Trainees learning to work a ten-key add listing machine have to translate chunks of numerical data such as ‚1278‘, ‚253467‘ (obtained, in real life, from invoices) into rapidly performed sequences of key depressions. At the outset, each chunk poses a realistic problem, and it happens that people learning this skill describe the problem in terms of more or less consistent descriptive attributes. Some of these are:
(i) Number of items in a chunk of data. (Two attributes.)
(ii) Whether the items entail horizontal runs on the keyboard (given the usual layout ‚123‘, ‚456‘, ‚789‘), or
(iii) Vertical runs like ‚141‘, ’25S‘, and ‚369‘.
(iv) Specific constraints such as ‚all items selected from the subset 2-8‘.
Hence a problem is conceived as something possessing or not possessing these attributes in varying degree and the skill is performed after the manner of an elaborate frog that deals with specified conceptual categories. We use a teaching system of the kind in Figure 21. The controller learns the effect of changing attribute values upon 0 (as before, computed from the student’s performance). Then, hill climbing in the attribute space, it aims to maximize O. The mechanical arrangement involves an attribute filter used in reverse. A given state of the controller specifies, for example, that the data presented at this instant shall have four items selected from numerals 2-8, and a horizontal run. The ‚dice thrower‘, which supplies the requisite variety, selects some problem from the ’specified set. Now, for any state of the controller, Figure 21, is a relabelled version of Figure 6 in Chapter 3. Hence, recalling the discussion, this teaching machine presents the student with a suitably adapted sequence of models of the environment he must eventually deal with.