The symbol grounding problem

The problem is that symbolists, those from the paradigm that believe that cognition is the manipulation of symbols, can't connect those arbitrary symbols to real life experience. And this problem carries over into the computational realm in the fact that a robot, currently, cannot make its own meanings based on its interaction with the world, the only meanings are our meanings. The symbols that we put into the robot are extrinsic, they depend on something external in order to have any meaning, and in this case that external thing is our mind.

Connectionism tries to get away from this problem by doing away with symbols altogether. They claim that meaning is an emergent property that is brought about by activation patterns. These activation patterns are based on weighting and links put in place by the programmer. Once in place, the system prunes and weights based on feedback. However, symbols are useful in the sense that they can be combined and built up to create new meanings, and this paradigm ignores that.

It seems like Harnad is essentially trying to combine these two methodologies in order to create a more robust computational method with the best from both world. However, as was noted in class, it was a rather theoretical article that has flaws which would prevent this method from succeeding.