One Selfridge Model: COUNT TESTING MODE OF OPERATION ONLY
This implementation of the COUNT model runs in LCSI’s Microworwld Logo.
It recreates a version programmed by Lawler in Apple Logo in the 1980’s.
NB: The keyboard IO routines (MWEX Logo “Question”and “Answer” primitives) do not yet
work properly, so this example of Selfridge’s idea is not yet available.
Think of the Logo Turtle as a finger, pointing at one of several sticks in a row.
Imagine the Learner knows the number names in sequence, but does not know how they apply
to the challenge of counting a collection of objects. The Learner (a Learning Machine) then has
a number name in mind (the mental number name).
The Learner knows how to do four things. Two apply for the finger, moving Left [L] and Right [R].
Two apply for the Mental Number name, thinking of the Next [N] and Previous [P] number. There is
one more primitive in the system, [A] Again means “do the last thing repeatedly until reaching
a natural boundary.” These are the COUNT facility’s five Primitive Capabilities (PCs). The engine
runs the generate and test strategy. When it succeeds occasionally, Learner asks if the solution is worth
remembering and requests a name if it is so. The suite of operations which jointly achieve
a solution are composed into a new procedure, a Learned Capability.
By proposing a judicious sequence of goals for the Learner, the Teacher is able to lead the machine
to Learn how to count a general set of sticks of any small, finite number.
Note well: this facility requires the use of the LCSI web-player interface (a small, free bit of code
available from them) which recommends that on Macintosh OSX systems it run with the Safari browser.
See Also: NEST, an Extension of Selfridge Models (Forthcoming).