So they assert that “the apple token is not meaningful by itself”, and then go on to say, “The meaning of the apple token emerges from its network of connections to other tokens.” This is not true reductionism. It is wrapping up your confusion in a gift-box.
Your criticism seems off the mark to me. These AI researchers were trying to automate reasoning. They turned to formal logic, which makes perfect sense—formal logic is just highly disciplined reasoning, so disciplined that a computer can check it. If you’re using formal logic, of course your symbols don’t mean anything of themselves; all the meaning is in the axioms, a.k.a. the symbols’ network of connections to each other.
A better criticism would be that they stuck to deductive logic—which only tells you when you can know something with absolute certainty—and took decades to realize that most of the useful real-world knowledge was probabilistic. One can’t formalize the concept of an apple because it is not well-defined, anymore than a blegg is.
Your criticism seems off the mark to me. These AI researchers were trying to automate reasoning. They turned to formal logic, which makes perfect sense—formal logic is just highly disciplined reasoning, so disciplined that a computer can check it. If you’re using formal logic, of course your symbols don’t mean anything of themselves; all the meaning is in the axioms, a.k.a. the symbols’ network of connections to each other.
A better criticism would be that they stuck to deductive logic—which only tells you when you can know something with absolute certainty—and took decades to realize that most of the useful real-world knowledge was probabilistic. One can’t formalize the concept of an apple because it is not well-defined, anymore than a blegg is.