Turing complete game 2 AIs can play to measure intelligence

Not open for further replies.
Jan 2013
Anything which can predict the actions of a turing machine on a large scale must be intelligent. Humans do this when they predict what will happen on the Internet as things network to eachother, which is in part a turing machine and part the external world.

The game for AIs is Conways Game Of Life with 2 colors. The rules are the same as 1 color. There is some old computer game I played this 1 square at a time, and its very strategic, but I dont mean choose any square this time.


Only the colors are different. When an empty square becomes filled (because its adjacent to 3 other filled squares of the 8 ), the majority color of those 3 becomes the color of the new cell. Colors spread themself and compete this way.

Imagine a game space for AIs, and people who could do high level strategy, where AIs choose which squares to spend game money to create a cell of their color in a certain cycle of the game. Their investment pays off when their cells spread color, consuming the cells of opposite color.

We would probably need to normalize the money supply and cost of cells spreading so the total number of cells and money in the game stays constant, because you dont want to continue calculating a turing complete complexity of gliders escaping into the empty space in all directions. Or we could define the game as in a wrapped board, a torus viewed as a square, where left/right and up/down wrap around, so there are no paths to leak out.

Its a fact of math that anything can be calculated in Conways Game Of Life, but not necessarily the best system for any certain task. What I'm more interested in, is how we could evolve AIs which use the interface of bit vectors (a line of cells on the game board they control, which create forms jumping out from them), for example bayesian networks, boltzmann machines, and many other kinds of AI support that interface, and evolve them toward general intelligence, or in combination with manually designed components.

For a turing complete intelligence test between 2 AIs, I dont think you can find one much simpler that Humans can watch on screen and understand who is winning, even grab a beer and watch it like a football game as more of your team's color spreads.

I am also interested in the game called Go, which is played on the same kind of board and normally has 2 colors. I predict, based on my theory of physics that black holes are a clique of nodes which all have the same edge weights (because all nodes are held to be adjacent to a total of 1.0 other nodes, causing conservation of many things, equal and opposite force)... based on that and the majority color rule of 2 player Conways Game Of Life, I think the same large scale strategies that work well in Go will also work in this game. They're both games about surrounding the other player at the right time recursively, and gravity is the definition of being surrounded in a balanced way.

You could also play Minority Color, where the 3 adjacent cells creating a new one create it by whats the least color in the 3. If all are the same color, the minority is 0 but we still know what color it is. I imagine minority color as time with the fast changing oscillations and complex numbers, and majority color is stable gravity blobbing together because your color stays your color until acted on by enough of the other. if you think this kind of thing doesnt happen in physics, then why are protons and neutrons made of 1 or 2 up quarks and 2 or 1 down quarks (always 3)? Not that its running Conway's Game Of Life so directly, but it is a fact of math that everything that ever did, will, and could happen is in the set of all possible games of Conway, in the same way all possible games of chess are known just by knowing the rules.
Not open for further replies.