I’ve gotten into a number of conversations about this recently and I just wanted to throw my 2 cents into the ring. And the first thing I have to say is that I consider Artificial Intelligence to be a misnomer. It suggests notions of sentience and consciousness that, despite what some random and panicky Google employee might think, is never going to happen. I’m not going to go into why Algorithms ≠ Sentience and never will. That’s a rabbit hole full of quibbles and false equivalencies. It’s like arguing how many angels dance on a pin. Suffice it to say, science can’t even currently define sentience/consciousness, let alone create it “artificially”. AI is just a set of algorithms no different—and I can’t emphasize this enough—no different, practically speaking, than a pocket calculator. If you ask any calculator what 2+2 is, it will tell you. 4! My God! Artificial Intelligence! Sentience! Consciousness!
No.
Correctly answering questions is not a sign of sentience. ChatPGT is just a glorified calculator. That’s all it is. Which is why I would call it algorithmic intelligence rather than artificial intelligence. Ask ChatGPT a question and you’re asking a calculator a question. The only difference between the two comes down to the sophistication of the algorithms used and the answer given. That’s all it is. You can unplug it or take out the batteries when you get bored with it. It’s just algorithms.
Now, onto the subject of algorithmic intelligence vs. art, music and literature. It ain’t gonna happen. Here’s why: Back in 2015, AlphaGo became the first computer Go program to beat a human professional Go player without handicap on a full-sized 19×19 board. AlphaGo accomplished this feat through the use of “Deep Learning“, what developers termed a “Neural Network”. The unfortunate upshot of all this terminology, like “neural”, is that it leads one to think that developers must have created something like a brain. But that’s not what they’ve done. What they’ve done is to write elegant algorithms that mimic perceived cognitive features in biological systems—and in a very limited sense. They’re not mimicking “consciousness”. They’re mimicking, at an algorithmic level, the way biological systems are perceived to organize and analyze information. I write perceived because AI is only mimicking one aspect of a biological system ascertained through observation. Neural networks in no way define or recreate “intelligence” or “sentience”. The reason that AlphaGo could master Go is because, though the algorithms were difficult to perfect, there was a fool-proof evaluation function that defined success. Either AlphaGo won or AlphaGo lost.
Period.
The same doesn’t work for producing art, a symphony, or poetry. Algorithmic intelligence, for example, has no way to evaluate the aesthetic/emotional success or failure of a poem. Given that human beings can’t even agree on what constitutes a great poem (mostly for lack of knowledge, ability or talent) an algorithm has no hope. I’ll occasionally be asked why I obsess over a definition of great poetry and why public appeal matters. It’s because public appeal is humanity’s version of an evaluation function; and it’s most effective when it functions over time. That’s why we can say that Shakespeare, Bach, Keats, Mozart, da Vinci and Beethoven are our greatest artists and why we can say that as an objective measure (despite all the hand waving among those who continue to insist that all art is subjective and a matter of taste). A work of art’s appeal, over time, is an objective measure. It’s the only one we’ve got.
The problem for algorithmic intelligence is that genius is rare.
This means that if Algorithmic Intelligence is tasked with creating a poem, its models—the thousands and thousands of poems it can sample—are going to be almost wholly mediocre. And because algorithmic intelligence has no concept of mediocrity, has no evaluation function pertaining to the artistic accomplishment of a poem, it will, at best, “learn” how to flawlessly mimic humanity’s mediocrity. By way of example, I get sent dozens of poems over the course of a year and, with few exceptions, they are all mediocre. But what is striking is how similar the mistakes of algorithmic intelligence are to the mistakes of mediocre poets. In short, algorithmic intelligence is rapidly “getting better at “learning to mimic” mediocre poets, and that’s because mediocre poets and algorithmic intelligence are both drawing from the same well. (Interestingly, part of what makes mediocre artists mediocre is they lack the ability to accurately evaluate their own output, called the Dunning Kruger Effect.)
You might object that if it takes time for humans to identify and agree on great art, then expect the same from algorithmic intelligence. The problem is that the next one hundred years are going to expose algorithmic intelligence to vastly more mediocre art, music and literature—along with a hundred years worth of confused human evaluation. That’s only going to make algorithmic intelligence even better at mimicking mediocre poets and readers. Ultimately, the mirror that AI will hold up to humanity, in terms of art, is not humanity’s genius but it’s bland mediocrity. And that’s because mediocrity, with rare exception, is what humanity produces.
To summarize, the only evaluative guidance algorithmic intelligence has as regards the “success or failure” of its art is human taste. God help it.
Pity the mediocre poet, composer or artist, because that’s who algorithmic intelligence is going to put out of business.
up in Vermont | March 22 2023