CogSci1 Top 10 List Chapter Summary
xex Home

The following Top Ten List of Big Ideas summarizes Chapter 7, "The Second AI Debate", of Stan Franklin's [[Artificial Minds]]. The the words are generally taken straight from the text.

  1. The Second AI Debate The Second AI Debate takes place between AI people advocating computer modeling and connectionist that argue for a brain model of mind. Symbolic AIconstructs its model of mind using computation as a metaphor while connectionism bases its model of mind on a nervous system metaphor. It is very important to remember that both sides sharethe same goal which is to understand and implement intelligence.

  2. The Connectionist Agenda The connectionist agenda is to find a better methods of solving AI problems, to model actual mechanisms of neural computation, and to explore mechanisms of human information processing.

  3. Brain vs. Computer Model of Mind The computer model postulates symbolic internal representation, an internal language of thought, while the brain model does not. The computer model postulates stored programs in the form of production rules while the brain model postulates activity guided by connections in networks turned by their weights.

  4. Lloyd's Cautions Lloyd states that "just because we can describe the behavior of a complex system with cognitive language does not make the system cognitive and certainly does not make the system a mind."

  5. Foror's Attack "Fodor says that thoughts have composite structure. ... Words are composed to form phrases, phrases are composed to form sentences, sentences to form paragraphs, and so on. ... Cognitive processes are sensitive to the structure of the thoughts they process. Symbolic AI systems represent entities compositionally and process them in a structure-sensitive matter. On the other hand, connectionist processes operate via statistical association and are no structure-sensitive. Fodor claims that connectionism can add nohting to cognitive modeling beyond the contributions of symbolic AI.

  6. Chalmer's Defense Chalmers states that connectionism allows you to operate holistically on functionally composed representations without first proceeding through the step of extraction. This proves Fodor's attack to be completely wrong.

  7. Is Connectionism Richer? Symbolic representations have primitive atomic components and compoitional structures while connectionist representions also have compositional structure but instead of having primitive atomic components they have complex, distributed microstructure containing much more information. This is not enough to prove to Franklin that connectionism is better. He states that even thought symbolic representations have primitive atomic components they can be chosen very finely.

  8. Passivization of Sentences "Chalmers claims that the endcoded distributed representation of a sentence still maintains the structure of the sentence, because that structure can be recovered from the encoded version by decoding."

  9. RAAM RAAM is recursive auto-associative memory which is an applied neural network architecture that associates patterns with themselves in a recursive way.

  10. Hundred-Step Rule Human recognition time is order of magnitude 500 milliseonds. This is the time it takes for a idea to come up, not the time is takes to act on it. Feldman's hundred-step rule states that "human reactions are physciologically constrained to require roughly 100 serial steps to