Simulations of connectionist networks aren't duplications
It's invalid to conclude that because a serial simulation of a connectionist network can’t think, that an actual connectionist network can’t think. The invalidity can be shown by modifying one of Searle’s own arguments about simulation.
Three Valid InferencesSearle argues that this is invalid reasoning:
- An actual storm can make is wet,
- Therefore, a simulation of a storm can make is wet.
But that implies that the contrapositive form is also invalid:
- A simulation of a storm can't make us wet.
- Therefore, an actual storm can't make us wet.
Thus, Searle's connectionism argument is also invalid:
- A serial simulation of a connectionist network can’t think.
- Therefore, an actual connectionist network can’t think.
See "
Simulations are not Duplications", Map 2, Box 23.
Jack Copeland, 1993.