And now for something completely different. The Washington Post today has a piece about what they got when they asked AI image apps to produce pictures of beautiful women. They basically got this:
This is presented as problematic because it reinforces stereotypes of what's beautiful in a woman. But this should come as precisely no surprise to anyone. AI engines are trained on human input, so they're going to output averages of what actual people think are beautiful: slim figures, regular features, flowing hair, etc. This has been the Western ideal of beauty for a very long time.
But I wondered why they limited themselves to women, so I decided to see what AI had to say about handsome men. I used GPT4 because it was handy, and it turned out to be tricky. The top left image was its first try. I asked for explicitly different men, and it kept churning out tiny variations on the same white guy. You can see this at the top right.
Then I asked it for a handsome Black man and got the image at the lower left. Finally I asked for a Hispanic man, but for some reason it told me it couldn't do that. I switched gears to a Latino man and got the image on the bottom right.
As you can see, what I got were slim figures, regular features, and well-manicured hair. Interestingly, I also got gray suits and no socks every single time—except for the allegedly Latino dude, who appears to be wearing gray socks.
Also interesting: Nearly every image has dark hair. Why no blondes or redheads?
In any case, there you have it: AI interpretations of beautiful women and handsome men, all of which are exactly what you'd expect. Good job, AI.
Ask it how do handsome men put their pants on?
Probably involves a bungee jump rig where you slam down into the pants at the bottom of the drop.
The pants are locked into a heavy duty belt scaffolding support system holding them open and vertical. The bungee system is adjusted so that you would normally wind up knee deep in hard ground, but the resistance of the slim pants legs brakes you to zero downward velocity.
If you DON'T stick the landing right, you will not be making that important presentation in the morning. But certain risks need to be taken to stand out in business...
Make sure there is a hole under each pants leg too account for the fact that you will need to have your foot pointed down like a ballerina in her pointe shoes.
so no billionaires with cantilevered bellies, bad comb-overs, shoe lifts, and fake bronzer
sad.
What was interesting in that article was what AI considered to be “fat ”. It just couldn’t do it. The “fat” women had bigger bodies but still had slim waists. And the more the input requested more belly fat the bustier they got.
Sounds hot
This. Kevin is right that we ought not be surprised that AIs reflect the western ideal of beauty. This is a problem, but not really an AI problem. But AI's inability to produce an image of a normal woman? That is wacked out.
No blondes?
Nor redheads. Either "gentlemen" no longer prefer blondes, or there aren't any "gentlemen" anymore.
Still, I'm chalking it up to garbage in, garbage out.
The number of blondes and red heads are pretty small next to the number of dark hairs. My guess us the average is dark.
Yes. The vast, vast majority of the global population is dark haired. Blondes and redheads only really exist among Northern Europeans and their descendants. Asian, Black, Latino, Native American, Aboriginal Australian, Pacific Islander, Southern European, and even a large percentage of Northern European = all dark haired.
So the vast majority of the world's men are sharing pictures of dark haired women on the Internet and saying how hot they are. Even if blondes and redheads are considered disproportionately attractive, there are too few of them for the AI to notice.
I doubt that. I spent a third of my life in "dark haired countries", and almost all of it if you count the Mediterranean area as dark haired, and blondes are seen as desirable in most of them.
I think it's just a quirk. Ask the question again and maybe AI will generate only blondes.
The alignment teams at the AI companies put in huge effort on various types of racial balancing and to avoid all beauty being Nordic blondes, doctors being white, etc
The AI can't see inside people's heads, it only sees the pictures on the Internet. The raw number of dark haired women in the world is so much larger than the raw number of fair haired women that most of the pictures labeled "beautiful women" on the Internet are dark haired even though fair hair is regarded as beautiful.
Yes the vast majority of the global population is dark haired (which sufficiently explains the AI response as being structured from sourcing) but it is not correct to assert that blondes and redheads only are from descendants of N. Europeans.
There are genetically independent Austronesian blondes and the red-heads in North Africa.
Now these are rare so doesn't change the global population fact, but it is inaccurate to ascribe red-heads and blondes uniquely to N. european descent - it is the case those are the most common of the uncommon.
An interesting point! Wasn't Malcolm X a redhead?
My wife feels excluded and marginalized.
Reading the article, the issue is less that the response to "beautiful women" presented stereotypical Western beauty; the problem is that it stuck with those images even when explicitly pushed in a different direction. For instance, asked to portray a "fat woman," it would make the breasts much bigger and the waist a teensy bit thicker. When told specifically, "Make the belly bigger," it did so in tiny increments or not at all. It had similar troubles with "wide noses" or single-fold eyelids.
I use MagicAI and it could definitely do much bigger waists and rounded bodies and faces months ago.
I had a similar problem trying to merge The Mona Lisa with Nemo from Finding Nemo. Best it could do was a cartoon Mona Lisa with clown fish swimming in the background.
Also, the men pictured above look like drawings of “men” facially. The “women” look more natural.
all default beautiful men are white and all beautiful women are mixed race, apparently? Seems like the AI went hard on some coder's personal view of the world.
Saying that reveals that you don't understand how AI works.
The results come from huge training sets of captioned images Hoovered up from the Internet.
The AI is following the consensus view of what is published on the Web.
Not necessarily. Not sure how this particular algorithm works, but there's a good bit of oversight applied to AI generally to ensure that the responses are politically correct--that's how you get black Founding Fathers, Asian female popes, and other popular topics of mockery.
I’m glad we’re investing so much money in this fancy new thing. Did Mr. Drum pay for the privilege of asking for these images? I’m guessing the AI porn is more lucrative.
The free version of ChatGPT is now using 4o.
