Sunday, October 31, 2010

Games That Do Some Good For The World

People spend a lot of time playing games, especially computer games. I just got an iPad, and I've been playing a lot of puzzle games, and a lot of them are really fun.

But I'm always trying to maximize my productivity, the productivity of everyone I meet, and I'd love to get everyone else on board too. There is a class of games out there that seek to do some good for the world.

Broadly speaking, games that have a function beyond the entertainment of the player are called serious games. These other functions could be education, training, or solving some real-world problem.

I've been very inspired by Luis von Ahn's work. He has a website called GWAP (Games With A Purpose). He's a genius at getting people to do productive work for him. All of the games on the site server collect valuable data for other computing applications. For example, in the ESP game, players have fun but also tag images with meaningful labels.

The ESP Game
Here's how it works. Two online players are randomly and anonymously paired. They both see an image from the web, and their job is to type in the same words. They can't communicate, and the only thing they see in common is the image. So they naturally type in words that have to do with the image. For instance, for the image at the top of this post, I'd type cookies, tetris, green, frosting, icing, wax paper, etc. If my partner types any of the same words, we get points and go on to the next image. Because we agreed on a word, it has a high probability of actually being meaningfully associated with the image. I encourage you to play it a few times.

What's good about it is that the players could be playing Tetris, or solitaire, or something that does no good for anybody. Instead, they're helping web search, and gathering important data for artificial intelligence applications. All of the games on the GWAP site follow this basic design: get randomly paired people to agree on content. It's worked beautifully.

The FoldIt game is downloadable and is based on a different paradigm. In FoldIt, players solve a puzzle, but the puzzle if informationally identical to folding a protein.

DNA creates proteins, and does little else. Proteins run almost everything in molecular biology. They come out as a chain, but fold into its lowest-energy state. This state is very hard to predict, because the space of possible folds is enormous. The behavior of the protein, however, is determined, to a large extent, by its shape. It's a major, unsolved problem in biology, and I even published an AI paper that tries to make some progress on solving it.

The great insight on the part of the FoldIt designers was to recognize that people's perceptual systems might be better at predicting how the protein would best fold than a computer. A player folds the protein on screen and the computer calculates its energy. The lower the better. The hope is that if people play this they'll find new protein structures for us. Big win.

Here's someone playing FoldIt (you don't need to watch more than a few seconds of it):

The Future
The proportion of games that are serious, in this sense, is vanishingly small. I see an opportunity. The concept of a serious game has blossomed only recently, and I think it's very fertile ground for innovation.  Now, when I play fun games on my iPad, (I'm enjoying WizardBlox and Cablink, lately, and of course I have to give a shout out to Makeout Mania) I try to think of what real world problem I could be solving by playing this game.

One of the problems is that for the computer to score your progress, or to determine that you've won, it needs to be able to solve the problem itself. If the computer can solve the problem itself, then we don't need a serious game. AIs can do all of the work for us already.

A productive game needs to do something that computers are bad at, but people are good at. Doing algebra is bad because computers are great at it. A game for solving the Middle East Crisis is bad because people are no good at it either.

von Ahn gets around the problem by making people agree. The computer does not need to check to see if the tag given is actually associated with the image (which is currently not possible), because the agreement between people does it just fine.

In the case of FoldIt, the computer can how good a given fold is, so it can score it automatically, but it's bad at deciding what folds to evaluate. This allows FoldIt to be a single-player game.

So lately I've been thinking about real world problems that can be dressed up into games that are informationally identical. 

I've developed one GWAP-style game (it is not public yet), and I have another in the works.

I'd love to hear ideas about what kinds of problems might be amenable to this. Don't worry about the game design-- I can do that. I need ideas for good problems.

This post is related to my philosophy of not wasting student work:


Pictured: Tetris Cookies

Bookmark and Share

1 comment:

Dustin Freeman said...

A colleague of mine (Shengdong Zhao at the National University of Singapore, had some students working on a project using this technique for navigation tasks. The idea is take a difficult-to-navigate environment, and map it as a dungeon level in any regular MMORPG and get users to play through it. However, I think they were having trouble coming up with any cases it worked well for, i.e. a "problem" robots had that could be mapped to a "problem" in a virtual world. Good food for thought though.

I often ponder what good/evil megalomaniacal projects I can use Mechanical Turk for. I used it for transcribing interviews over the summer, and it worked really well (if you want cheap, low quality transcripts).