Hunch

I signed up for hunch, and have been play­ing with it a lot over the past cou­ple weeks. It’s a fun sys­tem, very addic­tive at the begin­ning, while you are answer­ing the ques­tions about your­self (called “teach hunch about you” or THAY questions).

How it works is pretty inter­est­ing. Top­ics are the big ques­tions, “What lap­top should I get?” etc. Under these top­ics are ques­tions and results. Ques­tions are the fil­ters that lead to results. Things like “Should the mon­i­tor be big­ger than 17″?” for exam­ple. Results are the solu­tions. There are two types of results, yes/no, and spe­cific things. In the lap­top exam­ple, a result might be a Mac­Book Air. Some ques­tions, like “Should I learn to ski?” would have just two results, yes and no.

Results are “trained” which means for each result some­one goes through all the ques­tions in the con­text of that result and says which spe­cific answers apply to it. For exam­ple, the topic “What font should I use?” would have a ques­tion, “Do you want serif, or sans-serif?”. The result Palatino would be trained in each ques­tion. For exam­ple, Palatino, would be trained as yes for that ques­tion. The train­ing is done mostly by humans, but if the algo­rithm sees some­thing it thinks cor­re­lates, it will train on its own. The rank­ing of results is based on a com­bi­na­tion of the train­ing, and what peo­ple like you have rated things. (people-like-you being based on the THAY questions)

It works sur­pris­ingly well really. Even with­out the train­ing, and ques­tions to guide it, it can do a pretty good job bas­ing it’s answers only on people-like-you data. The team is also very respon­sive to sug­ges­tions and things in the forums, and is iter­at­ing a lot. It’s fun to see employ­ees so obvi­ously excited about the prod­uct they are making.

If I have one crit­i­cism, it’s that there isn’t enough to do, and no great ways to tell peo­ple what needs to be done. There’s also no good way to com­pare work and cre­ate any feed­back loops. I think these things help build a real com­mu­nity. I’m think­ing of this a lit­tle like Wikipedia, and maybe that isn’t a cor­rect com­par­i­son. Maybe they don’t need a com­mu­nity in the same way really. It’s cer­tainly very easy to use, and they have built the tools for peo­ple to cor­rect their own results. Maybe that’s enough.