Hunch (www.hunch.com) is a new "decision engine" that uses fancy math and user input to build a profile of personal probabilities aiming to help you figure out what to do when you're stuck for an answer. Hunch made its public debut on June 15, and as of this writing, it has 489 members in the official Facebook group and another 1,726 followers on Twitter (doubtless there is some overlap there). It's gaining popularity very quickly. And although only time will tell if the buzz will carry it through the long haul, the fact that Hunch is headed up by Caterina Fake (co-founder of Flickr), chief scientist Hugo Liu, and "a bunch of MIT nerds" with very impressive web cred bodes reasonably well for its future.
A Tour of Your Mind
Here's a walk-through of a Hunch experience. When you first visit Hunch.com, you see the message "Hunch helps you make decisions and gets smarter the more you use it" above a box that asks whether or not you like bumper cars. I do like bumper cars, so I click Yes. This is all part of "teaching Hunch about you"; there are twenty questions total in this first, most basic phase of establishing familiarity with the engine (and in establishing familiarity with you). Which do I prefer, it wants to know, "concrete tasks" or "less well-defined duties." The latter. The next question is about French fries (I skip it-my favorite style, steak fries, will remain an area of mystery for now). It asks my age, whether I like to work with numbers, and who I find most attractive (from a choice of four, I click on somebody who must be either Sandra Bullock or Eva Longoria). Then it asks "do you sing in the shower," questions about where I live, and questions about how I think and work (do I read instruction manuals or just wing it, etc.). Sometimes a question about alien abductions sneaks into the mix.
Once through the first volley of profiling, Hunch is ready to guide you through some decisions. There's a ready set of questions posted by other users (users who have created profiles, which we'll get to later); if none of those look interesting, you can click the "explore" tab at the top of the page to look at a directory of topics and the most recently posted questions. The search box in the upper right matches your term to any existing categories or questions.
I click through to "Which desktop Twitter app should I use?" in the Computers & Internet category. The questions are about screen real estate, operating systems, the cost of the app, the customization of search, direct messaging of other Twits, and a few others. At each point, about 3-6 answers are offered; you choose the one that best fits. This is called a "decision tree" in the business of interface design. Hunch suggests I'd like Qwit, Snaz, or Snitter. So far, this is all pretty smart and looks alright, but there's always room for improvement. You know, Qwit looks pretty good, so maybe I'll trust it with helping me pick a dog one day.
If you click "Why did Hunch pick this," you see how your answers have forced the engine in a direction of fewer and fewer options. "Would you pay for the app? You answered: No." That cuts out all the apps that have a cost, and so on.
"Bing & decide," by the way, is all hat and no cattle compared to Hunch, in my estimation. At best, the just-launched Microsoft Bing (www.bing.com) ends up being a very adequate "organizational engine," building in tabbed categories to help you see various levels and dimensions of the subject you're searching on. Bing is more serious than Hunch in that it assumes you want to end up booking a hotel at the end of a search session. But make no mistake-Hunch is the actual decision engine, not Bing. There's not a lot new from Bing that Clusty (http://clusty.com), the clustering search engine from Vivisimo, hasn't been doing for years. (For Bing details, see the NewsBreak "Microsoft's New Bing-The ‘Decision Engine,'" by Greg Notess at http://newsbreaks.infotoday.com/NewsBreaks/Microsofts-New-BingThe-Decision-Engine-54514.asp.)
Create a Profile and Participate
If you create a profile in Hunch, a host of new options and tools appear to you. For example, there is a "Remembered Answers" area in the lower right of your profile page where all of your responses to Hunch questions are stored. This can be set to private or be made publicly viewable. The Remembered Answers section is the base matrix Hunch uses to build a more sophisticated profile of you and of your probable preferences.
Chris Dixon is one of the co-founders and head of the business side of the project. He pointed me toward a blog post that illustrates one of the fundamentals of Hunch. "Broken Legs and Fake IDs" (http://blog.hunch.com/?p=1454) shows the importance of correlating personal stats and building predictors from those correlations. Users who answered yes to the question "Have you ever broken an arm or a leg?" are more likely to enjoy Madden NFL 2009 over Katamari Damacy. That's just how it is. It's not a 100% accurate map 100% of the time (I've had a broken arm, and I don't like the sports video games myself), but it's a strong enough correlation to move the decision tree on to the next crotch. What's one of the best predictors of whether you should switch from a PC to a Mac? Turns out if you like to dance, you're more likely to like a Mac.
Naturally, there's a secret and proprietary algorithm at the heart of all this. But what is open and mashable is the code for developing new Hunch apps. Here's the way they spell it out on their developers' site (www.hunch.com/developers):
"Hunch allows access to its data and embeddable content through an API. Developers can make use of this API in many ways, such as:
- Putting an embeddable version of a Hunch topic on your web site. For example, if you have a blog about dogs you could embed our ‘which dog breed is best for me' topic in your sidebar. Note: To embed Hunch on your site, you can also use the widget Hunch provides by copying the Embed code that appears on each result page.
- Porting Hunch to a mobile device or other platform such as SMS, Twitter, IVR etc.
- Querying our ‘taste database' for correlations, predictive variables, and more. For example, you could ask our taste database for the correlation between people who own guns and the people who like SUVs, or what the most salient personal trait is for predicting whether someone likes the movie Napoleon Dynamite. Over time we hope to provide a complete database of people's aggregated taste preferences. Note that all data provided here is anonymized and aggregated to protect our users' privacy.
Our API is free for non-commercial use. We ask that you attribute Hunch by providing a link to our Web site wherever you use our API."
Guessing the Future
As serious as the Hunch team seems to be about respecting its users, it is just as serious about building this as a viable and popular web tool in the long term. Dixon says, "All the founders sold previous companies (Caterina sold Flickr to Yahoo!, me and some of the other guys founded SiteAdvisor, acquired by McAfee) and this time we are very intent on building up an independent company/website."
It seems to me that Hunch is set to move forward more quickly into the world of apps on Facebook, etc., than to remain a stand-alone powerhouse. But we'll see. Welcome to the web, pop-oracle app. Now I know which foreign language I should learn next (it's Persian, of course).
TechViShow's interview with Hunch's Hugo Liu at www.youtube.com/watch?v=_9Orh4NiUmw.
Microsoft Bing promotional video at www.youtube.com/watch?v=et0rUzRAXGE.