Age prediction's child diseases

In this (second) machine age, it was only a matter of time before machines would start guessing our age. Researchers from Microsoft have recently released an online tool called “How Old Do I Look?”. When given a photo, the tool will not only tell you how old you are, but also your gender.

Since one is never too old to learn — especially about oneself — I gave it a go. Contrary to this rather biased article in The Guardian, I wanted to use more than a single data point per person to evaluate the predictor’s quality. (Data) Science oblige!

Enter my often mocked selfie archive, painstakenly collected over the years for exactly this kind of scientific purpose. Thirteen pictures from the last year were pseudo-randomly selected, with some variation in hairstyle, location, sunlight, season, hours of sleep, bespectacledness, and headwear.

Let’s start with the good news: in 13 out of 13 cases, my gender was determined as being of the male kind. To whom it may concern, and for the record: this prediction is 100% congruent with the actual situation.

Then again, according to Microsoft, my average age over the last year was (38 + 39 + 42 + 43 + 43 + 45 + 47 + 48 + 54 + 54 + 55 + 55 + 58) / 13 = 47,8 years. This is almost 2 years more than the actual average age I had when the photos were taken. Not brilliant, but not that bad either, given the relatively small test sample size.

With a standard deviation of 6.7 years (about 1/7 of the average), the age predictor does allow itself a fair amount of variability. As the saying goes: with age comes wisdom, but sometimes age comes alone. I’m talking about the predictor, of course — it may still have some child diseases to outgrow.

On a less serious and scientific note: both my daughters will no doubt revel in joy that funny hats do not make a middle-aged person — let alone a father — look any younger. Quite the contrary.