Could an artificial intelligence understand texts?


It is said that an artificial intelligence (AI) would not interpret a context correctly. The programing of AI to understand languages is, according to experts in this field, much more difficult than to teach human being that. Besides, I often make a mistake in reading English sentences so as to understand it in Japanese language.

One is asked to read (1) and then to answer (2).

(1) Shakespeare wrote Hamlet
(2) Who was the author of Hamlet?

We expect to answer (2) readily if understanding (1). In contrast, AI could not answer (2) even if programed with (1). To see why AI would fail to answer (2), it may be necessary to explain how we reach the answer.

Let me assume that we have an algorithm to match "was the author of" to "wrote". But the word, "wrote" could be used for a sentence, "John wrote Mary," while the phrase, "was the author of" would be used for a sentence, for example,  "Babe Ruth was the author of the book on how to hit home runs." For this conjecture, the algorithm is not suitable for AI to answer (2).

This may lead us to a process of the understanding of a context by human minds. When one has learned rules of a language, he will have made sense out of a sentence by using the rules effectively and memorized the derived meaning. He stores the meaning in a different form from an original text he reads. Because an event still remains itself, regardless of the exogenous frames (languages, sentence structures and writing style), the stored memories are irrelevant to its representing forms. The representations used by us will therefore involve just the very meanings. In words, only the representations of meanings do we conceive when understanding a context.  (In my view) Perhaps this is why we are able to understand a wide variety of forms of description and why AI cannot answer the simple question (2).

As we grow, we accumulate experiences and knowledge on the world surrounded us, get sets of meanings, lay down rules and concentrate on using that. We all have common knowledge on which the inference of answering (2) could be made, and a script of which has played a crucial role of preventing us from making contradictory conjectures.

So?

I should study hard....

この記事が気に入ったらサポートをしてみませんか?