#038x: Even Wins - Cybernetic Version

Release Date: July 1973

Platform: Mainframe (BASIC type-in)

Genre: Puzzle (Nim Variant)

Developer(s): Eric Peters

Publisher(s): Digital Equipment Corporation

Trust me, it sounds way cooler than it actually is.

History

This little side quest is a follow-up from last week's Even Wins article. As I stated there, 101 BASIC Computer Games has two versions of Even Wins. The one from last week's article is the original, which uses a conventional AI. This one, which is called EVEN1 and Game of Even Wins by the two versions of the book respectively, does something a little different with its AI: it learns.

This version of the game was written in-house at DEC by Eric Peters, who also developed the Lunar Lander variant, Rocket. MobyGames dubs his version of Even Wins the "Cybernetic Version" (I think based off of the game's description in-book), uses a different AI pattern that begins only knowing the basic rules of Even Wins, and progressively learns how to play the game better over successive rounds.

A quick refresher on Even Wins.

The whole idea of "machine learning" in this sense isn't new to the blog; we've previously seen this form of AI learning in a couple of previous programs. Animal is likely the most notable, although is different in the sense that the player was the one teaching the computer different questions and animals, instead of with games like Awari and this version of Even Wins, where the computer learns by itself. Animal, as a side, happens to be one of my most-read articles, though I honestly don't know why - it's not even a game. This is supposed to be a gaming blog!

The Game

Anyway, I said this was going to be a short one, so let's get into it.

Look familiar?

First of all, the game is almost entirely re-written from the original, and has significant differences as a result. Visually, the game presents the objects as chips instead of marbles. Mechanically is where the majority of changes are, however.

No longer is the pile of objects set at 27 - it's now a random, odd number. The game code suggests that a pile size of a single chip could be generated (there is failsafe code in place for this occurrence); in practice it tended to be between 7 - 21 chips.

The player does not have the choice of going first or second anymore. Instead, the computer always goes first. I find this a curious choice, with first being the advantaged position. However, it makes sense in the idea that it showcases how the AI begins squandering its advantaged position, but progressively learns the game and gets to the point where it can rarely be beaten.

Bro is pretty stupid.

On the note of the AI, it does indeed start out woefully bad at the game. It makes some truly bizarre and illogical choices, and continues in that pattern from quite some time. Sometimes it will completely blow a winning position, like being at 5, myself taking 1 chip, being on an even number, and it'll just take the whole pile like it's giving up.

...but he does get better. Eventually.

Eventually, it does start to get good, and did start taking game off me. However, I seemed to have found a flaw in its algorithm, as there was a particular strategy that it just couldn't figure out. I would force the pile, if I was on an even number, to 9 chips remaining. From there, the AI would always take 3 chips, putting me in a position I would always win from. It never learned how to beat this strategy - I must have beaten it 10 times in a row at least.

Closing Thoughts

No scores for Even Wins: Cybernetic Edition. For one, I think it's somewhat redundant as it's functionally the same game as the source. It's more of a showcase of AI programming, which is inherently interesting to me and worthwhile not skipping over just because of "redundancy." Every game matters.