We may be a small step closer to the robot wars thanks to former Apple designer Mike Matas.
Matas’ previous work includes user interfaces for Apple’s Maps, Photos, and Camera apps, as well as the Nest smart thermostat. And he showed off his latest creation, an artificial brain called (appropriately enough) The Brain, via a quick demo on YouTube. It’s a neural network that with an expectedly sharp and clean interface, and in the video, he shows how he can teach The Brain to spit out emojis based on different shapes that he draws.
Check it out below.
This is a pretty basic test, to be sure, but it’s nonetheless impressive because of the simplicity of the interface and the ability to actually see the machine learning over time. Matas starts by telling the computer that it will learn to output various emojis based on shapes, and then he shows it the shapes that correspond to each emoji. So a smile-like, curved line produces a happy picture, and a Z shape represents sleep.
“There’s no preprogrammed rules here that are telling it a curved line means this or that,” Matas says. “It’s all being figured out automatically based on the examples we taught it.”
You can see this concept at work when the program sees a shape that it doesn’t “know.” At around the 1:20 mark, Matas draws a heart and a teardrop, neither of which The Brain has learned yet. So it makes its best guesses and tosses out a smiley face and a frowny face, respectively. Matas goes back to the teaching mode and provides the proper responses for those shapes, and it doesn’t mistake them for others anymore.
The video also shows off the individual connections and processes that go into associating this shape with that emoji, and the workings underneath that super-clean interface look impressive and complicated.
Matas built The Brain in Quartz Composer, a node-based processing language that’s part of Xcode, the suite of tools Apple and developers use to create both iOS and OS software. We aren’t entirely sure what it can do outside of emoting, but Matas sees a lot of potential.
“It’s sort of crazy on the inside, but on the outside it’s pretty simple,” he says. “The same process that allows it to learn and recognize what this shape means also allows it to learn a whole lot of other stuff. And so it would be interesting to think, if this kind of thing were made a little bit easier to use, what kind of ideas might come out of it.”