Representations without rules
To adequately model cognition, connectionist networks must exhibit represented representational structure without following "obligatory, hard representation-level rules".
This new paradigm for cognitive science must include:
"representations with complex internal structure that can be related in various ways that go beyond mere association, and processing that depends on this structure and those relations. But, for large and significant areas of cognition there must be no rules adverting to representational structure or content."
Terence Horgan and John Tienson, 1991.