proponents of A-life have argued for the superiority of bottom-up thinking over top-down as being more like the way life works -- based on interactions of populations without a "master plan."
Freeman Dyson points out that these two styles of thinking have historically been characterized as Baconian or Cartesian. The former starts with experimental data and tries to set up a theory to account for them. The latter does not depend very closely on experimental work and consists of having some basic beliefs and trying to incorporate them into one theory. (Thomas Kuhn calls the "Baconian sciences" the cluster of research areas that owed the status as sciences to the characteristic insistence of seventeenth-century natural philosophers upon experimentation and the compilation of natural histories, including histories of the crafts.)
Attempts to create artificial intelligence (AI) have been characterized by two different approaches: top down and bottom up.
for top down AI, the material substrate of the brain is irrelevant, and so are the patterns of neurons firing, etc. Instead, strong AI researchers of the top down bent try to construct "rules of thought" through representation in symbols and and rules to combine symbol strings into new, cognitively meaningful ones. Newell and Simon's GPS (generalized problem solver) epitomized the top-down approach of the 1960's. (During this time the paradigm of the computer was the centralized unit accessible only through the "priesthood" of computer programmers.)
Bottom up research believes that the "general architecture" of the brain does matter. For neurophysiologists and computer connectionists the characteristic features of the brain that are important for AI are: large numbers of simple processors, operating in massive parallelism, that are unprogrammed and adaptable.
More recently, artificial life, or A-life, has been working from a "bottom up" approach. In 1989, Rodney Brooks at MIT proposed that robots sent to other planets be "fast, cheap, and out of = "control.html#25"> control" These ant-like "mobots" would exhibit "swarm" behaviour. (see Kevin Kelly, Out of Control, pp 36 - 40) Even at the level of the individual unit, a distributed reflex-based intelligence seemed to work better than a central control unit. "Genghis", a cockroach-like unit with six legs, twelve motors, and twenty-one sensors walked without any central guidance. Like real ants or cockroaches, each leg thinks for itself with feedback loops linking it to the other legs. Brooks describes this bottom-up structure as "subsumption architecture". This design approach has resulted in a reembodiment of intelligence. (see body )
Oliver Sachs invokes Gerald Edelman's "bottom-up" theories of consciousness which looks to the biological orgins of consciousness in the perceptual processes and their mappings in the organism. Alterations of local mapping are sufficient cuases for alterations of consciousness. It is not necessary to invoke any additional cause, such as a coexisting "top-down" neurosis, like hysteria, or psychosis.
Do primates build and maintain their communities in the same way as corals form oceanic reefs -- that is, blindly, without a concept of the end product?
where does the micropolitics of power that Foucault and D+G analyze fit in to the discussions of top/down or local/global?