Music is personal. It's tied to our identities, our emotions, even our friends. So the idea that a complex algorithm could make us smarter about music is counter-intuitive. The creators of Cone, a wireless speaker that learns what you like and builds on it, think they've cracked the code.

Cone is the first product by Aether Things, a San Francisco-based company that draws on decades of research in machine learning to build "things that think." When I met its Chief Product Officer Duncan Lamb last week, he summed up the company's mission with a question: "How is it that we've got all this power and all this data, yet we still have to tell our machines, word for word, what we want them to do?"

That's a tall order, but it's not as tall as it would have been just a few years ago—before products like Twine, Automatic, and Nest Thermostat made the concept of thinking things accessible to the public. The Aether team comes from companies working on the edge of this burgeoning field: Nokia, IDEO, Google, Apple, and even a NASA alum who deals with Aether's voice recognition ("he builds voice-controlled drones as a hobby," says Lamb).


Lamb, who himself worked as Head of Product Design at Skype and Creative Director at Nokia, describes Aether as a group of people who go out "looking for trouble." And Aether's first product, Cone, responds to the first trouble they found: Streaming music.

Too Much Choice

So, what could possibly be wrong with services that make it possible to stream any song at any time?


That's just it: With so many degrees of freedom, what used to be a simple act of flipping a dial on a radio set while you're making coffee in the morning is complicated by toggling between computer and smartphone, making playlists, staring songs, and shuffling through choices. "As stuff gets more capable, it gets more complex," Lamb explains. "And that complexity gets dumped straight onto us. TVs are a perfect example. In the 50s, there was a nob for on and a nob for tuning. Now, we have four or five remotes."

Cone wants to simplify the process down to a single action: Just turn the dial. The three-pound speaker has a circular face that functions like a big click wheel. Turn it to the right, and you hear music. Turn it one more time to change the song. If you want to be surprised, give it a big, Wheel of Fortune-style spin. If there's something specific you want to hear, just press the button in the middle of the speaker and request it. At my demo, asking Cone to "play the Rolling Stones" had the Stones blasting within a few seconds of the request.

It's also beautiful, with a smooth copper skin and a curving profile that hearkens back to antique phonographs without seeming corny. When it ships this spring, it'll arrive with a 1GHz processor running on a Lithium ion battery that packs eight hours of play before it needs charging. Right now, setup is iOS and OS X compatible only, with Android coming down the road.

Most importantly, its three-inch woofer delivers crisp, warm sound—even at high volumes—which backs up its steep $400 price tag.

The Machine Behind the Music

Those details are about all you need to know to interact with Cone. But behind its copper skin, it's hiding a complex piece of propriety software written over the past two years by Aether's development team.

You see, Cone listens to what you like, what you skip, and what you request, and builds a profile based around your activity. Aether hasn't confirmed exactly which streaming services Cone will pull from yet, but it says it'll announce multiple content partners, along with podcasts and internet radio, before the device ships later this year.

Cone also learns from how you use it. Thanks to its built-in accelerometer, it generates data points about where you listen to what, as well as when you listen to certain things, and bases its playlist on that information.

For example, if you listen to NPR in the mornings in the kitchen, Cone will know to play that rather than music. If you're particularly prone to instrumental electronic tracks while you work every afternoon, it'll learn that too. Cone's algorithms also include ambient data points. It might suggest something else if it's raining, say, or if you turn up a particular song.

The more you use it, the smarter it gets. The idea, says Lamb, is to "take all this choice and all this power that exists and is available to us in different forms, and put it in a physical forms that are really simple, really natural, and really direct."

A Teacher or a Tyrant?

The broader question about Cone is whether people are ready for a device that knows what they want to listen to before they even know it. We give other "connected" products control over aspects of our lives that we don't want to think about, like the temperature of our house, our car's gas mileage, or when we should water our plants. Music, on the other hand, is bound up with our personalities, and is closely related to mood and emotion.

Aether's decision to build a new piece of software from scratch, rather than partnering with a proven company like EchoNest, is a risky one. Existing services have millions of users and years of data to back up functionality. Cone, on the other hand, will by flying solo. That might not be a bad thing, depending on how sophisticated the algorithm is, but it could be a pitfall, too. Another issue for some users might be related to portability, since not everyone wants (or needs) a speaker that's sensitive to its location.

Hopefully, Cone will feel like a trusted friend who likes the same bands as you and always seems to know about the next big album, and not like a tyrant who can't be predicted or controlled.

From what I saw over the brief hour I spent with it, the results seemed promising. Still, without having had the time to let it "learn" about me, it's hard to say anything definitive about how it'll really work. But if it functions as it's meant to, Cone could be a magical piece of hardware that hides a complex piece of software—designed not to be seen, but to be heard.