'Sesam sesam'

eller: Om kognition, sprog, bevidsthed og deres formodede lykkelige forening i det 21. århundrede*)

Abstract

Are there limits to AI? Is 'new' AI better than 'old'? And why? What is the difference, anyway?
This paper examines some of the claims made by the proponents of the 'connectionist' or (POP) model. In particular, it criticizes the implicit assumption present in much of Cognitive Science today that humans are properly modeled as 'information processors'. Two essential characteristics of humans which make computer modeling difficult are: intentionality and sociality. The individual human rests his/her identity on the former notion; the social human on the latter. Another problem is posed by the indiscriminate use of the notion 'functional equivalence' in connectionist modeling. Here, the output is not the only valid criterion: the way one obtains that output is important, too.
Also, much of connectionist thinking, while explicit at lower levels (the 'subsymbolic'), fails to take the higher levels of mental representation seriously. Finally, a possibility of dividing up the territory of AI between classical 'GOFAI' and new
connectionist paradigms is suggested. I conclude with ten 'theses' for summation and further discussion.