In November 2016, Google released a cute little game called Quick, Draw! on its AI Experiments website, where it showcases fun or unusual AI experiments for consumers. Quick, Draw! challenged you to draw–in 20 seconds or less–items ranging from tennis rackets and wine glasses to yoga and the Mona Lisa, all for the purpose of advancing machine learning research.
Since then, 15 million people have generated 50 million drawings–what Google is calling “the world’s largest doodling data set”–that are now available for researchers, artists, and designers to use in training algorithms to do things like distinguish a scribble of a boomerang from a doodle of an elbow. Just scrolling through the site that showcases hundreds of these drawings is a joy in and of itself–and a glimpse into how certain objects seem to share certain unalienable details.
Take many of the electronics-focused doodles, like “stereo,” for instance. Nearly everyone who drew a stereo in Quick, Draw! drew some sort of box with two round circles inside it, more akin to the boomboxes of the 1980s than any modern stereo. “Headphones” is similarly intriguing–most drawings feature two circles with a line between them, code for old-fashioned (or noise-cancelling) headphones, even though headphones with two earbuds are far more pervasive. The vast majority of the drawings of a “television” have two-pronged antennae–when was the last time you saw one of those?
It’s a reminder that culture is often a step behind technology. An image of an old-fashioned, antennaed TV is far more visually dynamic and easier to understand than a drawing of the flat, nondescript boxes that serve as televisions in many households today.
There’s something similar at play in emoji–after all, there are 11 separate emoji relating to mailed letters, and only one for email. And, it turns out, the emoji icon of a TV is an old-school vacuum tube box with dials.
Maybe machine learning algorithms will only learn to recognize drawings of 1950s televisions as a result. But while the different doodles are fun to peruse (just check out the “beard” drawings), they raise questions about how Google as a company uses its user data. In one sense, players provided free labor to Google to create this data set. When Co.Design asked how Google plans to use the data set, a Google spokesperson sidestepped the question, instead saying that Quick, Draw! and the rest of its AI Experiments projects are ways to make AI and machine learning more accessible to the public (though it has likely already used this data set to create Autodraw, which guesses that you’ve tried to draw a birthday cake and turns it into a polished drawing).
It might seem like an innocuous game and a fun data set, but the existence of both raise a greater question: How will data that users create for free be used by companies like Google as it continues to push toward a world where algorithms reign supreme?