Kirk Report showed off a game called Guess-the-Google. The author presents a bunch of pictures and the user is encouraged to guess the keyword which spawned them. I remember seeing something like this at MS research, but I can't find it for the life of me.A machine seems "smart" if you tell it to do something and it does it. The more arcane or odd that request is, the smarter you will think the machine is. But the nice thing about structured data and the Internet is that search engines receive billions of odd or arcane requests every day. All you have to do is figure out which the user thought were successful and you can just mine those results into the machine. That's exactly what a game like this provides. It's fun to play, but what you're really doing is typing the keywords that should return those pictures. Unless you're lying, which is possible for one person but practically impossible for thousands of people, when you see a bunch of orange pictures and you type orange first, the machine now knows that's what it should present when someone types orange. It's a nice way to take a lot of distributed knowledge and perceptions (namely in our heads) and record it in such a way that it's useful for a lot of other people.