Microsoft new face analysis tool called “How Old Do I Look?” took the internet by storm last night. Within just of few hours of Microsoft sharing it, 35,000 users (29,000 of them from Turkey – don’t ask) had used the app. Microsoft engineers Corom Thompson and Santosh Balasubramanian said in a blog post that they “were shocked” by the reception.
But is this fad the real deal or is it just app plugged in with nonsense algorithms, intentionally designed to waste people’s time, and at the same time, make people feel rubbish about themselves?
The app is simple – you upload a picture and it predicts your age. To satisfy my own natural curiosity, I decided to try it. My interest was heightened by the fact that ever since I can remember, I’ve always been called “baby face”. The older I get, the more the compliment seems backhanded. Now, at 23, it’s just downright insulting. Even my 17-year-old sister often squeezes my cheeks and calls me “cute” – enough already. So, I thought maths might be able to determine how old I really look. The result was not what I expected, to say the least:
Seriously? Do I really look 36? Even the photo behind me is less mature than I am. I thought to myself “surely they’ve made some horrible mistake – perhaps the angle of my face is wrong” – I proceeded to upload more pictures, yielding more or less the same result. Perhaps I’ll take the baby-faced comments as a compliment from now on.
I wasn’t the only one who was less than pleased by the result. Social media responded to Microsoft with outrage, even those for whom the site gave the young and sexy thumbs up (there were around 15,000 tweets about #HowOldRobot within 24 hours of the site’s launch). Microsoft has installed a default apology after each age guess which reads: “Sorry if we didn’t quite get the age and gender right – we are still improving this feature.”
This morning, I spoke to Professor Benedict Jones, a research psychologist at the University of Glasgow, and he said: “I absolutely agree that it is important that these apps, even though they may just be a bit of fun, have a strong empirical basis.” If the app is to have any mathematical credibility, functions and the standards used to measure age must be improved.
I also spoke to Dr John Collomosse, a senior lecturer in computer vision at the University of Surrey, and he said:
Facial analysis using crowd-sourced data, as in this app, is attracting significant research interest in the Computer Vision and AI fields.
Recently projects such as Facebook’s DeepFace have been unveiled, trained over millions of user-contributed images and approaching human-level performance at face identification (presented at the CVPR 2014 conference in Columbo, Ohio).Demographic estimation from photographs, as seen in this app, is a closely related task. It impossible to say without further information which algorithms are in use here, or meaningfully comment on their accuracy over previous work. However the state of the art in this kind of technology is often delivered through recent advances in AI referred to as “deep learning” e.g. through neural networks that need huge volumes of data for training. These are beginning to become practical using the deluge of tagged visual data available online e.g. in social media.
Thompson and Balasubramanian also expressed surprise that people are uploading their own photos rather than just random photos online or, ahem, crude places, but in a society that’s obsessed with youth, it should hardly be surprising.