Cognition: Memory, Attention, and Learning - Foundations for Designing User-Centered Systems - page 151

5.4.6 Transfer
Transfer of learning is important. After a skill has been learned, the goal is often to
reuse or apply the knowledge to a new situation. If no transfer occurred, then every
situation would be a new situation. If transfer was perfect, then few situations
would be novel. Studies have shown that transfer is usually far from perfect, that it
can be underestimated, and that it is possible to predict transfer effects.
Perhaps the earliest study on transfer had subjects read a story about how a king
wished to invade a town, but his army was too big to come through a single gate in
the town. So he split his troops up into smaller parties. Subjects then read about the
use of lasers to attack cancer, but the problem was that the light was too intense to
directly fire through the tissue at the cancer. What to do?
A surprising number did not think to use multiple lasers, which is what the
transfer of knowledge would suggest. This lack of obvious (to the experimenter)
transfer effect has been repeated many times. Problem solvers have trouble
transferring knowledge or strategies where there are structural but not surface
similarities. Thus, users do not always transfer useful information.
The second study to keep in mind is that of perverse Emacs. Singley and
Anderson ( 1989 ) trained users for several hours with a powerful text editor called
Emacs. Then, on day 2, some subjects were trained on Perverse Emacs, which was
just like Emacs, but the keystroke commands were different. Subjects, we can
imagine, did not like this at all, but at the end of the day, their performance was
indistinguishable from the people who had used Emacs in day 1 and day 2. Thus,
transfer can occur, but users do not always see it, and they do not like it when
transfer leads to small mistakes (as would have often happened in the first hour for
the users of Perverse Emacs). Thus, changing key bindings is likely to lead to
frustration, but might not hurt users in the long run.
Finally, there are some tools to measure potential transfer. Kieras and Polson
( 1985 ) created a simple language to express how to do a task, and many task
analysis techniques can be used in a similar way. Each instruction in the language,
such as how to push a button, takes time to learn. Thus, the number of differences
between instruction sets indicate how much knowledge can be transferred and how
much must be learned. Analyses of popular operating systems suggest that there
are real differences between them, with one being a subset of the other (and one
thus being easier to learn).
5.4.7 Implications for System Design
All users will learn and get faster at doing a task as they repeatedly do it. This is an
important way that users fit themselves to a task, covering up errors in design or
compensating for trade-offs made in interface design that do not favor the user.