Page 55 - Deep Learning

38 Introduction

Retention

Perception
Declarative
knowledge Thinking
Retrieval
Encoding
Central executive
and working
memory
Skill
acquisition
Practical Planning
knowledge Action

Figure 2.2. An overview of the cognitive architecture as currently understood.

system that explains all behavioral regularities and phenomena. We specify
the mental representations, the repertoire of basic processes and the control
structure once and for all. The only component that varies from explana-
tion to explanation is the content of the representations; that is, exactly what
knowledge we hypothesize that the person or persons brought to bear in the
course of the cognitive performance to be explained. Ten years after Newell’s
proposal, John R. Anderson coined the label “cognitive architecture” for such
a specification and the term stuck. The ACT-R theory proposed by Anderson,
arguably the most influential theory in 20th-century cognitive psychology, is
a sustained attempt to carry out Newell’s theoretical program. Figure 2.2 is
35
a sketch of the cognitive architecture as understood by most cognitive psy-
chologists. The trick in the cognitive architecture game is to specify exactly
the right system. A specification is correct if the resulting system solves cog-
nitive tasks in the same way (or ways) as people, and if it exhibits the same
behavioral regularities and phenomena as people. Newell’s vision was, and
remains, the only clear, complete and coherent concept of what it means to
have a unified theory of human cognition, analogous to the Final Theory of
Everything that some physicists suspect might be within their reach. 36
Representation-process-control explanations differ radically from other
forms of psychological explanation such as behaviorist, dynamicist, Gestalt,
Freudian, Piagetian or psychometric explanations. A comparative study of
strengths and weaknesses of these different explanatory programs would require
too many pages. The main reasons to prefer the Turing-Newell conception over its
competitors are its clarity and the specific guidance it provides for building mod-
els of particular cognitive processes. Due to these strengths, the Turing-Newell