Artificial Intelligence is Bad for Humanity
Our Faustian Bargain
This is the first post in a sequence. You can find the second part here and the third part here.
I’ll start with the simple claim: artificial intelligence1 is bad. It’s bad even though, and in fact especially because, it is frequently extremely good at what it does. If it were merely subpar at the tasks to which it is put, it would be just another failed tool. A cessation in advancement, and a giant bubble pop, would be the best possible outcome at this point — one which, alas, we are extremely unlikely to get.
Unfortunately most takes about artificial intelligence are terrible. If only the worst thing about AI were its water and power usage! And the notion that AI is a useless parlor trick whose main impact is instigating an economic bubble is cope, not criticism. No, the reality is that many people are integrating AI into their lives, and it’s starting to actually have the capabilities that its boosters have been promising, and these are very bad things.
If your ideas about today’s AI are shaped only by using the free chatbots, or even worse by meme screenshots of silly errors, then you have a badly misinformed picture of the world.
I wrote in October that frontier AI models were better at math than almost all people, able to solve well-defined and self-contained problems better than something like 99.99% of humans. They have only gotten better since then. We’ve moved from frontier models solving most new Project Euler problems to solving essentially all of them, and top mathematicians are saying that they sometimes create novel proofs containing “the kind of insight I would have been proud to produce myself”.
In software, major AI labs are claiming that their AIs are now writing nearly all of their code.
Consumer LLMs can write short “flash fiction” stories that are judged better than, and are hard to distinguish from, works by published authors.2 And AI-produced images have come a very long way from the seven-fingered monstrosities of three years ago, to the point where the chief complaint about “AI art” from human practitioners is that it’s stolen their style; many people consider it an acceptable or even superior substitute to human-created art.
In other words, today’s artificial intelligence can compete, across several areas, with extremely talented humans working in their area of expertise. While it still cannot replace the work of a top professional in their field, we might do well to remember that in the legend, John Henry did beat the machine one last time. The capabilities of frontier AIs seem to be increasing at an accelerating rate, in ways that have been quite surprising. It’s hard to see assertions of an imminent halt to that improvement as anything but wishful thinking.3
But my primary purpose here is not to convince you that AI tools are extremely capable, or that they will become even more so in the future. It’s to explain why this is, on net, a very bad thing. I’ll go into details in my next couple of posts, but the gist is this:
The use of artificial intelligence, on both the individual and societal level, is something of a Faustian bargain. In exchange for very real benefits — I am not saying that those benefits do not exist — we accept an ultimate outcome that is very bad for us, in ways that are harder to measure than the benefits, but are nevertheless just as real. This is true at the individual level, where (notwithstanding exceptional cases) extensive use of AI will lead to cognitive atrophy, precipitate unwanted changes in thinking, and open one up to novel ways of being deceived. It’s even more true at the level of our whole society, where we will face an unprecedented crisis of meaning even in the best case, and possibly something far worse.
For what shall it profit a man, if he shall gain the whole world, but lose his own soul?
Of course, “Artificial Intelligence” is notoriously hard to define. When I refer to AI, I mean the obvious thing: the cluster of models and tools that have exploded in capability in the last four years: LLMs and their tooling; image, video, and audio/music generators, and the like. Or, of course, any future technology that is capable of at least as much as these are.
And this was last summer, an age ago in the wild world of current AI progress.
I know that all the talk about AI surpassing human experts sounds like something out of a science fiction story. That doesn’t mean that it’s not true. You’d best start believing in science fiction stories. You’re in one.
Haven’t read the whole series yet but I think you do put AI in a too positive light here regarding capabilities.
> And the notion that AI is a useless parlor trick whose main impact is instigating an economic bubble is cope, not criticism.
On the subject of cope i am grimly amused that the criterion of whether AI had human-level intelligence was always the Turing Test. Until AIs clearly passed it, of course, at which point people tacitly agreed that was never the criterion.