Larry Learner: An AI Program

Tony Kuzola: Designer
David Leech Anderson: Designer
Robert T. Arrigo: Programmer
Andy Schmidgall: Programming

Larry Learner is an artificial intelligence program that plays a simple game with 10 pencils called, "Last One Loses." There are two versions of Larry. The first version is a very accomplished player. If Larry goes first, you will lose. If you go first and make a wrong move, Larry will also beat you. There is a second version of the program called Larry Learner. Larry Learner doesn't know how to win, at first, but it learns. It gets better as you play it. You can play both Larry & Larry Learner in a browser (they are Java script programs).

Larry is good at "Last One Loses" but is Larry intelligent? Larry is very effective in classroom situations to help students explore the possibility of machine intelligence. Can a Machine Think? is a module which uses Larry Learner to explore questions around machine intelligence.



The game Last One Loses can be played with any number of pencils (or any other kind of object for that matter). We use 10 pencils. On each move you must take at least one pencil and you can take no more than three. (Thus you have three legal moves: You may take 1 or 2 or 3 pencils.) The object of the game is to force your opponent to take the last pencil. The game is just that simple.

Larry (Winning Version)

This version of Larry is programmed with the winning algorithm. If Larry starts, he will never lose. If you start and make a mistake, Larry will win.

Larry Learner (Learning Version)

This version of Larry is not programmed with the winning moves. All it knows are the rules of the game. Over time, with trial and error, eliminating losing moves it eventually "learns" to win the game.


This module was supported by National Science Foundation Grants #9981217 and #0127561.