AI-assisted Abstract Game Design

General discussion on board games
Post Reply
User avatar
MarkSteere
Posts: 279
Joined: 09 March 2021, 19:21

AI-assisted Abstract Game Design

Post by MarkSteere »

AI has facilitated some designs I couldn't have achieved on my own.

Unitile (alpha) - Like the sliding tile puzzle, but two-player game with two colors of tiles, red and blue, initially arranged in a roughly checkerboard pattern, with the center tile missing. Object: Slide your tiles together into one group. I worked on this for a couple weeks a year ago and was plagued by cycles and deadlocks. Another guy jumped in and tried to solve it. Neither of us got anywhere. Claude Fable 5.1 Max figured out a solution. Legal moves are ones that reduce the average number of random moves needed to bring the game to a conclusion. This is an exact number, just as 3.5 is the exact average of a die roll. Unlike the die roll, calculating legal moves in Unitile is beyond human comprehension. It takes a program. Unitile is a great example of what I would call a pseudo random game. Beyond the current turn, you have no idea what moves will be legal, any more than you'd know future dice rolls. So roughly in the same class as Backgammon, even though there are no actual random elements.

Clear Path (alpha) - Square, orthogonal connection game. Red tries to form an orthogonal path connecting the top and bottom edges, and Blue tries to connect left to right. The mechanism is to guarantee your own connection while simultaneously preventing your opponent's connection. This was even harder to design than Unitile. Five guys took a shot at it over the past several years. I did find a related solution which involved chain reactions - multiple placements per turn. But Claude really solved it with single placements.

Yonkoma (alpha) - Four checkers, two of each color, start off in the corners. Non-capturing bishop moves. The goal is to not form a crosscut:
X O
O X
The mechanism is to either 1) Move a checker closer to the enemy checkers, so that the sum of the king distances decreases, or 2) keep that distance constant and move one of your checkers farther from the other one. Seems pretty easy to design, but it wasn't. Part 2, Claude's contribution, is what prevents deadlocks.

Counterflow (not submitted to BGA) - Non-lethal breakthrough game. The odd-sized square board starts out densely packed with red checkers in the bottom half, blue checkers in the top half, and only the middle row empty. Non-capturing, forward queen moves. You win when all of your groups have a checker in your far row. I think this is my most magnificent mechanism ever (figured out by AI of course). The flow can become so constricted that sometimes one player doesn't have a move and has to skip a turn. And then the other player might only have one possible move. But magically, it never grinds to a halt.

QBF (not submitted to BGA) - Square board, initially empty. One checker is the seed, and more checkers are added, branching out from there. Placements are restricted by quantified boolean formula considerations. For this design, I supplied the seed, and Claude ran with it. And AI really needs a seed. Once it figures out how to generate seeds, I'm out of a job.

Colorfield (alpha) - Related to Dots and Boxes, but with two colors of segments, no chain reactions, and areas larger than a unit square can be claimed with a single placement. I solely designed the 2D version, but it didn't neatly scale up to 3D and 4D. For that I need help from Claude.

I had a major design surge over the past few months, that is only now tapering off. All of these games are hard to understand, hard to explain, hard to identify legal moves, hard to play... But they're all beautiful mechanisms. Welcome to MSG in 2026. The rule sheets are on my website, https://www.marksteeregames.com/index.html.
Post Reply

Return to “Discuss about board games”