"Computer computer, on my screen — what's the fairest face you've ever seen?"
Presumably, that's what the folks at Youth Laboratories were thinking when they launched Beauty.AI, the world's first international beauty contest judged entirely by an advanced artificial intelligence system.
More than 600,000 people from across the world entered the contest, which was open to anyone willing to submit a selfie taken in neutral lighting without any makeup.
According to the scientists, their system would use algorithms based on facial symmetry, wrinkles, and perceived age to define "objective beauty" — whatever that means.
It's a pretty cool idea, right?
Removing all the personal taste and prejudice from physical judgment and allowing an algorithm to become the sole arbiter and beholder of beauty would be awesome.
What could possibly go wrong?
Of the 44 "winners" the computer selected, seven of them were Asian, and one was black. The rest were white.
This is obviously proof that white people are the most objectively attractive race, right? Hahaha. NO.
Instead, it proves (once again) that human beings have unconscious biases, and that it's possible to pass those same biases on to machines.
Basically, if your algorithm is based mostly on white faces and 75% of the people who enter your contest are white Europeans, the white faces are going to win based on probability, even if the computer is told to ignore skin tone.
So, because of shoddy recruitment, a non-diverse team, internal biases, and a whole slew of other reasons, these results were ... more than a little skewed.
Thankfully, Youth Laboratories acknowledged this oversight in a press release. They're delaying the next stage in their robotic beauty pageant until they iron out the kinks in the system.
Ironically, Alex Zhavoronkov, their chief science officer, told The Guardian, "The algorithm ... chose people who I may not have selected myself."
Basically, their accidentally racist and not-actually-objective robot also had lousy taste. Whoops.
This begs an important question: As cool as it would be to create an "objective" robot or algorithm, is it really even possible?
The short answer is: probably not. But that's because people aren't actually working on it yet — at least, not in the way they claim to be.
As cool and revelatory as these cold computer calculations could potentially be, getting people to acknowledge and compensate for their unconscious biases when they build the machines could be the biggest hurdle. Because what you put in determines what you get out.
"While many AI safety activists are concerned about machines wiping us out, there are very few initiatives focused on ensuring diversity, balance, and equal opportunity for humans in the eyes of AI," said Youth Laboratories Chief Technology Officer Konstantin Kiselev.
This is the same issue we've seen with predictive policing, too.
If you tell a computer that blacks and Hispanics are more likely to be criminals, for example, it's going to provide you with an excuse for profiling that appears on the surface to be objective.
But in actuality, it just perpetuates the same racist system that already exists — except now, the police can blame the computer instead of not taking responsibility for themselves.
Of course, even if the Beauty.AI programmers did find a way to compensate for their unconscious biases, they'd still have to deal with the fact that, well, there's just no clear definition for "beauty."
People have been trying to unlock that "ultimate secret key" to attractiveness since the beginning of time. And all kinds of theories abound: Is attractiveness all about the baby-makin', or is it some other evolutionary advantage? Is it like Youth Laboratories suggests, that "healthy people look more attractive despite their age and nationality"?
Also, how much of beauty is strictly physical, as opposed to physiological? Is it all just some icky and inescapable Freudian slip? How much is our taste influenced by what we're told is attractive, as opposed to our own unbiased feelings?
Simply put: Attractiveness serves as many different purposes as there are factors that define it. Even if this algorithm somehow managed to unlock every possible component of beauty, the project was flawed from the start. Humans can't even unanimously pick a single attractive quality that matters most to all of us.
The takeaway here? Even our technology starts with our humanity.
Rather than creating algorithms to justify our prejudices or preferences, we should focus our energies on making institutional changes that bring in more diverse voices to help make decisions. Embracing more perspectives gives us a wider range of beauty — and that's better for everyone.
If your research team or board room or city council actually looks like the world it's supposed to represent, chances are they're going to produce results that look the same way, too.