The current huge leaps in the capabilities of so-called AI (artificial intelligence) systems is astonishing. Essays, pictures, cartoons, and conversations are coming out of these machines that are a giant leap forward, far surpassing the machines of just a few years ago. To read the hype, you’d think humans were about to become irrelevant; why employ writers, painters, cartoonists, and essayists when you can just ask ChatGPT?
The answer is simple: In spite of the excellence of these machines, they’re still not creative, and never will be. Not with this technology. For creativity, you still need a human.
Before I go on, a bit of background.
I’m a long-haul skeptic of artificial-intelligence claims. Way back in 1986, I took two graduate-level classes from Professor Terry Winograd at Stanford, in which we studied linguistics as it relates to computer algorithms to process natural language. His main thesis wasn’t that computers aren’t amazing; clearly they are. Instead, Winograd took the opposite approach: he showed convincingly that the human brain is vastly more complex than we realize.
And because we consistently underestimate the power of our brains, we also consistently overestimate the power of computers.
Winograd came to this conclusion after his initial success with SHRDLU, his clever program (1970) that allowed users to type in English-language commands that were understood by the computer. In his “blocks world,” you could type in ordinary sentences like, “Put the red pyramid on top of the blue block,” or “How many blocks are green?” His program was widely admired and is still considered groundbreaking.
But when Winograd and others tried to extend the technology into more real-world situations, it just didn’t work, to almost everyone’s surprise. And that surprise stemmed from two errors. First, we anthropomorphize: we imbue all sorts of objects with intent and intelligence where there is none. Whether it’s a car that won’t start in the morning, or a pesky fly buzzing around, the evil plot to annoy us is entirely in our own minds. Second, we vastly underestimate the power of our own brains. When we see a computer doing some limited task, it creates the illusion that there is some parity between the computer and human brains. But in fact, the gap between the human brain and a computer is enormous.
Winograd saw this, and devoted much of his subsequent research and teaching to this topic, delving into the incredible complexity of human language and our ability to understand one another in spite of this.
How enormous is this gap? Well, it took five more decades to achieve Winograd’s basic goal: parse human language over a wide range of subjects and answer correctly. ChatGPT and its ilk can now answer question on almost any common subject, draw pictures and cartoons, and write long and complex essays.
But really, all we’ve done is move Winograd’s skepticism to the next level. AI is now processing questions correctly, but it is still incapable of creativity or true understanding. The gap between humans and machines is still enormous. Parsing language correctly is a big step forward, but humans are still way ahead.
This gets to the first point above: because we anthropomorphize so much, we’re imagining that ChatGPT is being creative. It’s not. ChatGPT is drawing from a vast repertoire of facts, synthesizing an answer using the writing styles of millions of humans (arranging its essays to be like those documents), and finally (again drawing on billions of documents) creating grammatically correct output.
But is it being creative? Inquisitive? Can ChatGPT create anything that is truly outside of the body of knowledge already loaded into it? Can ChatGPT create a new branch of painting like the pointilists of the late nineteenth century? Can it create a new voice like Hemmingway did? Could it have figured out Einstein’s Theory of Relativity, or written the music for Fiddler on the Roof? Not a chance. Sure, it could write songs in the style of Bock and Harnick. But something new?
Would ChatGPT even try such things? Is it inquisitive? Does it desire companionship and dialog? Does it get bored?
The answer to all these questions is a resounding “No.”
These new AI programs are amazing. They’re a true breakthrough, and will have vast consequences.
But don’t mistake this for true intelligence. Don’t make the mistake of anthropomorphizing. Today’s AI is not creative. It’s just using a mash-up of terabytes of human knowledge. These programs plagiarize the creativity of the rest of us, mixing, matching, and rearranging the current state of human knowledge. And like all plagiarism, nothing new results.
Maybe someday, but as of today, we humans are still important.
Recent Comments