Page 95 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

1. Dichotomies 83

adaptability; and work by discrete dynamics. Their physical implementation is
irrelevant in principle; they exhibit sequential processing and the information
processing happens mostly at network level. Brains are self-organizing devices;
they are structurally nonprogrammable; they work by both discrete and continuous
dynamics; their functions depend strongly on the physical (i.e., biological) substrate;
the processing is parallel; and processing occurs for both network and intraneuronal
information.
Inspiration from the brain leads away for emphasis on a single universal machine
toward a device composed of different structures, just as the brain may be divided
into cerebellum, hippocampus, motor cortex, and so on. Thus we can expect to
contribute to neural computing as we come to chart better the special power of
each structure. The brain may be considered as metaphor for sixth generation
computing, where the latter is characterized by cooperative computation, perceptual
robotics, and learning [57].
We now know that (mostly part of now the collective wisdom, but see, e.g., also
Ref. [21]): (1) brains are not digital computers; (2) brain does not have a central
processing unit, but rather uses cooperative, distributed computation; (3) memory
organization is based on dynamical (as opposed to static) principles, (4) brain uses
the combination of discrete and continuous time dynamics, and (5) the synaptic
organization of the brain is very unique, and may be the key element of the biological
substrate of human intelligence.

1.3 THE COMPUTATIONAL THEORY OF MIND
Third, the computational theory of mind holds that the computational metaphor is
the final explanation of mental processes. The classical version of the theory
suggests that the mind executes Turing style computation. As is well-known, the
birth of the formal AI was the Dartmouth Conference held in the summer of
1956 (an important year, in many respects) and organized by John McCarthy.
The goal was todiscuss the possibilities to simulate human intelligent activities
(use of language, concept formation, problem solving). The perspectives of the
cyberneticians and AI researchers have not been separated immediately. Some of
McCulloch’s papers also belong to the early AI works, as the article titles reflect:
“Toward Some Circuitry of Ethical Robots or an Observational Science of the
Genesis of Social Evaluation in the Mind-like Behavior of Artifacts” or “Machines
That Think and Want.”
“Connectionism” [22] emerged an ambitious conceptual framework for a general
brain-mind-computer theory movement, but it is based on principles of “brain-style
computation” that ignore many of the “real brain” data. The connectionist move-
ment is thus directed more to the engineers of near-future generation computer
systems and to cognitive psychologists.
There are recent debates about the meaning of the term “mind computes,” and
“embodied cognition” seems to be a radical alternative [23]. The central hypothesis
of embodied cognitive science is that cognition emerges from the interaction of