Slashdot Mirror
Microsoft's AI Judges Age From Snapshots, With Mixed Results
mikejuk writes: A Microsoft Research project that lets users upload photos and estimates their age and gender has attracted more attention than expected — not all of it complimentary. The How-Old.net site demonstrates of some of the capabilities of the Face API included in Microsoft's Project Oxford that was announced at Build. It may have been expected to be a source of amusement but instead it backfired when people started to upload their own photos and discovered just how wrong its estimates could be. It demonstrates not only that machine learning has a long way to go before it's good at estimating age, but also that machine learning may not be the most politically correct way to go about answering the question 'How Old Do I look'. It might be better to employ and algorithm that built in all the rules of how to make a polite answer to that request — such as always knock a decade off the age of anyone over 28. Perhaps this particular neural network needs to learn some social skills before pronouncing how old people look. However it is capable of telling some truths — a photo of Barak Obama in 2005 gives an estimated age of 46, close to his real age of 44, but just 9 years later in 2014 the age guessing robot places him at 65. It seems that Mr President aged 20 years in less than 10 years of office.
It's probably more accurate to say that Presidents look haggard and appear to be older than they are while in office. Probably something to do with the responsibility and the stress. Thing is in this Microsoft software, if it doesn't have a means to address the difference then it probably will err on the side of older.
Someone did a faces of pornography shoot where they took headshots of pornographic actresses before and after their makeup was applied. I wonder how substantial the differences there would be with the subject at the same age.
Do not look into laser with remaining eye.
The only guy I've ever met who could judge the age of people well was a man who made some money on the Stroget in Copenhagen more than ten years ago by betting with pedestrians that he could guess their age (he probably did it mostly for fun, though). But he didn't use the face only, he used all kinds of clues, especially he looked at people's necks. He was really good, but everyone else just sucks at it.
No wonder A.I. can't do it when ordinary people can't do it either.
I think it's mostly hair.
A Presidential candidate is trying to look like a young man, a virile man, a man with great energy to do great things. He's probably in his late 40s or early 50s. He may have some gray, but if it's a lot he's probably dying it the fuck away (you'll note most of the GOP field has no gray at all, despite the fact they're all in their late 40s at least, back in '08 the Dem field was just as non-gray). If he's a she of course she's dying the gray away (I love you Hillary, but we all know your hair today is not precisely the same shade it was back when you personally murdered Vince Foster and then arranged for the paperwork to say "suicide" back when I was 12).
OTOH, as his term goes on a) non-gray hair becomes less socially acceptable (just imagine what Jon Stewart would have done with the tidbit that Dubya was dying), b) it's harder to claim that you actually aren't dying (Obama will leave office at 55, and he's on the young side), and c) he wants to look like the Alpha Male Silverback Gorilla who has Already Accomplished Great Things. So you remember a 50-year-old from the campaign with no gray, then he gets on HD TV at 57 with the normal level of gray for a 57-year-old, and you think :"Gee the past seven years have really aged our President," when in fact all that's happened is he's stopped using Just For Men.
I think it's more an effect of the people we see pictures of most, celebrities, put a lot of work into appearing young, so we don't expect people in the public eye to age as quickly. Tom Cruise ages slowly because his career demands it, Barak Obama on the other hand probably looks more serious the older he looks, so there's less reason to make himself appear young.
Even compare to Jon Stewart in 2008 vs now. There doesn't seem to be a huge difference, until you realize the gallon of makeup applied to Jon Stewart's face, it's hard to appeal to GenXers and Millenials looking like you're over 50.
I'd actually be curious to see how this algorithm does with celebrity photos.
I stole this Sig
The criticism of website is unfair. Some people in some photos indeed look younger. Instead of criticizing AI algorithms, I would suggest take a test by using a random photo where you don't the true ages beforehand but have a way to find it out. Compare you guess with the website and see who does better. I tested several photos from my collection by asking my colleagues to guess the ages of my extended family members and the website. On an average the website was actually better than my colleagues.
This is from the website's TOS, which happens to be the TOS for Azure as well:
"Microsoft does not claim ownership of any materials you provide to Microsoft (including feedback and suggestions) or post, upload, input, or submit to any Website Services for review by the general public, or by the members of any public or private community (collectively "Submissions"). However, by posting, uploading, inputting, providing, or submitting your Submission, you are granting Microsoft, its affiliated companies, and necessary sublicensees permission to use your Submission in connection with the operation of their Internet businesses (including, without limitation, all Microsoft services), including, without limitation, the license rights to: copy, distribute, transmit, publicly display, publicly perform, reproduce, edit, translate, and reformat your Submission; to publish your name in connection with your Submission; and to sublicense such rights to any supplier of the Website Services."
So, how are you, future microsoft's advertisement stars?