am not a computer scientist but have played around with it. So my explanation might not be correct but I can at least give you a sense of the

The difference between AI and photoshop is how they are created. The various online AI image generators basically allow you to give a series of prompts, such as "Trump riding a triceratops in Rome", and then it will basically create that image for you; the image will be entirely new although draw upon whatever material the AI uses or was initially fed to it. There is a whole training process with this that is above my ability to explain, that ensures that the AI "learns" what and how to generate the correct images. It doesn't really understand the images, which is why you sometimes get abominations with extra legs and warped faces. But I have also seen images of people invented by the AI which could easily fool folks into thinking they were of real people.

Photoshop basically is just cutting and pasting or warping images. You are taking images from somewhere else and directly sticking them. To do a good photoshop requires the right programs and a degree of skill that you don't need to do with AI. You can also have it do other things. A major issue in the last year has been students having Chat GTP literally right there papers for them, with just a simple prompt. My friend caught a student doing this just last semester; The student was caught because when Chat GTP gets confused or can't find the data it needs, it invents it. So Chat GTP basically made up several papers and researchers out of the blue, which is how my friend spotted it.

 

[pppppvvvv]

I suspect the AI technology could also be used to input a ton of false data onto the likes of ebird. You could make an army of virtual birders going round virtually birding! Posting on here lol

 

[dantheman]

So we already have accusations of people faking/hoaxing AI images? (ie using photoshop to pretend it's an AI created image)

The irony ... ;-)


(All a terrible slippery slope imo)

 

[Xenospiza]

I suspect the AI technology could also be used to input a ton of false data onto the likes of ebird. You could make an army of virtual birders going round virtually birding! Posting on here lol

This is already happening with the 1000s of people using software like Obsidentify and unquestionably publishing utter nonsense records.
This can be checked by people, but the few volunteers sifting through the records just cannot cope.
There is no AI to correct the AI...