Kaldy, Z. (1997). On the proper treatment of Microcognition. Psychology (Pszichológia), 17, 181-193. (In Hungarian)

Eight years have passed since the publication of Paul Smolensky’s seminal paper On the proper treatment of connectionism (Smolensky, 1988). In these eight years, connectionism grew from the passion of a small, partisan group into a major alternative movement to cognitivism, the mainstream cognitive science approach. Andy Clark in his book Microcognition describes this movement and contrasts it with the classical cognitivist, or symbol-processing view. The main difference is that while in the classic view symbols are the smallest units of processing, in connectionism, symbols are activity patterns of several smaller units called microfeatures. Clark shows several advantages of seeing concepts this way. Concepts become flexible (informational holism), in case of partial or incongruent input the system does not break down (graceful degradation), and partial input can activate a whole concept (via default values). However, the same characteristics can be disadvantageous, when we model rule-based cognitive behaviors such as playing chess or proving theorems.

Clark describes some important evolutionary constraints that a cognitive agent has to possess: online, real-time performance, robustness with partial input, and active interaction with the environment. The emphasis on real world compatibility is unique in present day cognitive science: robots that can act upon their environment are not in the forefront of research. Clark presents a persuasive argument why this should change.

Early connectionist researchers claimed that connectionist systems resemble neural networks in the brain; therefore, they are more biologically plausible than symbol-processing models. Clark, following Smolensky, points out that these similarities are only superficial, and the main appeal of connectionist models is not this, but the fact that they more closely simulate an intermediate, subsymbolic level of processing. For the purposes of a psychological explanatory model, one needs to find the appropriate description level. According to Clark, our best metaphor for understanding the mind, at the moment, is a connectionist machine that, for certain purposes, can act as a symbol-processing virtual machine, the same way a computer can sometimes act as an internet portal or sometimes as a word processor.