Nice. Like.com is owned by Riya, which once tried to solve the facial recognition problem but then gave up. Now they’ve turned to an application where their less-than-accurate technology is acceptable – shopping.

Here’s how it works – you search for a product on their site (jewelry, watches, shoes) and then it will show you products that look like it. Try it with watches or women’s handbags; it’s quite nifty.

Aside from the visual similarity algorithms, this is what I find compelling: attribute search. Narrowing search by color, style, price, material, etc is extremely valuable and something Amazon is not very good at. If Like.com can be the world’s authoritative source on attribute data, that’ll be a huge win for them.

The quick summary: Each day, Jellyfish will choose one product as it’s Smack of the Day. It’s a dutch auction, which means that the price starts high and drops every few seconds. So what, you ask? Dutch auctions have been done before, but here’s the twist: there is an undisclosed quantity of items available for sale. In addition, there is a real time chat so you can communicate with others watching the same auction.

Here’s why that’s important. When there is only one item for sale, once someone decides to purchase the item, the auction is over. Hence, auction participants have a clearly defined understanding of when the auction ends and how to end it. With Jellyfish’s Smack, however, deciding to purchase the item does not necessarily end the auction; instead, the is one less unit available for sale and the price continues to drop.

Let’s think about that for a minute. First, let’s assume that there is more demand than supply; someone isn’t going to win. On the other hand, it’s in everyone’s best interests to cooperate and hold off buying anything until the price drops to $1. This sounds suspiciously like an iterative Stag Hunt game.

So now let’s work out the SH game. Assume person B believes person A waits, then the optimal strategy is to Wait. If person B believes person A Buys Now, the optimal strategy is to also Buy Now. The same is true for person A. If person A believes person B waits, then the optimal strategy is to Wait. If person A believes person B Buys Now, the optimal strategy is to also Buy Now.

Solution? Participants will choose the payoff dominant solution over the risk dominant solution if they can punish those who Buy Now.

Flaw: it’s not iterative for the individual. Once you buy what you want, you’re out. Only iteration is across days. Hm. The punishment, then, will have to carry weight across days. That’s a bit tricky. In the interest of publishing this post, I’ll leave that problem open for another day.