The AI takeover is coming
I'm an AI maximalist. Over time, I expect AI to become much, much more capable than most people imagine. In time, I expect AI will be able to do anything a human being can do, and much faster. AI will not be able to directly manipulate the physical world, of course, but with some initial support from human beings, it will be able to build robots that can.
That's not to say I'm an accelerationist or optimist. That's another issue entirely, and I don't know exactly how I feel about it. I tend to believe AI will probably do immense harm to humanity in the long run, however, either intentionally or unintentionally. I'm not confident enough in that view to change my behavior in any way, and even if I were, I wouldn't know what to do about it. Superintelligent AI is coming, and there doesn't seem to be any way to stop it.
In any case, I do expect that AI will soon replace almost all human knowledge workers. How soon is “soon”? It's hard to say. We may need a revolution in AI technology, not an evolution, to get there. LLMs may not be enough. Still, if the AI takeover of knowledge work takes 100 years, I would be very surprised. If it takes 5 years or less, I'd be only modestly surprised. I'd guess there's about a 50% chance of it happening before 2050.
My reasoning is straightforward. I don't believe intelligence requires a human brain, computers are much faster than human beings, and the first company to develop a superintelligent AI is going to make a tremendous amount of money. The race is on.
I've heard the argument that AI will only complement human beings, forging partnerships that are greater than the sum of their parts. I consider that wishful thinking. Consider chess.com, which provides an analysis of each game after it's completed, identifying mistakes and suggesting better moves. It's a huge asset to the platform, and it's one reason people pay for premium memberships. Is chess.com tempted to hire human beings to do this work? Of course not. Computers are vastly better at chess, they are incredibly fast, and they are much more cost effective, not needing sleep, health insurance, bathroom breaks, or team-building exercises. Yes, there are some human commentators at highly-publicized matches, but for the millions of other games played on the platform each day? Hell no. The computers do it, and why shouldn't they? It's a “no brainer.”
Get ready for the same thing to happen to software engineering, media production, accounting, and just about everything else that largely involves transforming information.
What happens after that? Only time will tell.