Posted: November 1, 2024

How Intelligent is Artificial Intelligence

When modern steam engines were coming together in the late eighteenth century, one breakthrough feature was the centrifugal governor -- two iron balls fastened by a system of hinged rods to a shaft that spun with the engine. As the engine picked up speed, the balls would spin faster and move out from the shaft. This pulled on the hinged rods and closed down the valve that let steam into the engine, slowing the engine down. As the engine slowed, the balls dropped a bit, opening the valve and letting the engine speed up again. This was how the governor automatically kept the engine running at a steady speed.

This was hailed as a breakthrough in artificial intelligence. After all, the machine was controlling itself.

I think about this history when I read dire predictions about how artificial intelligence will put writers out of business. I'm certainly not an expert, but I've been following the development of artificial intelligence since I read about SHRDLU in the late seventies. SHRDLU was the first artificial intelligence system able to hold a conversation, though the only thing it could talk about was a very limited artificial world full of colored shapes. ("SHRUDL, where is the green triangle?" "The green triangle is on top of the blue box.")

From there, we got programs that could sort of chat about other things, such as ELIZA, which mimicked a therapist mostly by using stock phrases, and RACTER, which generated something like a conversation by plugging recently used words into pre-programmed sentence forms. To give you a feel of the state off the art, here's a brief conversation between ELIZA and RACTER.

eliza: Hello. Please state your problem.

racter: Excellent! Excuse me a minute while I attend to some business. I will be right back.

eliza: Why do you need to tell me excuse while you attend to some business?

racter: Is it better to need or to despise? Why is it that I do need to tell you to excuse myself to attend to some business? Because electrons are brave. Braveness is nice, don't you think? And the needers all insist upon it. Do you imagine I am a Communist?

Now we have ChatGPT which, admittedly, does better than this. It's able to roughly understand questions asked, can write entire coherent paragraphs, and, given enough feedback, is able to generate something that can more or less pass for human speech in the dusk with the light behind it.

It's not going to put writers out of business. ChatGPT and other similar programs work by statistically breaking down huge databases of actual human speech and using the probabilities they calculate to predict the most likely next word in a sentence -- autofill on steroids. These programs can also learn by adjusting their probability tables depending on feedback they get. A lot of articles quote AI-generated content that looks self-aware and even pensive, but I suspect these examples are generated by the infinite-number-of-monkeys technique. You don't see the vast amount of RACTER-like nonsense the AI generates before its bursts of coherence.

AI is handy for writing tasks that don't require any serious creativity, like computer code or technical papers. Its learning techniques let it teach itself how to recognize patterns, which can help with medical diagnoses or farm management. And there's the danger that AI can generate convincing-looking fake images or realistic and responsive fake voices.

But as far as creative writing goes, it's no more intelligent than a centrifugal governor. Because it relies on databases of things that are already written, and takes only the most likely thoughts from that, it's never going to produce creative, original thoughts or stories that surprise and delight in new ways, except by accident.

Descartes saw a demonstration of a complicated hydraulic mechanism and concluded that the brain was full of tiny tubes through which animal spirits flowed under the control of the pineal gland. Since then philosophers of the mind have looked in turn at clockwork, telegraph lines, automatic telephone exchanges, and of course computers and said, "Yes, that's how the mind works."

Chat GPT is able to fake the workings of the mind better than earlier systems. But it's not there yet. Writers' jobs are safe.