Why AI doesn't work without humans

It was coming for our jobs! And to some extent it has. But it needs us. More than we need it. Here's why.

The one-word summary: Inbreeding.

Inbreeding does not actually create genetic mutations, but it drastically increases the chances that harmful ones will be be expressed. Everyone carries hidden, recessive gene mutations that are usually masked by a healthy dominant gene from an unrelated parent; however, because close relatives share much of the same DNA, offspring of inbred pairs are far more likely to inherit two copies of the exact same damaged gene, causing hidden genetic disorders and health defects to manifest.

This is what will happen with AI models as they eventually feed off less and less human input and more and more information generated by other AIs.

To use another analogy, it's like making a photocopy of a photocopy of a photocopy of a photocopy. All the flaws, so tiny at first, become magnified, along with new flaws that are created with each pass through the copy machine. Soon the noise overwhelms the signal.

The vacuum effect

The remarkable success of AI so early out of the gate can be attributed—as with most tech hype—to it being tested on simplistic models. Those models lived in a vacuum. Real life is messier, in ways we still haven't leaned to model. People can create genuinely new ideas and adapt when reality refuses to cooperate. But so far AI can only recombine information that already exists. Ironically, AI's early success came from operating in a world clean enough for it to master. Real life isn't that world. AI might do well to become acquainted with Chaos Theory, where a hurricane can arise from a butterfly's flight. But even then it would model Chaos Theory too simplistically, because AI is inside it, and you cannot fully model something with complete accuracy when you are confined within it. Philosophers and mathematicians call this Gödel's Incompleteness Theorem, and it's a powerful, probably insurmountable limitation. But I'm fairly confident "geniuses" like Elon Musk and Sam Altman have never heard of it.

Maybe someday this will be proven, as Einstein proved Newton's theories, to be incomplete, and a way "around" Gödel's IT will be found. I'll even go so far as to say AI itself will find it. But until then—and even in many cases after then—this limitation exists.

"Aha!"

AI can weave tapestries of astonishing intricacy, and it can do it far faster than any human. But the cloth—and the fibers that make up that cloth—are made by humans. AI can discover combinations and patterns that would take people much, much longer to recognize. But it cannot have an "Aha!" moment when its entire pool of knowledge is confined to the internet's universe of experience. It can't think outside the box like an Einstein, a Beethoven, or a Leonardo. Its known world is the box, and it can't even really comprehend that there is an "outside."

How do humans do this? That's the ephemeral thing we call creativity, genius, or inspiration—something we still cannot really define. It's like Justice Potter Stewart's definition of pornography, of all things: "I know it when I see it."

Recognizing brilliance...It's not always easy

Mistaking complexity for genius is a pretty common thing.

About 15 to 20 years ago I started hearing about this astonishing new genius who was supposedly the next Einstein. So I started following the words and thoughts of Elon Musk. I wondered almost immediately why everyone thought he was so brilliant. I was told by commenters—and I really don't think they were all bots—that I was stupid for not seeing it. It's obvious, they said. When I pushed for examples, I heard things along the lines of, "How much are you worth?" or "He's founded all these billion-dollar companies. What have you done?"

Well, in truth, he didn't found those companies. He was arguably the least important prong of PayPal, and from there he pushed his way onto multiple other companies. He was not the intellectual fount of any of them. From what I can see, he simply cracks the profitability whip harder than most others while making wild, completely unsubstantiated claims that keep him in the headlines. Almost all of those claims have turned out to be ridiculous, dead wrong, or both. Yet for a very long time he was able—and in many circles is still able—to convince enough people that he is the genius they already wanted him to be.

The human advantage

And I fear AI output itself will be able to do much the same thing—dazzling people with elaborate, seemingly profound ideas that persuade those who assume genius must be complicated and incomprehensible. We won't really be able to judge AI's eureka moments until long after the dust has settled.

The supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience. —Albert Einstein, Herbert Spencer Lecture, Rhodes House, Oxford, December 1933

I am fairly confident that, because AI is fundamentally a simulation of human intelligence and cognitive processes, it will never be able to produce truly original ideas without mining human thought. When those human inputs begin to disappear, AI will gradually churn into nonsense, only it won't know the difference.

But we will.

AI purveyors already know this, of course. You may have heard that they're trying to stuff every book ever written into AI's brains for it to scrape content and knowledge. As Techradar reports in this infuriating article, AI companies much prefer books published before 2022 to train their dragons, because those works predate the widespread use of AI-generated content.

It's a cleaner source, in other words. So the AI bros trying to rub out the very content that they know is superior quality and replace it with their slop. The difference is that the "slop" is, for them, profitable. (At least it's profitable in theory; no one's actually made any money off a business model hedge funds are investing billions into yet. And we used to be told Wall Street guys were smart.)

If you ever needed evidence the whole thing is headed for a giant implosion one day, that's it. We just have no way of predicting how soon. Or how later.

Of course this is cold comfort for people who are seeing their jobs and their livelihoods disappearing now. And it's not mean to be comfort exactly. But it is meant to point out that the sands shift, and the changes going on in grant writing and philanthropy in general are not absolute. AI advocates will say it is, but again, they cite models of the future that exists in a vacuum. Just google Irving Fisher, a top Yale economist who sported all sort of expert and authoritative bling, to see how predicting the future is a bad idea. Always.