[ACT-R-users] Playing Tic-Tac-Toe

[db30 at andrew.cmu.edu]

--On Monday, September 17, 2012 3:59 PM -0400 Philip Ritchey 
<pritchey at cs.purdue.edu> wrote:

> Hi all,
>
> Do you think it is possible to get ACT-R to play TicTacToe?
>

That depends on what you mean by "get ACT-R to play".  Assuming
the question is simply can one write an ACT-R model that plays
TicTacToe, then the answer is certainly yes, but there are a lot
of details that will determine how difficult such a task will be
and how issues of representation, memory, and action will need
to be addressed.

Some things to consider are:

- Does it need to play well?
- Is a simple searching model sufficient?
- Is implementing a specific set of rules for playing "perfectly"
  all that's needed?
- Should the model learn to play the game through experience?
- Should it learn the game starting with just instructions?
- Should it learn to play the game and be able to generalize
  that knowledge to similar games?
- Is there human performance data which is to be modeled?
- Does the model need to be able to do the task in the real
  world using pencil and paper?

I don't know of any TicTacToe models that exist already, but
one can find information on issues like those above in other
tasks (including other game playing models for some of them)
in the publications that are available from the ACT-R website:

<http://act-r.psy.cmu.edu/publications/index.php>

Dan