# How to build a simple computer from a cardboard box (or 2) and some marbles?

Quite a few years ago (around 36-40) I read about a simple computer, made of a cardboard box, some dividers, and using 2 groups of colored marbles; whereas the object was to tilt the box  and one marble would fall through.  Depending on which of the 2 colors came through (pretty random) a move was made in the game (a simplified version of tic-tac-to).

Wrong moves were removed and eventually the "computer" learned to win the game every time.

BUT I can find no reference to this, nor the book it was in.   I don't remember the title nor author, but it was very much like the book:  "1, 2, 3...Indiniry" by Gamow (it is not this book, I have it and have read it through several times looking for it).

I have been looking for this info for about 20+ years, and decided one of the more sciencey readers amongst you all might remember  it, the book or something to give me a clue to where to find this.

sort by: active | newest | oldest
rickharris4 years ago
http://stackoverflow.com/questions/10705044/which-algorithms-are-available-to-solve-tic-tac-toe
Goodhart (author)  rickharris4 years ago
My cardboard box would not understand that. :-D
bwrussell4 years ago
How would it "learn"? The out put is random so when you put the marbles back in the box for a new game you could very well repeat the exact chain of moves every time you play.
Goodhart (author)  bwrussell4 years ago
It learns because, each "bad move" is removed (the game it plays is MUCH simpler than TicTac To, so the variables are minimal) at the end of each "game". Eventually, it only gives winning moves.
4 years ago
That only works if the game plays out the same way every time and the marbles fall out the same way every time, minus the ones you removed. If the marbles come out in a different order then removing the bad moves from Run 1 does not change anything in Run 2 and in fact after several runs you could end up "teaching" the computer in such a way that it can't ever win.

How can a game be simpler than randomly placing a piece in a square (tic-tac-toe)?
The only variable is which space to place it in. At the end of each turn either someone has won or they haven't. If you remove any more variables it isn't a game, it's repeating a pattern. Since it's a game then there is an opponent which brings a level of complexity you can't combat with cardboard and marbles alone, no matter how simple the game.

If the game is so simple that any bad move can be tagged as bad as soon as it is made then what advantage does the computer bring? If that is the case then the outcome is wholly predicated on the order of the balls. This means the user knows the best output before the computer and you might as well just load the correct sequence in the first place.
Goodhart (author)  bwrussell4 years ago
The game is simpler in that, although there are still two choices, the board has much fewer squares then tic tac to

The advantage of this computer is a physical demonstration of "survival of the fittest". It demonstrates that, when an organism drops unuseful or detrimental attributes by means of death (losing the game) the organism soon becomse "much better at the game" winning or drawing every time.
4 years ago
So it's not really a computer, I think the use of that term threw us off. The only way that I see this working is you have two marble colors and the point of the game is to get all of one color. After you play once you remove all of the losing color and put the rest back in the box. Eventually you will only have the winning color. For this to work you have to control every variable and the only variable you can control is what marbles you load into the box.
Goodhart (author)  bwrussell4 years ago
It computes, but ok, sorry.

Well this is why I needed a reference to it (or the the book it was in). The process was slightly more complicated IIRC, but not much more. It HAS been at least 40 years since I saw the book (which indicates the age of the book approximately).

It kind of bothers me that I have forgotten something that I had thought about on and off for awhile now.
4 years ago
Mechanical systems can't "learn" unless you adjust some variable as it plays. I can't see how this might work with marbles,
Goodhart (author)  rickharris4 years ago