Notes from Today’s Xindao Salon, 09-07-04

Today, NKM’s grand visit saved me from turning into a solitary castaway. I showed him his logic paper, “Counterfactual Conditionals in Situational Semantics,” and we tossed around a few things about counterfactual conditionals and possible-world logic. I am completely a layman when it comes to logic, so my opinions probably won’t be of much help to him either (exchanging ideas with laypeople is beneficial for philosophical thinking, but I do not know how it works in a strange specialty like logic). 

Personally, I do not concern myself with technical details such as the construction of related formal semantic systems. What I mainly care about, as far as counterfactual conditionals are concerned, are questions about scientific laws. Many of the most basic scientific laws appear in the form of counterfactual conditionals, so how are their meanings to be understood, and how are they to be applied rigorously? If one says that scientific laws are applied approximately in the real world, while they can hold literally in an ideal world, such a statement is easy to understand in everyday discussion. But strictly speaking, what is actually going on? If one says that, in an imagined world, a counterfactual scientific law can be true as an ordinary factual conditional, then how is the self-consistency of that truth guaranteed, and how is its meaning to be applied? Logicians rarely give examples, and can only operate well in worlds containing two or three elements in the form of p, q, p&q and the like. But the question is: let alone the infinitely complex real world, even in highly abstract sciences such as theoretical mechanics, I am somewhat skeptical that these logical models can really be applied. 

Also, I always feel that logic is of no benefit to research on artificial intelligence, even though computers themselves are a byproduct of modern mathematical logic. But I think the direction artificial intelligence should take is to simulate intelligence (in the behaviorist sense), not to reconstruct intelligence. So if the computer’s model is still classical mathematical logic—in other words, a Turing machine—then using computers to realize artificial intelligence should amount only to behavior simulation by relying on database technology, and one should not expect them to simultaneously imitate the internal workings of human intelligence.

Latest Comments

• NKM

  2009-07-04 23:50:01 Anonymous 124.205.76.91

  I also think that artificial intelligence should simulate intelligence, so the view I hold of everyday counterfactual conditionals may, to some extent, seem to lack much originality, because I think it precisely captures the way human intuition makes judgments.
  But one cannot say that logic is of no benefit to research on artificial intelligence either; research in natural language carried out within logic has greatly promoted the development of artificial intelligence.
  Also, speaking of the essence of scientific laws as counterfactual conditionals, I think it should be like this: take Galileo’s ideal experiments as an example. To a very large extent, he was not, as in everyday counterfactual conditionals, subtracting facts—finding the nearest scenario—adding the antecedent—and then deducing results from the conditions already present in the scenario; rather, he was to a great extent making conjectures on the basis of his intuition, because what he overturned was Aristotelian physics, which had long been treated as the ultimate principle. So the laws he reconstructed could not rely entirely on ready-made grounds, but required processing beyond mere reasoning.

• Gu Qi

  2009-07-05 10:39:29

  Could you give an example of how research in logic has greatly promoted the development of artificial intelligence in the area of natural language?

Translated from the Chinese original with AI assistance. The original text is authoritative.