Page 117 - Artificial Intelligence in the Age of Neural Networks and Brain Computing
P. 117
4. Wearable Personal Assistants 105
Of course we could build a digital equivalent of a governor from electronic com-
ponents. It may seem that a computational model offers an easy solution to problem
of metacognition and symbolic thought. The computational governor, unlike the
centrifugal one, consists of structures that measure (represent) things. To explain
how such a device might think about keeping the flywheel steady or invent a system
of symbols to stand for various steps in the procedure, we need only to explain how
the device looks down on its own cognitive processes and makes them accessible to
consciousness. In this view, metacognition and explicitly symbolic thoughts are built
out of the very same representations that do the job of more ordinary cognition.
This kind of explanation is pervasive in cognitive science today. Phonemes are
postulated to be the representational units of speech perception and speech produc-
tion. Phonemes are also, more or less explicitly, represented by the letters of the
alphabet. Alphabets thus can be viewed as the external representations of internal
symbols and their invention as evidence for phonemes’ internal reality. Teaching
reading, writing poetry, theorizing about phonology, and talking pig Latin can
also be viewed as using the very same representations that underlie the perception
and productions of speech. Thus all cognition is one kind of thing: the explicit
symbolic manipulation of doing logic, drawing maps and counting are overt versions
of what all cognitive process is. As Liberman has shown, representational entities
such as phonemes do not exist in palpable form, like flywheels or letters on a piece
of paper. If we were to look into brain, we would not see symbols.
Thelen proposes that symbolic thought, like walking up and down slopes, like
interpreting novel words, is emergent in activities and in the products of those
activities in the physical world and on us. As we act and explore our world, our
actions will produce concrete physical changes in that world that we may perceive.
By perceiving the products of activity and the reentrant mapping of that perceiving
onto the ongoing activity from which it emerged, we create external symbols
and discover their value. This is another description of Merleau-Ponty’s action/
perception arc.
In Thelen’s view, development does not unfold according to some prespecified
plan, there is no plan. Developmental change is caused by the interacting influences
of heterogeneous components, these are not encapsulated modules; indeed develop-
ment happens because everything affects everything else. But time-locked patterns
of activity across heterogeneous components are not building representations of
the world by connecting temporary contingent ideas. We are not building represen-
tations at all.
4. WEARABLE PERSONAL ASSISTANTS
If we combine the field approach of neural resonance with the motor-based theory of
cognition, we arrive at a powerful new platform for the next generation of wearable
personal assistants. Unlike pure AI which is trying to make computers intelligent,
neurally based personal assistants are focused on a more attainable goal of