For Google, it’s not enough that its products rely on machine learning and artificial intelligence. The company also wants you, its customer, to understand how these technologies work.
Last year, a few months after it open sourced its deep learning engine, a Google researcher partnered with The New York Times to create this data visualization explaining neural networks. Now, Google has rolled out AI Experiments, an online collection of tools and games designed to help you understand the inner workings of machine learning.
Take the game, Quick, Draw! It works like Pictionary; the game gives you 20 seconds to draw an object on screen, and Google shouts out guesses as the time ticks down. And Google is good. It asked me to draw camouflage, a microwave, a hexagon, an umbrella, a baseball, and a crocodile, and the neural net guessed correctly every time.
But the game’s accuracy, while impressive, isn’t what makes it a powerful learning tool. It’s how, by observing the way Google responds to your doodling, you can get a better sense of how its technology works.
Here’s what happens when the game tells me to draw a tree: I start by sketching leaves. In its robotic voice, Google guesses: “squiggle.” As I add more leaves, it sees a bush. Finally, I draw the trunk-and it clicks. Google’s neural net says: “Oh, I know, it’s a tree.” (After the fact, you can scroll through other players’ tree drawings to see what illustrations Google has used to inform its guesses.)
Similarly, when the game prompts me to draw a donut, I start by drawing a circle with a smaller circle in the middle. Google hesitates, unsure what I’m going for. But when I add frosting and sprinkles, the game gets it. The lesson: To understand the whole, neural networks need pieces of data that not only connect but build on one another, piece by piece.
Of course, AI Experiments isn’t just a free education for neural network nitwits. Every interaction, be it with Quick, Draw! or one of the other applets in this virtual playground, improves Google’s ability to more nimbly recognize images and language. That makes the company’s products stronger, but it also services users. The data fuels apps like Google Photos, which uses AI to swiftly organize all your pictures. It’s a system of give and take-and with games like Quick, Draw!, it’s fun, too.