Meat Robot cont. Wright’s Noughts & Crosses Parable

I thought this post of John C Wright’s was quite good:

While I disagreed with a lot of it, it was thoughtful and structured. Indeed, it makes the contrast with his posts on Islam, politics or sexuality seem worse in comparison – Wright is capable of arguing on the basis of something other than appeals to fear and mangled history.

There are two elements intertwined here.

  1. An appeal to Aristotle’s concept of final causes
  2. A very interesting example of the issue of meaning.

I’ll come back to point 1. later as this is the issue on which I made comments and to which there were interesting replies.

Point 2 was this:

Suppose you had a deck, not of 52 cards, but of each move of the 255,168 possible unique tic-tac-toe games. The cards are kept in a chest of drawers. Each drawer is also marked with a diagram of a possible board. Within each drawer is every legal move that can be made in response.

Two instructions are carved into the top of the chest of drawers. The first says to compare the diagram on the drawer with the board, open the drawer, and play a card within. The second says that, as you play each game, whenever you lose, you throw away the card representing the losing move from the drawer representing the previous move. That way, next time you play the game, the move that lost is not the one among the available options to play from that drawer.

Now, suppose your grandfather avidly played tictactoe and carefully threw out every card that represents a losing move. The Naughts and Crosses chest would now only contain cards leading to victory or stalemate.

Again, suppose you inherit the chest and do not know the rules of the game, and no one ever told you the victory conditions. Nonetheless, merely by following the instructions carved in the lid of how to open the drawer and play a card taken blindly from inside that drawer, you can win or stalemate every game.

Now suppose a philosopher strolls up while you are on the last move of a winning game. There are two crosses in in the upper left and right squares, and your opponent has not placed a naught in the upper middle square to block.

You inspect the board, find the drawer that represents it, and open it. Inside is a single card showing the winning move. You follow the card’s instructions and place a third cross in the upper middle square.

The philosopher asks, “Why did you make that move? What was your aim?”

You explain the mechanisms of the chest carefully to him. You show him that your grandfather threw out all the other cards which would have you place a cross elsewhere on the board. There was only one card in this drawer.

The philosopher says, “No, you have told me the mechanics of how you select which move to play. You have not told me what the purpose, point, or aim of the move itself is.”

You look carefully over the chest of drawers. Nowhere are the rules of ticatactoe written down, nor the victory conditions.

The other player says, “The victory condition is to place three a row, either horizontally, diagonally, or vertically.”

Now, studying the chest of drawers a second time, you do notice that all the possible moves of the cards your grandfather did not throw away do, in fact, fit the pattern of being intentional moves meant to bring about the three in a row for crosses while preventing three in a row for naughts.

The word ‘pattern’ here refers to no material property of the chest or the cards, but to the model you carry in your head that you use to deduce the purpose or aim of the grandfather’s chest. The word ‘pattern’ refers to the form.

The pattern is not in the chest, but in the head of you, the gameplayer.


Firstly, it is worth pointing out that it certainly is possible to create a machine learning noughts-and-crosses machine using very basic technology. I can’t find the reference, but in one of Martin Gardener’s many books of recreational maths, he had an example that worked stochastically with marbles in matchboxes. The mechanics of Wright’s example are not at issue.

Some commenters drew comparisons between Wright’s example and John Searle’s Chinese Room analogy. Wright’s example is both stronger and weaker. It is weaker in the sense that it isn’t directly addressing an artificial intelligence and stronger for the same reason. He is using an example of something that could be built and then pointing out that even though this thing could ‘play’ tic-tac-toe very well there is no meaning to be found anywhere in it.

Wright returns to the first point at the end of the post:

Now, having gone over this argument countless times, I will point out that no materialist deigning to present an argument has ever once given even a single example of describing a why as a how, or describing a final cause in terms of mechanical cause, reducing a quality to a quantity, or defining a quality to a quantity.

I think this has a clear answer and that answer helps us start to untangle what is going on, but I’ll save that for another post.



  1. thephantom182

    “Firstly, it is worth pointing out that it certainly is possible to create a machine learning noughts-and-crosses machine using very basic technology.”

    The Ontario Science Centre used to have one running on a PDP-11. Don’t know if they still do, but I remember it as a kid. (PDP-11 was -new- then. Holy. Shit.) PDP-11s did not understand Tic Tac Toe. They used to have a machine that said “coffee” too. That machine did not understand the word “coffee.” It just said it.

    The above is what the Artificial Intelligence community is talking about when they say “artificial intelligence.” Which is hilariously wrong, obviously. there’s nothing even slightly intelligent about it. It’s a -program-. It doesn’t know anything, it merely remembers values. It doesn’t comprehend its environment, it just totals up sensory inputs and applies a pre-written program to them.

    Machine learning is a fancier pre-written program, with parts that get modified with each iteration. But in principle, still a program. It can discover patterns, but not meaning. The programmer, that is who finds the meaning. Meaning requires cognizance.

    Liked by 1 person

  2. David Langford

    The Martin Gardner essay, “A Matchbox Game-Learning Machine”, is collected in his The Unexpected Hanging (1969; UK title Further Mathematical Diversions). To reduce the game space to something that could be handled with 24 matchboxes, Gardner switched from noughts-and-crosses to hexapawn, a deadly conflict of six chess pawns on a 3×3 board.

    Fred Saberhagen used such an improvised learning machine as an sf device in his 1963 Berserker story “Fortress Ship”, later retitled “Without a Thought”. From the timing I suspect he was inspired by Gardner: the Scientific American columns in that collection appeared 1961-1963.

    Liked by 1 person