An example model of a simple cognitive task, and of the style of explanation AI offers. Why is it easier to recite the alphabet forwards than backwards? Note Neither Sloman nor I are proposing this as a serious cognitive model, merely as an example of a style of explanation.
A mechanical analogy: we decide, for no good reason, to store the
letters of the alphabet in a filing cabinet. To store A
, we write
it on a card, find an empty drawer (say 8), and put the card inside.
Make a note of this as the starting position. Later, we store a
B
. We pick another drawer at random. Say it's 15. We put the card
in 15, and then go back to drawer 8, writing 15 on A
's card.
Later still, store C
. We find another drawer - 1 say - put
the card for C
there, and then go back to write 1 on B
's
card. And so on, to Z
, which goes into drawer 435.
Now, consider somebody who's asked to start at A
, and recite the
alphabet forward. This is easy. Start at 8, pick the card; find the
drawer it mentions, and repeat. Going backwards is harder...Start at
435, and then what?
I could, using what computer-scientists call ``singlely linked lists'', build a robot which stored sequences in an analogous fashion. Such a robot would find it much quicker to recite forwards than backwards. Alternatively, I could build one where each cell (card) bears two numbers, one pointing at the previous cell, and one at the following cell. Now there'd be no difference between the two directions. So the way we represent information crucially affects its processing.
This example comes from The Computer Revolution in Philosophy by Aaron Sloman (1978). In the Bodleian (one copy as 26684 e 1106 and one as 266 e 603 unless reclassified). Chapter 8, especially sections 8.1, 8.3 - 8.10, and 8.16, is an account of the kind of representation that might build up in a child learning to count. This is Sloman's view of how cognitive psychology ought to answer questions like ``Why is it easier to count forwards than backwards?'', and it is typical of the AI approach to cognitive modelling.