The Failure of Symbolic AI
- Until roughly 1990, the vast majority of research in AI
followed the symbolic paradigm.
- We used symbols like *cat* to refer to things.
- We then used systems, like logic, semantic nets, or just
good old code, to reason with the symbols.
- A foundation for this as intelligence was Newell's hypothesis
that humans are symbol processing systems.
- I actually agree with the hypothesis, but think that
we're a lot more than that.
- We humans base our symbols on the environment; we
ground them.
- This relates to Searle's Chinese Room problem.
- So, our symbols are a lot more than just *cat*.
- It has been said that the failure of the Cyc system
to solve the problem spelled the down fall of symbolic AI.
- I think symbolic AI is still important to solve the Turing
test.
- It's still the most popular form of industrial AI.
- Clearly, people do some complicated symbolic reasoning with
math, language and logics among other things.
- It's just that symbolic AI is not complete.
