Phenomenology of Artificial Intelligence (AI)


This year AI celebrates its 50th anniversary. Its first task was to represent reasoning by rule-based systems. John McCarthy (1956) coined the name AI. Ted Shortliffe  created  expert  systems for  inference in medical diagnosis and therapy.  It was the first attempt to  formalize medical reasoning.  Despite intensive effort in the so called computer assisted diagnosis, its relevance to the physician’s reasoning was marginal.  Representing reasoning by a computer program is simply hopeless.   Man is neither a machine as Descartes believed nor can his reasoning   be reduced  to logical elements.

Even in robotics this machine reasoning failed. The performance of   robots equipped with sophisticated expert systems was disappointing.  In the 1980s  Rodney Brooks revolutionized robot reasoning  with subsumption architecture. Robot behavior was stratified into layers of increasing abstract and complex behaviors. These studies paved the way to embodied AI dominating robotics today.  We may thus distinguish between pre-embodiment AI of the founding fathers, and today’s embodied AI.

Intelligence requires a body. Embodiment is the way in which human reasoning arises from the brain's and body's state. It is an attribute of life and operates even in an ameba. Even the environment in which the body operates is part of the embodiment since triggering its state. Embodiment marks the entry of phenomenology into the exact sciences.  Heideggers Dasein (explained in the previous section) is the hallmark of embodiment.

In my studies I represent embodiment by the Wisdom of the Body (WOB) metaphor.
http://www.what-is-cancer.com/papers/ca/ca93.htm

CA are essentially embodied  since the entire state determines its subsequent state. However Dasein is more than that. It includes also the environment which CA generally ignore.  Embodiment means also interaction, which depends on  the number of neighbors (r)  involved in the computation of the subsequent state.  The greater ‘r’  the more environmental cells are involved in the computation of the subsequent state. Most  CA in Wolfram’s  book are more or less isolated (lonely) like the class-4 structures (p. 282).

In the previous section I explained why Wolfram classified CA only into four classes. We may now add the fifth, consisting of embodied CA which completely interact with their neighbors. Yet they have an “unpleasant” quality. This class-5 cannot be reduced to its elements and may only be treated phenomenologically.

v. Embodied CA:

Back to complexity index