We can no longer complacently say, “Seeing is believing.”
Discuss. That is, if you can believe your eyes.
Aw, you mean Superman wasn’t really flying in 1978?
Who said that?
There were a couple of articles I read somewhere commenting on this year’s uncannily high level of quality among AI-generated deepfakes compared to last year, and to the casual observer - and that’s most of us - they’re pretty good. Reportedly all the cats in the following composite pic (from this article) are AI-generated:
Among the fifteen there are seven that are for me detectable; but except for one, I still had to take a good hard look not to be fooled. I also think you have to be very familiar with cats to know what’s realistic and what’s probably not, and compared with the terrible quality of before, now the tipoffs can be subtle.
Reportedly all the cats in the following composite pic (from > this article> ) are AI-generated:
What’s worse is that we still need to find them homes.
Make that eight.
If AI can make cats, it can make the real estate. And the litterboxes.
Eleven. Yes, I have no life.
I don’t even know what a deepfake is. I’m so out of touch with the world.
Here you go:
https://en.wikipedia.org/wiki/Deepfake
Strictly speaking, I suppose the cat images might not be considered deepfakes per se, but only constructions. Still, that’s part of the game.
Thank you Nano. I should’ve known. I am a bit gullible.
I don’t see why; it’s only recently become something of a household word, and it won’t be every household.
We all are. It took some really hard looking for me to realize why those pics are fakes, and I didn’t get it all on the first pass. There are only four left so far that give me little reason, other than being told, to be certain that they are in fact artificial constructions.
More sophisticated eyes, particularly those in the field of AI, might be better at it; I have to rely on my familiarity with cat physiology, and on things such as lighting and consistency. Some say that blurry backgrounds are a dead giveaway, but I think not necessarily, and in any case AI is bound to catch up with that.
I’m going to have a go at what I see as ‘wrong’ with those cat pictures. I think with a few of them, I would have seen that there was something off, but still wouldn’t necessarily have thought they were fake - just American. 
Anyway, here goes:
1 The eyes seem to low, and the forehead too flat
2 The eyes are too small and the wrong shape
3 The eyes are too small, the wrong shape, too wide and too low
4 I would never initially think this one was a real cat. That mouth is nothing like that of any other cat I’ve ever seen … I can’t beyond the mouth, really …
5 There’s something wrong with the way the fur is lying, and the ‘mascara’ and ‘eyeshadow’ just doesn’t look convincing
6 The noes it too broad
7 If that is supposed to be a cat, then it’s been crossed with a bat
8 The eyes are the wrong colour, the wrong shape and too far apart, and the nose looks more like a dog’s nose than a cat’s
9 Too human. Fairly convincing, this one, for me, though …
10 This one is very convincing. I’m slightly put off by the hairs in its ears, but apart from that, that one is convincing …
11 This one doesn’t look like any cat I’ve ever seen. I have never, ever seen a cat that colour, for a start, plus it looks stuffed
12 Convincing. Maybe the nose is a bit broad at the forehead, but pretty convincing …
13 There’s something absolutely wrong with this one. The eyes don’t look remotely in the right place to me
14 Convincing. There again, it’s a bit blurry
15 Convincing.
How does that measure up with what you were thinking, Nano?
Some people haven’t said that since shortly after the invention of photography. As the Deepfake Wikipedia page says “the act of faking content is not new”. A century ago people who could have been regarded as ‘influencers’ of their time were fooled by some crude efforts - https://en.wikipedia.org/wiki/Cottingley_Fairies Normal folk going to a show of ‘lantern slides’ given by an explorer may not have noticed if some had been ‘touched up’ a bit to better illustrate the story (I have seen examples )
Forewarned is forearmed. We know what is now possible, and as before we know that we might not be able to tell from the final product.
As for the cats - how do I know you can tell? As a test it would need a mix of fakes and real cats, some slightly weird. I was going to look for weird cat photos but remembered this;
So far as I can tell it’s real and not an April 1st story that got out of hand. But I’m sure the eyes would be easily done in Photoshop. So either way seeing it I’m not going to believe it without reservations.
As for the AI bit. Meh. The guys who made Superman fly did some clever stuff and could probably explain it to human beings in plain English (or whatever).
It’s very interesting: In some pics you and I see the same things, but in others we don’t, and in others yet we have very different cues as to why we think they’re fakes.
Here are my observations:
1: The back meets the head too high up.
2: The body behind the head isn’t really a body at all; it’s just … random furry stuff.
