AI - part 1 - Thought and language

/ by
Reading Time: 3 minutes

In his latest blog post, Richard Develyn, CloudTrade CTO, kicks off his grand opus with this first chapter of five connected explorations into the world of artificial intelligence. Here he examines the nature of thinking and the complexity of understanding human language.

“I think therefore I am.”

We have all heard of this famous statement. It was made by a philosopher called Descartes back in the middle of the 17th century. It doesn’t actually say very much about intelligence, though it does at least try to draw a first line in the sand.

The problem with it, however, is that it seems to suggest that thinking can come before existence. But without existence, what can you possibly think of?

Imagine if a thinking thing was to appear in the void: what thoughts might cross through its “head”?

“Is this a nice void,” perhaps?

Or if two of them were to make contact somehow, what sort of intelligent conversation might they have?

“Is your void voidier than my void?”

(I’m looking forward to seeing how our French translators cope with this one, by the way).

I believe it was Sartre, in the middle of the 20th century, who kindly pointed out what a load of nonsense all of this was. Thinking, and language (which is, after all, merely the communication of thinking), is very much tied into the context of our existence. In fact, it is probably more accurate to state: “I am therefore I think” rather than the other way around, as we cannot consider ourselves to be thinking beings if we have nothing to think about, and having things to think about implies that we exist in a world full of interesting, and thought-provoking things.

We do exist in such a world, of course, and the ideas that we hold in our heads, and that we exchange with each other through language, are intrinsically dependent on having a shared view of what that world is like. If I make the statement: “It is hot”, then not only will you immediately know that I’m talking about the weather, you will also gauge how hot I mean by relating it both to where I live and, if you happen to know me, to how much the heat bothers me since I probably wouldn’t have made that statement unless I was finding the temperature uncomfortable. That’s a ton of implied context to accompany those three little words.

(And if I had I said, “he or she is hot”, you would have needed a whole ton more contextual information in order to make sense out of that one - again, I will be looking at the French translation of this sentence with some interest!)

In fact, our brains are so sure of themselves when it comes to interpreting communication that they frequently don’t bother listening to anyone beyond the first word or two and the occasional word after that. If you were to say to someone “It’s not hot” on a hot day, they may well mis-hear you and believe that you told them that it was “hot hot” rather than “not hot”, and reply “yes, very hot”, almost as if to correct your grammar. Actually, people don’t so much mis-hear you as not hear you at all; in this case, once the brain had picked up the word “It’s”, it would predict what was coming next and stop paying attention.

Memory works in a similar way. We may feel that we replay images of the past when we’re feeling nostalgic. In fact, 99% of what we’re doing is reconstruction rather than recollection, and reconstruction depends on who we are now, rather than who we were then. Again, it’s the brain taking short cuts, which in this case leads to a phenomenon called Confirmation Bias, i.e. a tendency for us to remember things in “particular ways”.

There is nothing in the world of artificial intelligence that can hold within it a world view that is anything remotely human. Neural networks that do language translation have a world view of sorts, but it’s one based on an analysis of word association gleaned by trawling the internet and other sources of literature, not on an experience of the world as they, and we, live it.

Language translators can certainly be very impressive if you’re careful what you ask of them, and when they get things wrong, they can be very amusing; but you wouldn’t want to use them to give instructions to a surgeon in a foreign country that is about to remove one of your kidneys. Artificial intelligence as a technology used to understand human communication is still very much in the realms of “impressive toy”. It’s fun to play with, but not something that you would want to use as the cornerstone for either your life or your business.