= 27d567bc2d48d9f40e9ccc20ea370b15)
Emile comments on Rationality Games & Apps Brainstorming - Less Wrong Discussion
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (58)
Graphical calibration game
1) Show the player an image with a bunch of simple images: rectangles, smiley heads, circles of various color, etc. for about 5 seconds.
2) Hide the image, and ask the player to give a 90% confidence interval for a value such as the average size of a figure that he saw.
3) The correct answer is then shown, along with the image.
Variety can be added by asking for median size, or average width, or whether there are more red or blue circles, or correlation between width and height, or between size and color (if the circles all vary from light blue to dark blue), or the average size of red circles, etc.
This allows for very tight feedback loops between guessing and seeing the answer, and the game can be replayed pretty much infinitely.
This could be important. Some card games teach calibration, such as Bridge and Spades. (Although it's not quite the same, because after you guess how many tricks you'll take, you have some control over it - if you were underconfident, you can throw tricks away, if you were overconfident you can take unusually large risks.) But they just ask for a single number, and later you see how close you were but you can't look back and see how confident you were. If you give a confidence interval, it's much easier to see whether you're well-calibrated.