The author isn’t talking about singularity doomerism or whatever.

Replacing programmers already sort of implies some degree of singularity, I can see AI assistants putting some pressure on some programmers, probably will tighten budgets a lot, but fundamentally replacing what programmers do already puts us real close to self-developing AIs, which yeah, you have to enter this conversation of singularity or whatever.

As someone guilty of not fully reading the article before commenting, let me respond now that I’ve sat down and read through it with a real-world example: Github’s Copilot.

When this was first announced a wave of doomerism washed over the tech world as people debated whether it would put the average software developer out of a job once they’d ironed out the kinks and gotten it into a fully autonomous mode. Arguments over this were stupid as it turned out, because the thing can ultimately only do one thing: pull code from Github that kinda does what you’re trying to do and put it into place. As someone who’s used it, it’s literally more trouble than it’s worth even as a code suggestion plugin because it rarely matches up like you’d want it to and you end up having to rewrite a bunch of the suggested code anyway.

One could argue this is just one of those kinks to be ironed out before a full-blown replace-all-the-workers movement kicks in, but that’s still missing the point. It can fundamentally only produce code given code that already exists in a place where it can read it from. This makes for a good replacement for a coder looking up something on StackOverflow because they forgot JQuery again, but that’s not what coders are paid to do. The fundamental job of a programmer is to translate real world needs (usually “business logic” as it’s called in industry) into code, and these needs are overwhelmingly novel and original in some way or form. This is equally true for a journalist or any other “creative” white collar job. And while there are tools out there that attempt to automate this, they’re overwhelmingly bad (just go try out a no-code platform and see how well it works when you throw real-world use at it, or go read an autogenerated tech comparison article and see how long it takes for you to realize nothing of value is being said).

The issue with automating this sort of creative work is, like all AI suffers from, a problem of aping existing work rather than creating anything new. AI and automation is really good at doing repetitive and noncreative work, like manufacturing the same circuit board a hundred thousand times, because a human can come program it to do that thing really well and then leave it be. Putting automation towards creative work inevitably outputs something that can only be described as polished-but-vapid, like a term paper written by someone who didn’t pay any attention and is just scrambling to meet their word count.

And this issue is something that can really only be addressed meaningfully by artificial general intelligence, because creativity is something you simply can’t code into an algorithm. The closest you can get is some degree of randomization, but that falls back on the vapid results again.