Yes, back then AI researchers thought that propositional logic was how human intelligence worked. Later, they abandoned this idea and instead turned to Baeysian probabilities as the most likely mechanism. Later they abandoned this idea and instead turned to connectionism as the most likely mechanism.

Suffice it to say that we didn't have a clue back then and we dont' have a clue now, either. We don't have any good model of human intelligence. We don't understand it. We can't reproduce it. We dont' know how.

And if you think that modern AI research is closer to that target because we now have good classifiers- well, think again. In the past, we had good inference engines. We had good ways to reason with uncertainty. All those things are useful, and they may even be components of intelligence, but they do not, on their own, constitute intelligence.

Clearly.

Depends on your definition of "symbolic".

If you mean symbolic in the sense that computers use symbols, then no, it won't involve that. Humans are self-reflective (at least some of the time), and they don't think that way (with the possible exception of mathematicians).

But words can also be considered symbols, and when we think, we usually think in terms of words. "I" is a symbol, and it's hard to self-reflect without using it.

I think in this context he’s talking about a very specific meaning of symbolic, vs connectionist the two main classes of models of thought. Symbolic is something where you can make out the structure of what’s going on (a requirement for analysis or “introspection”; hence clearly). The alternative is bottom up modelling eg with neural networks where only the outcome is known and the internal processes can’t really be analysed.

My frank thought on this is the AI researchers that placed their bets on symbolic logic were placing their bets on the idea that symbolic logic is at a higher level than connectionist logic.

If you're not emotionally invested that argument is laughable. It's total basis is that human suck at math.

The original artificial neuron, the Pitts-McCulloch neuron, was a propositional logic circuit, conceived entirely on the basis of a model of biological neurons as logic gates, with a threshold function that controlled their true or false value. It was even called the Threshold Logic Unit. The paper where it was first described is titled "A Logical Calculus of the Ideas Immanent in Nervous Activity". In fact, the perceptron too was a logic circuit, a function with two outputs, "positive" and "negative ... or "true" and "false".

AI researchers always used the tools that all scientists use to model their subject, in this case, intelligence: maths. Logic is maths, just as probabilities, calculus, algebra, etc etc.

I think you’ve completely misunderstood what is going on here.

> If you mean symbolic in the sense that computers use symbols, then no, it won't involve that

Why not? Not consciously, but as part of the underlying process that produces consciousness. Particles and fields that make upboir brains carry no more intrinsic meaning than the charged bits in digitial computers.