Tonic is a new app with a radical approach to recommendations personalized for you: It doesn't want to know who you are, make money off your interests or hook you on using its app more and more. Instead, Tonic wants to recommend five, and only five, things every day, like an article or photo essay.
It's definitely the first app I've ever been pitched that might recommend me a poem.
It's also the first app to launch from, a startup comprising veterans from companies like Spotify, Instagram and The New York Times. Many came to the startup from tech giants, disillusioned by how an earlier promise of the internet has turned dark or exploitative. Canopy's mission is figure out what recommendations are like when they're private by default and aren't optimized for addiction.
"We're all feeling pretty down about the state of the internet ecosystem, the apps we use and our relationship to our devices," Matthew Ogle, Canopy's head of product, said in an interview and demo last week.
Your options, he said, shouldn't be either checking Twitter every five seconds for the latest hate reads or throwing your phone in the sea. "There's still good stuff on the internet, and there are other experiences that are possible."
Society has spent the last few years re-examining technology's hidden costs, exposed by some of the nightmares tech has created. Personalization, for one, is built on the premise that tech companies can delight you with recommendations if you let them get to know you. But that approach has evolved into users handing over troves of intimate details about themselves, without much visibility or control over how that data are used.
How Tonic works
Tonic's recommendations are built by a combination of on-device machine learning and differential privacy. To understand how they work, imagine the recommendations Tonic suggests as a jigsaw puzzle. Your device is the only part of this system that knows your jigsaw puzzle should be of, say, a picture of kittens. But your device tells Tonic's servers to send all kinds of puzzle pieces -- for kittens and puppies and baby otters too. Then your device discards all the puppy and otter puzzle pieces and uses only the correct kitty pieces to put your puzzle together.
That's one way to think of Tonic's system, where all the behavioral data needed for recommendations stay on your device, and a model of what you like is sent to Canopy's servers masked by other information. While most other recommendation systems ingest your data and do all their algorithmic crunching on their own servers, Tonic's servers send back a bunch of possible recommendations and then the final selections are finished on your device, because that's the only place where the model of your specific preferences lives.
Tonic and Canopy never know who you are, or even that you love kittens, but you still get your kitty jigsaw puzzle put together.
Human vetting is also another backstop. While people don't curate your recommendations, humans vet all the content that's in the pool of possible suggestions on Tonic. The company said that so far Tonic has recommended articles from more than 550 publications and websites. The majority of recommendations will be articles, but the app will add in other forms over time, like photo essays, digital art, videos and short poems.
The app is free to download and use starting Wednesday for Apple's iOS, with an Android version coming within the next six months.
So what's in it for Canopy? The company may launch paid apps in the future, but its business model is to license its recommendation technology to partners. The idea is to power the personalization of companies that can't build something from scratch that stacks up against Facebook. Tonic is the first way Canopy will see how its approach to recommendations works when unleashed into the wild.
"We hear a lot about creating a better internet for people, and for that to be successful, it's a shared endeavor," Annika Goldman, Canopy's senior vice president of strategy and operation, said in an interview last week. "We're not going to do it alone."