Flickr Co-founder Unveils Her New Startup: Hunch

For the decider, there is the appeal of having a system to approach a difficult decision; of learning from others; of mastering a challenge and having confidence in your decision.

For people responding to a question and refining the decision tree, there is the appeal of demonstrating expertise and understanding how other people think.

And there is a more abstract appeal for deciders, responders, and just plain readers: validation that you're facing a challenge that you shouldn't be ashamed of, that you can ask for help with, and that can be overcome.

I'm not sure if hunch really does appeal effectively in these ways (I've only just requested an invite) but the potential is there. Starting with that sort of potential and really listening to users seems to be how flickr evolved into a product+community that is really prized by its members/customers, and another app made with that same level of attention and attentiveness could really earn a lot of devotion.

The large-scale analysis and AI could certainly be amazing -- but if you want to learn from this as a model for your own future work, pay attention to how well this app connects to the emotions of its users, and how they grow the community around it. If you're making something that doesn't have a straightforward save-time/money selling proposition, making those things work is your biggest challenge.

What should I eat for lunch?

A. McDonalds (sponsored by McDonalds)

B. You get the picture...

Crowdsourced machine learning isn't new, but because Hunch has enough weight behind it, it might eventually become much more useful than a lot of other 'silly crap' out there. A native iPhone client for Hunch would then be quite welcome.

For example, I don't use Twitter right now. I don't have a need to write 140 character messages to the general public. But I have changed in that I now see that Twitter could improve some situations in my life. I could imagine using it to send text messages to all my friends to see if they wanted to grab lunch. I could also see myself getting it if a future boss or coworker used it extensively. So while I could not see much value in Twitter when I first heard of it, I see value in it now.

So I am just assuming that I either don't fully understand how I could use Hunch, or I don't really need it right now. That just means I am not going to be an early adopter of their service. I would not say that the service is total crap.

Even now I can see some situations where I could use expert systems to make decisions. I remember when my parents got a new car my dad took the cost of the car, the mpg for the car, and the amount that he drove, and used that to find out which car he expected to be the cheapest. I am sure he would be willing to input some numbers into a web page and use the site's recommendation instead of doing all the work himself, assuming he would be able to actually view the decision tree.

Everyone has trouble making decisions sometimes--it's human nature on contact with a world more complex than the one we evolved in, not a character flaw. I don't just mean major decisions either--sometimes the hardest ones to make are things like where to have dinner or what movie to rent or what color shirt to wear. People are constantly encountering situations in which the cost of making an optimal choice exceeds the consequences of making a sub-optimal choice.

More generally, this is the problem of bounded rationality. If you can't make perfectly rational decisions, you have to find efficient heuristics to approximate those decisions by reducing the amount if information that has to be collected and considered. We've reduced the cost of information gathering to nearly zero, but that only exacerbates the problem of processing it into something useful for making decisions.

Most attempts to mitigate this aim to rank, rate, and prioritize, in order to narrow the firehose and give you only the relevant information. This is great if you know what information you need to make the decision and how you weight various aspects of that information. It's useless when you don't know how to decide: what information to consider and how to rank it. It's like having a research assistant when you want an expert. So what is useful in that situation?

It turns out we already have technology for that: expert systems. They're good for reasoning through a problem because they actively ask for information as it becomes relevant rather than requring the user to know beforehand what is relevant. They're great for situations involving uncertainty because they don't require answers to every question in order to provide potentially useful results.

Where expert systems fail it is because they lack common sense, breadth of knowledge, and the ability to adapt and expand themselves. These are all things which the collective intellegence of networked humans have demonstrated great aptitude at. Driving the former with the latter seems likely to produce interesting results, even if Hunch itself fails.

I have no idea if Hunch will be "huge" or hugely mediocre -- that is heavily dependent on the yet-unseen functioning of the site itself and users that contribute to it (the who, the why, and the how), but I at least find the approach novel and interesting, which is more than can be said for many startups.

Must be nice to have a huge name to tag onto some silly idea that'll make insane money no matter what.

Not as nice as it is to take shots at people for being successful, right?