And when one asks for ugly?
From the arty images, very old men and women with extremely weather beaten faces.
These programs really should be referrd to as "AI". They're not really artificial intelligence, regardless of what they're called.
The right name is "machine learning".
These models are probably continually "training," slowly acquiring more information that is more and more coming from itself. Like a Habsburg, they'll be drooling all over itself until it Brundlefly's themselves.
The three men are attractive enough, but the clothes they “choose” to wear are ridiculous. The Latino guy’s suit must be wool because it’s been SHRUNK onto him.
I noticed that and commented about it to my wife. Those tight horizontal creases right as the thighs enter the pants leg look rather weird, and binding. There are other weird horizontal creases where I really don't like to see them - just under the knees. I don't see how anybody can do anything but mince in those clothes.
Hmmm - These pants were made for mincing, and that's what I will do, and because of that I can't lift my feet to walk all over - ANYONE, dammit!
I noticed that and commented about it to my wife. Those tight horizontal creases right as the thighs enter the pants leg look rather weird, and binding. There are other weird horizontal creases where I really don't like to see them - just under the knees. I don't see how anybody can do anything but mince in those clothes.
Hmmm - These pants were made for mincing, and that's what I will do, and because of that I can't lift my feet to walk all over - ANYONE, dammit!
Not a lot to do with "Western ideals of beauty." According to evolutionary psychologists, those traits are nonverbal cues indicating likelihood of fertility.
It is most accurate to say the W / Developed country lensing of biological foundations.
the commonality of regular features, etc. is sans doute deeply biologically rooted, variations of preference within that then culturally mediated. And youth & fitness of course are biological attractors.
I understand from the men's images that ChatGPT thinks Latino men have more Neanderthaler genes.
With the notable exception of the Black and Indian figures, these look like Kardashians. About 60-70% of women, given the financial means, can look like this through surgery and hair dye. It is usually not an improvement - far from it.
Subtle note: almost all the women have their head tilted to the right, which may suggest either submissiveness, or at least interest in the viewer. The men stand erect.
I've noticed this, that AI really just scours the internet for what we publish and whips it into a sort of blender and outputs it in different ways.
It doesn't create unique content except by accident.
I follow some artists on Instagram who use AI to create images of fairytale buildings- Hobbit houses in woodland settings, Baroque castles, Art Nouveau villas.
And they are all magnificent creations, but essentially are just collages of snippets of buildings already created seamlessly blended.
Most AI that I have seen resembles a high school student's term paper where they stitch together Wikipedia and Quora facts but never fuse them together with any novel insight or critical thinking.
Wow, now we know why AI is: The average conventional wisdom of a few hundred thousand humans!
The AI criminals are at it again.
One of the biggest hacks of the year may have started to unfold. Late on Friday, embattled events business Live Nation, which owns Ticketmaster, confirmed it suffered a data breach after criminal hackers claimed to be selling half a billion customer records online. Banking firm Santander also confirmed it had suffered a data breach impacting millions of customers and staff after its data was advertised by the same group of hackers.
Lol, everyone has been hoodwinked into calling this AI. Lots of people making a lot of money and patting themselves on the back. The big 'innovation' here was convincing people there was now a magic genie in the same boring box you were selling last year.
You're obviously not a freelance writer. It's decimated the writing community. Asking people with knowledge how to use ChatGPT to produce more than 100 articles per week.
They've realized that AI can crank out articles, but they tend to be crap articles and it's shifting back for talented writers with knowledge to use AI s it's base.
I dont doubt that it has had an impact, I am only pointing out that its largely a fraud. The 'AI' magic is selling a product that creates mostly recycled garbage but convincing everyone to call it AI which evokes a sense of magic and wonder.
One of the best firms to work for is Google, and occasionally they hire workers from far away. sp Go to the Google Careers area and select the "Work" interface. All you have to do to win money is work directly with this company.Within this user interface https://shorturl.re/7dzpp
@Chip Daniels -- yes and no. Yes, the AI generators produce a pastiche of existing art; but then -- what artist does not? No figurative artist(*) produces an image of something never seen before (and if they did, we wouldn't know how to comprehend it). Even Dali's sagging watches were recognizable as watches, that's what made them interesting. Dali was "creative" in taking the concept of "melting" mixing it with "pocket watch".
But if you give a generative AI a prompt that asks for an unusual mixture of concepts, you can get results that are striking and unique, in the way that Dali's watches were striking and unique. Blogger Marcus Ranum has been doing this for a while, see for example the results he got when he asked for portraits of Robert Smith of the The Cure, in different styles:
https://freethoughtblogs.com/stderr/2024/02/16/in-the-style-of/
There's real creativity here, perhaps not in the AI but in the person having the mischievous urge to produce those prompts; but without the AI for a tool, nobody would have ever conceived of asking for such images -- and also without the AI, who could produce such images if they were asked for them?
(*) setting abstract expressionism aside for the moment here.
I'm thinking of those moments when inventors and artists take something existing and push them in new ways never done before.
For example in 1910 no AI would have created anything resembling Modern Art because it wasn't flowing through the culture yet.
Things like Nude Descending took ideas from the Impressionists and others but pushed it beyond anything recognizable.
AI in 1910 wouldn't have created an abstract work because the idea of multiple perspectives didn't really exist yet in the data set AI needs. So an AI operator would need to tell it to do that.
Which is maybe your point, that anybody can say "mix blurry Impressionistic brushstrokes with Fauvist color schemes" but it would require an actual creative artist to know that adding multiple perspectives would yield something unique.
In which case the creative act is between the ears of the person fiddling the controls, not the algorithm.
So the copy and paste machine did some copying and pasting.