Using the Wisdom of the Crowd in a Computer Game
How does the mind represent the sizes of things?
Posted Apr 07, 2014
The field of cognitive science has made it quite clear that we represent things with symbols. From linguistics to problem solving, people reason with variables, and the values they have. For example, in English, we have different words for the colors red and orange. We classify colors according to the words we use--other cultures have different color words, and they see the world a bit differently.
But we also seem to be able to represent magnitudes. For example, when playing hide and seek, we can look at a box and know if the child we are playing with could fit in it. We have some memory of lengths, weights, sizes, and all sorts of magnitudes. How do people represent them? We don't know.
My laboratory is trying to make computer models of imagination. Part of this involves being able to re-create realistic but novel scenes. For example, if I ask you to picture a living room, you might picture a particular living room you have seen before, but would also be able to picture a new living room that you've never seen before. But this room would have common characteristics of living rooms in general.
To get a computer model to be able to do this, we require knowledge of distances and sizes of things. But there is no database of these quantitative magnitudes. So even if we knew that an imagined living room should have a sofa, chair, TV and coffee table, how do we get the system to know how big these things are, or where to put them, how to orient them, and so on?
One way to solve this problem is by getting people's opinions on these quantitative magnitudes through a fun computer game. Our laboratory has made such a game, called Quanty.
In Quanty, players are randomly paired with one another, and asked to provide guesses about quantities such as heights, lenghts, and weights from photographs. For example, you might see an image of a duck and be asked the weight of the duck. You and your partner both guess, and you get more points if your answers are closer. You go through as many pictures as you can in three minutes.
How does this provide us with the magnitudes we need to make an imagination engine? Because of the wisdom of the crowd. It turns out that with everyday things, the average guess of a number is often closer to the truth than any individual's guess. So if we have enough people playing the game, the average guess, of, say, the weight of a duck, will be pretty close to what a duck actually weighs.
In the picture, the players are asked about how true it is that the tea box is between the switch and the coffeemachine. Quanty also asks about English prepositions such as "between" so we can get quantitative data on what people really mean by the term.
We did a study of this kind of guessing that Quanty uses, and found that the results are pretty accurate for everyday things (not so much for the weight of a building, for example). We also asked participants how fun the game was, and people said it was as fun as Tetris! Though I kind of doubt it's that fun, the game is more fun than you'd think. It's like playing trivia--it's fun to guess.
Feel free to give it a try--you'll not only have fun, but you'll be helping the science of imagination at the same time! The results of this study are being submitted to the Human Computation conference.
(If you can't play, it's because nobody else is online at the same time--you can try again later.)
Jim Davies is author of Riveted, available now for pre-order.