'Keen' is Pinterest with machine learning and the power of Google Search

Google has quietly launched Keen, a service akin to Pinterest with one twist: It crawls the web for you using machine learning coupled with the power of Google's Search Index and surfaces content you might be interested in based on the topics you've told it you care about. The idea is you get a personally curated selection of links and content that can be as hyper-specific as you like. It's like Pinterest crossed with Flipboard for obsessives and hobbyists.

For instance, maybe you want to learn how to make sushi at home. With Keen you can create a new "Keen" — which, yes, is a lot like a Pinterest board — and give it the topic name, like "Sushi cooking." Keen will then suggest some web search terms related to your interest and use those to find content you might like.

The machine learning magic suggests additional, adjacent search terms related to what you've asked for. So, for instance, the Keen we created for sushi suggested we also add "sushi cooking kit" and "sushi cooking tutorial" to our list of interests. You can pick and choose these terms and change them after your Keen is created if you're being presented with links that aren't relevant, or aren't specific enough.

Find the hidden gems — Within your Keen you'll find a board of web content that matches your interests. When you see a link you like you can save it as a "Gem" so that you never lose it. Basically, Keen does the initial work of finding interesting content tailored to your hyper-specific interests, then you do the work of sifting through all of it for the best stuff.

Every Gem can have a description explaining why you've saved it, and you can also break up a Keen board into sections, such as a section for sushi how-to videos and another for recipe lists. You can share your Keen with others and even invite people to collaborate on it with their own links.

Alerts addicts, this is for you — If you've ever used Google Alerts to keep track of a topic, Keen is a more visual version of the same concept. Alerts take search queries you're interested in and periodically emails you when new results appear in Google search. Now, with Keen, you have a home where you can save and curate these links in an organized, graphics-heavy fashion. Keen learns from what you save so it can better curate your recommendations over time.

Keen was built by Area 120, Google's in-house incubator for new ideas. The division has released a slew of experimental products over the years to varying success — Google News came out of Area 120, but so have products that came and went without most people noticing, like Shoelace, a "hyperlocal social network."

Algorithm worries — Google has a spotty track record when it comes to algorithms, however, taking heat over the years especially on YouTube for sending users down rabbit holes of progressively more fringe and conspiratorial content. These sorts of things happen when recommendation algorithms are designed to serve up more of whatever you respond well to — if you liked one crazy extremist video, it'll look for more of that without considering the implications.

Pinterest already has a massive base of users doing curation on its service, but Keen is starting out from zero and algorithmically surfacing content is a cheap way to kick-start things. Plus, Google already has the world's most comprehensive models for linking related search terms together. Users on Keen can curate their boards, but the content presented to them is initially surfaced by an algorithm.

That could prove problematic if Keen turns into another channel through which people become radicalized in their views and aren't exposed to any viewpoints outside their existing echo chamber. Or worse, the more extreme views that exist on its edges. It's also going to make Google's massive advantage when it comes to the data it has on users' habits even more unassailable. Though, for this to happen, Keen will need to take off and not suffer the same death in obscurity experienced by services like Google Wave.