Intro “machine learning” for kids via Martin Gardner’s article on hexapawn

Last month I had the nice surprise of finding Martin Gardner’s book The Colossal Book of Mathematics at the Omaha Public Library’s book sale:

I’ve been flipping through the book and thinking about how to share some of the ideas with the boys.  Chapter 35 – “A Matchbox Game-Learning Machine” really caught my attention.  In particular, Gardner’s discussion of the game “hexapawn” inspired me to try this introductory “machine learning” idea with kids.

I had the boys read the (approximately) 2 pages on the game and the approach and then we talked through the game to make sure they understood it:

Next we started playing. We were very lucky to have a coffee table that allowed us to easily show the 24 cases and their snap cubes. This video shows the first two times through the game. I hope that it shows that playing through the game is something that is accessible to kids:

The next part shows 3 more turns of the game. My main reason for showing these three turns is so you can see some of the parts that kids find challenging. I think these parts are a big part of what makes sharing Gardner’s idea with kids so fun. The pattern matching and the general walk through the game keeps their attention while they learning about machine learning.

Next we played for a while with the camera off. After a while the kids (and the computer) learned something:

Next we played a bit more with the camera off and the before long the “computer” learned to win the game every time. Amazing!

In the last 3 min of this video the boys talk about some of the things that they learned in this project.

This is one of the most fascinating projects that we’ve ever done. It does require a bit of set up and probably a bit more careful supervision than usual to make sure that the kids don’t go down the wrong path, but wow is there a lot to learn here. I think that opening the door for kids to see how computers / machines might “learn” is an amazingly valuable lesson.