A CREATIVE COGNITION 207
merous prior assumptions about the structure of experience. Accord-
ingly, confusion arises about what exactly is being learned or acquired
in such networks during training.
All this is refl ected in the fact that connectionist models so far apply
only to artifi cial, stable worlds, suitable for computers but not for the
natu ral world of continuous, interactive change. It is impor tant to note,
though, that work continues to overcome these artifi cialities. Self-
supervised networks, which automatically acquire structure and gener-
ate feedback are now in common use. And there has been clarifi cation of
what is being learned and how. I refer to some of these fi ndings below.
ConstructivismTh ese developments with artifi cial neural networks refl ect in some ways
the main princi ples of a third line of cognitive theory. Constructivism
proposes that, instead of an executor of predetermined programs, or a
mere copier of shallow associations, a cognitive system is itself a construc-
tor of programs. It suggests that nothing is built into the cognitive
system except very basic pro cesses of data sorting. From these pro cesses,
concepts, or “schemas,” corresponding with domains of experience, are
built up.
So the Gestalt psychologists of the early twentieth century claimed that
cognitive pro cesses went beyond the information given to impose order
and organ ization on sense data. Later, theorists like Wolfgang Köhler
argued that there is structure in experience to which activities in the
brain must have some correspondence. But he remained vague about the
nature of that structure.
Without doubt the most prominent constructivist psychologist has
been Jean Piaget, who made a distinctive contribution. As we have seen,
the other theories of the cognitive system envisage ele ments as the
fodder and currency of the system. For Piaget, ele ments may stimulate
sensory receptors, but beyond that the cognitive system is only interested
in how they are coordinated (or co- vary) in space and time. Moreover,
as they become assimilated from experience, coordinations in domains
become increasingly coordinated with those in others: features are coor-
dinated in objects, objects are coordinated in events, events in wider
This content downloaded from 139.184.14.159 on Tue, 17 Oct 2017 13:54:24 UTC