Connectionist networks can’t elaborate capacities quickly
Humans and AI systems are "elaboration tolerant"; they can quickly extend their abilities to take into account new phenomena—connectionist networks aren't and can't.
Whereas humans (and AI systems) can quickly elaborate their capacities, connectionist networks can’t.
For example, when an English-speaking human is given a simple rule of Chinese pronunciation (e.g. say "ch" when you see "q", he can immediately use the rule to speak differently. This kind of learning couldn't occur in a connectionist network, which would have to instantly adjust thousands of connection weights.
John McCarthy, 1988.