Page 210 - Deep Learning

The Growth of Competence 193

140
2
y = 134 *X^(–0.725) R =1
120
Time for Task Completion (secs) 100 Phase 2 Phase 3

80
60

40
20

0 Phase 1
0 5 10 15 20
Trial
Figure 6.3. The three phases of skill acquisition proposed by paul M. Fitts superim-
posed on a learning curve.

by first identifying the relevant types of information that might be available to
a learner in the course of practice. The second step is to specify, for each infor-
mation type, how the mind might use it to improve an existing rule set. The
possible inputs constrain the possible mechanisms, because no mechanism
can learn unless there is information to learn from.
This view suggests that theories and models of skill acquisition can be
organized by slicing each model into its learning mechanisms and sorting the
mechanisms by type of input. Furthermore, it is clarifying to group the various
types of information and the associated learning mechanisms by the stage dur-
ing practice in which they are most important. Building on the formulation of
Fitts, i distinguish three main phases, which i call getting started, mastery and
optimization; see Figure 6.3. As Fitts emphasized, the stages are idealizations.
50
“it is misleading to assume distinct stages in skill learning. instead, we should
think of gradual shifts in … the nature of the processes … employed, as learn-
ing progresses.” That is, there are no sharp boundaries between the stages
51
and no big switch in the head that determines which stage a learner is in. The
appearance of stages is created by gradual shifts in the relative abundance of
different types of information, with consequent changes in the relative fre-
quencies with which the different learning mechanisms are triggered. The mix
of mechanisms that is responsible for most of the change in any one practice
period gradually changes as competence grows. The division into three distinct
stages superimposes artificially sharp boundaries on this gradual shift but the