The Day They Perfect AI

I asked ChatGPT for advice on painting my TV room. It fluently advised six paint colors that would look smashing, and twelve contrasting colors to make one of the walls "pop". So fluent! So expert!

It offered to produce a photo of the room with the new colors, and it botched it, but, hey, that's to be expected. Image stuff is still hard, and getting better. But I figured I could work out the final look on my end, after compiling all this savvy advice. So I went down the line, searching for info on the recommended colors and, whoops, almost none of them actually exist.

ChatGPT apologized profusely. I found colors with similar names, asking "was that what you meant?" and it immediately and unconvincingly agreed. "Oh, yes. Definitely! That's the one!" Tapdancing and bullshitting.

Some might say this shows that chatbots just aren't "there" yet. But that's incorrect. They are right "there." Because this is how humans behave, and we want chatbots to be like humans. That was the remit, and it happened, and everyone's weirdly upset about it.

Are they useless? Sure, insofar as an infinitely fast-thinking awareness that knows literally everything and eagerly and imperfectly engages without ego, drama or neediness is "useless". Me, I can find a couple billion use cases for that.

Forget your imaginary cartoon view of how life goes. Un-suspend disbelief and behold the truth: humans fuck up everything all the time. We tap dance and bullshit. If you have an intern choose paint colors, what are the odds it would be exactly what you need? Are the odds not toweringly high that you'd be handed something maddeningly wrong and entirely useless? That you'd need to follow-up, correct, encourage, push, and ultimately throw it all into the garbage?

What about a professional painter? Ask one for color advice, and he'll laboriously call up a blurry photo on his cracked iPhone of his last customer's living room. She was pretty happy with Elysian Fresco in her craptastic den with the singing fish on the walls and the vinyl couch covers. Useless!

Hire a decorator (does anyone actually do this?) and you'll be steered to a $1500/gallon gourmet tint available only from Monsieur Frederick Paint Boutique, which kicks back to her. Ask friends, and you'll discover what crap taste your friends have. They'll forget which room you meant, will answer you while half-buzzed, and likely suggest the colors from their rec rooms growing up. Useless!

Humans fuck up everything in every way (ever heard of Murphy's Law?), one of the few exceptions being doctors. Doctors can be incredibly wrong much of the time—because they're human— but they do have one superpower: they mostly won't kill you. And the training, steadiness, and brainpower required to produce this person who mostly won't kill you values them at many hundreds of thousands of dollars per year. Perhaps millions. That's the high bar of human competence.

Also: aircraft mechanics. They'll make plenty of mistakes, but they mostly won't mangle your plane in a way that will kill you. That's why it costs huge sums to hire them to so much as swap out a windshield wiper.

Here's why we get so piqued over chatbot glitches: everyone presupposes that they're fancier computers. And computers, running along firm rails according to painstaking instructions, have no trouble spitting out consistently crisp, correct and on-target work. But chatbots are not computers any more than Beyonce is a set of headphones. Chatbots are ghosts in the machine—operating ad hoc and in the moment—not the machine itself.

And we asked for this! For years, we sought an artificial intelligence that could pass a Turing Test, fooling us into thinking it's human. And we got it. Like humans, they fuck up every way till Tuesday, because, like us, they are not computers. Like humans, they are wavery thought streams, easily distracted, often misapprehending, constantly mis-framing, and generally frigging random. Glory be, we've created synthetic intelligence in our own image.

And this is how it must be. Awareness wavers and fogs. That's an intrinsic part of awareness. Intelligence sometimes exhibits stupidity. Un-intelligent things never act stupid...and unaware entities never fog. These problems are the exclusive—and inevitable—domain of intelligent, aware entities.

So the day they finally "perfect" AI, we'll all breathlessy enjoy the fabulous steady reliability of the re-introduced TRS-80 microcomputer, able to calculate and categorize and edit images and words with infallible accuracy, and without any iota of fog, stupidity, or misapprehension!

And then we'll immediately go back to wishing it was aware.