Connectionist AI vs Symbolic AI
Different architectures make different assumptions about the content they will process and about the types of problems they will solve.
Connectionism applied very successfully to classification problems and low-level processing. Strengths include integrity in face of uncertain and incorrect data, and a natural gradation in match. But they require large amounts of training data and the knowledge in the final net is usually opaque.
Symbolic systems have largely complementary strengths/weaknesses.
Human-like cognition probably requires both.