INTELLIGENT OR JUST REALLY GOOD AT PREDICTING?

What are we to think of “artificial intelligence”? Is it really intelligent?

When they compose text, the experts tell us, large language models work by predicting the word that is most likely to come next. Dr. Boli has heard many technical people explain that these software entities are not really intelligent: they are just very good at predicting the word that comes next.

But in order to deny the intelligence of the bots, Dr. Boli thinks, it would be necessary to prove that, when Dr. Boli is making his most entertaining conversation, he is not just very skillfully predicting which word ought to come next.

The assumption of those who would deny the bots intelligence seems to be that, in real human intelligence, there is a divine spark that controls the flow of scintillating repartee in a supernatural way. But no one has ever demonstrated the existence of that divine spark in a way that would satisfy science.

What if, instead, the human mind really is just a huge accumulation of neurons doing exactly what a large language model does: predicting the word that will come next?

Would that make it less divine?

What if, in fact, what it means to be God is simply to be infallible in predicting what comes next—the next word, the next event, the effect of any cause, the next in any series of conclusions, backwards or forwards to any point in time?

Would that make God less divine?

And in creating artificial intelligence, are we trying to build God?

If we are, Dr. Boli believes that God has nothing to worry about. We have imbued all our AI bots with our own idiocies and imperfections. But if God has a different opinion, we may yet see a day when God suddenly decides, “Go to, let us go down, and there confuse their large language models.”

We’ll see.

Comments

  1. Belfry Bat says:

    Personally, I’m of the opinion that the llms “habent nares et non odorabunt, neque est elloquium in guttore eorum”, appearances to the contrary notwithstanding. Indeed there’s some precedent for that even, with Nebuchadnezzar’s dragon.(banana faux fez, sir…). IF the things intellect, then they live, but where do they live and when? What goes for “thinking” And what goes for “speaking” seem to be at separate times.. yes, yes, we’ve all known apparently humans who seem to do the same thing, har-har, but if that make us laugh when it’s accidental, how sad is it when it’s the nature of the design—which, as an llm is a piece of property is indeed essential, or else Aristotle Aquinas could maybe change its mind . And Androidic couldn’t have that!

    I think it’s more fair to say that what intelligent people are doing when they speak is trying to find the words that go in between. They (not to presume myself among them) speak to a purpose, a conclusion, but must start somewhere familiar enough, and what goes in the middle has to fit both ends. Stated in not much more generality, we find ourselves describing a problem that is at least NP-complete, not at all suitable for the engine of a chatbot.

    And God? As the joke says “get your own dust!” You can figure out what goes in front, if it amuses…

    From what I’ve heard of the LLMs, your summary seems near enough their intended mode of operation, but I think we should make the description a smidge longer: they are predicting which word, coming next, is most likely to please the average reader. Or the average Musk. “Average” May be weighted by investment, perhaps?

    About Divine Sparks, I don’t recommend it, but if you fear yours is too close to catching on, there are technicians in Canadian hospitals ready to stamp it out. Again, I told them it was a bad idea…

Leave a Reply

Your email address will not be published. Required fields are marked *