3: Convincing.
4: Same as Ben. That lower jaw is all wrong. Plus the backlighting effect on the side of its ruff isn’t seen anywhere else.
5: No discernible chin; the face just sort of blorps into the neck.
6: Convincing.
7: Same as Ben. If that is a cat, it belongs to Cthulhu.
8: Going by typical Siamese coloration (which this only approximates after a liberties-taking fashion), the neck coloring is quite wrong. Plus the neck has an implausible formation. And I agree with Ben that the eye coloring is unnatural; if there is a first tipoff, that should be it.
9: The whisker pads aren’t realistic; they should be more prominent. Otherwise, not bad.
10: Convincing.
11: Convincing, with slight reservations about what’s going on below the head. On second look, though, I’m inclined to agree with Ben that its color is unnatural.
12: The muzzle is skewed. It could bite around corners if the corner is on its right.
13: Same as Ben. Too much asymmetry with the eyes.
14: The face’s plane is too vertical.
15: The ruff is asymmetrical, and its edge on the left (the cat’s right) is not as feathered as it should be.
I don’t know; I’m pretty impressed that AI can make such pictures independently with no human manipulation of the product required. Makes me all the more liable to keep my eyes open - and I’ve been doing that for years already, but, I must admit, not always with success. I remember a certain tuba …
Well then, AI can scoop the litterboxes. (It can come practice on mine… which, unfortunately, is quite real…)
Do we know if a human chose those 15 and if so what the rejects where like?
I’m impressed by the technology but irritated by the media hype over AI. Does this ‘intelligence’ tell us anything about cats? Improve our understanding of them ? Could it tell us more about making images of cats than the artists who advertise ‘paint your pet’ services in the local vet’s waiting room? I guess it could be a spin-off from something more useful (or maybe we need fake pictures of cats?). Similarly (useful but off topic) do the AI systems that can match radiologists in recognizing things of concern in a screening X-ray tell radiologist anything new about X-ray interpretation - maybe they do but the media doesn’t report it.
I’m all for computers helping us understand the world. Not convinced most AI, as presented by the media, does.
I think it’s safe to assume that a human did indeed choose the selection, for it was compiled for human amusement - and that the rejects will have been much on the order of Cthulhu’s moggie (#7). From this sample alone, the actual percentages are anyone’s guess. That there are convincing successes at all, including close calls, is, for me, more the point.
Well, what I’m getting out of it isn’t the verisimilitude of cat images, but the exponential increase in AI’s capabilities. The implications far outstrip mere cat pics.
They do, and it does. My awareness of AI in medicine has come about solely from the news media. While the following isn’t “media” as some might call it, it’s being taken seriously:
As I understand, there are already cases where AI has made correct diagnoses that humans missed. That’s a pretty important finding. I don’t think it means AI is our savior, but it has proven to be an important tool if you have the means to use it.
I think media presentation is aimed at poor dolts like myself who can’t tell ham from a hemostat, but can nevertheless put two and two together. I’m now persuaded that the sum might, in time, be 5. It’s both exciting and scary.
Nice article thanks, but regarding the “They do and it does” I read it that it doesn’t or they don’t. Is the AI, after having learned from many examples, able to ‘explain’ what the characteristics are that it is spotting that a clinician isn’t? Much as you and Ben, in the converse situation, can explain what you think the cat-drawing-AI is getting wrong.
I like the bit about whether people would prefer to be misdiagnosed by an AI or by clinician if the AI had been shown to be better at diagnosis than a clinician.
Getting back to cats. The reason I didn’t post examples of weird looking cats is that it we do an image search for “weird looking cats” they have almost all been manipulated and it’s hard to be sure any particular one hasn’t. An image search for cats yields quite a few that look odder to me than some of the 15 above but I guess we need an array of pre-1980 cat photos to be more confident.
I’d say you’re right. For me, the only plausibly real samples under “weird cats” are either those with birth defects, or engaging in behaviors that I would call perfectly representative of a cat’s physical capabilities but happen to be entertaining to the human onlooker. All the rest are clearly manipulated - and, I ask myself: For what? A cat is not improved by painting an unnatural smile on its face. But then, maybe I’m a wet blanket; I’ve never heard Photoshop’s siren call.
BTW, when I saw the doggy pic my first thought was that it had been manipulated: notably the eyes, and I’m not so sure about that bandana, either. Come to think of it, when you take a good look it appears as if the whole dog’s been assembled from parts.