Higher-Level Cognition - Computational Explorations in Cognitive Neuroscience - page 399

(roughly from the 1940s to the '60s) because the ef-
fects of frontal damage did not show up on standard-
ized tests. It turns out that these tests largely tap routine
knowledge and cognitive procedures, and not the kinds
of temporally-extended, novel tasks where the frontal
cortex plays its largest role. Thus, despite the tempta-
tion to associate intelligence with frontal processing and
higher-level cognition, the standard IQ measure appears
to tap more posterior, well-learned cognitive processes.
This fact should serve as an important reminder against
the temptation to attribute too much power and impor-
tance to the frontal cortex — intelligence is truly dis-
tributed and the posterior cortex plays a very important
role in “intelligent” processing.

tion firing, such that the productions end up falling back
on perseverative and noisy firing biases that operate in
the absence of other specific information. Thus, they
build in perseveration as the behavior that the model re-
sorts to after a frontal lesion. In contrast, we see perse-
veration as a result of learning in the weight-based pro-
cessing of the posterior cortex. Nonetheless, this paper
makes a number of more general points that resonate
well with the framework presented here. For exam-
ple, Kimberg and Farah (1993) emphasize the idea that
frontal cortex can be understood as performing a sin-
gle function, that, when damaged, produces a range of
different behavioral manifestations. Furthermore, this
common frontal function has something generally to do
with working memory, which is consistent with our em-
phasis on activation-based processing.
In addition, there are many points of overlap between
the framework outlined above and theoretical frame-
works in the literature that have not yet been specified
at the level of explicit neural network models.
Perhaps the most commonly cited theoretical frame-
work for frontal function is the supervisory attentional
system (SAS) of Shallice (1982; Shallice & Burgess,
1991). The SAS is a central-executive mechanism
(Gathercole, 1994; Shiffrin & Schneider, 1977) that is
deployed for nonroutine behavior, and it has been char-
acterized in the context of a production system architec-
ture, in which the SAS is responsible for maintaining
goal states in working memory to coordinate the firing
of productions involved in complex behaviors. These
notions of frontal involvement in attentional and non-
routine processing are similar to the top-down biasing
we discussed above, and we have discussed the main-
tenance of goal states as a likely role of the specialized
frontal activation-based memory system. Thus, many
aspects of the SAS are consistent with our approach,
but its rather underspecified framing in terms of a pro-
duction system hinder any attempt to provide a more
detailed mapping between the two frameworks.
There are several more biologically based theoretical
frameworks, which lack specific computational mech-
anisms but do provide important general themes. For
example, Fuster (1989) suggests that frontal cortex
plays an important role in mediating sensory-motor
mappings at the highest levels of a hierarchy of many

11.5.2 Other Models and Theoretical Frameworks

There are a number of other models and theoretical
frameworks for understanding frontal function in the
literature. Perhaps the most relevant for the dynamic
categorization model is the framework of Ashby et al.
(1998), who focus on the different brain areas involved
in category learning. They discuss the role of the pre-
frontal cortex in terms of verbalizable, rule-based cat-
egorization processes, whereas the posterior cortex is
involved in more holistic, similarity-based processes.
This is very compatible with the ideas that we have
presented here. One contribution our model makes is
in providing a self-sufficient, learning-based searching
mechanism — the Ashby et al. (1998) leaves these
mechanisms biologically underspecified.
Dehaene and Changeux (1991) developed a model
of the Wisconsin card-sorting task (WCST), which is
closely related to the dynamic categorization task we
explored previously. They proposed that the frontal
deficit impaired the ability to use feedback to select al-
ternative responses, which is also similar to the idea
proposed by Levine and Prueitt (1989). Both of
these models share basic principles in common with the
model we developed.
Using a model based on a production-system frame-
work, Kimberg and Farah (1993) accounted for a
range of frontal deficits, including perseveration on the
WCST. The essence of the model is that frontal damage
reduces the influence of specific information on produc-