We live in a time of peculiar and shaking newness, when the world around us sometimes seems to be changing faster than we have the bandwidth to process. We know, we all do, that so much of our economy is predicated on planned obsolescence, the idea that a given product is only going to be useful for a short amount of time. If you build a widget that does its widget-job perfectly, and that is so well built that it’ll be doing that job perfectly fifty years from now, you’re going to go out of business. So what we get are glass phones and cheap plastic, things that break or must be replaced.
"I'm sorry," says the tech company that sold you your router. "We're no longer supporting your five year old and still perfectly functional Machine X with software updates. That will, of course, possibly expose you to criminals who will get access to all of your passwords and drain all of your accounts. Here's a coupon for 10% off our brand new Machine Y, as a courtesy to you for being such a loyal customer."
But things are getting weird.
As AI models get more and more competent, and the big AI companies increasingly monetize their services, we're encountering a strange phenomenon. AI does certain things, like coding and the testing of code, far faster than human beings. Orders of magnitude faster, in fact. Large Language Models and GPTs are now being used to improve themselves, a process we're calling Recursive Self Improvement. The AI helps build a better AI to replace it, which then builds a better AI to replace it, and so on and so forth. The RSI process is, obviously and recognizably, the camel's nose of what tech futurists once called "singularity."
Singularity is similar to the way a supermassive star collapses into a black hole, becoming a point of unmeasurable density, in which our physics breaks down. It's a state of matter that we can no longer meaningfully interact with. There'll come a point where artificial intelligence isn't just faster and smarter, but is so much faster and smarter that we can no longer understand how it works. We're not wholly there yet, but our feet are crossing the threshold.
In this odd time, one sign is that acceleration is overwhelming our cultural yearning for the new and the shiny. Planned obsolescence is falling apart. If a product cycle is annual, with minor upgrades, users can keep pace with it.
But if a product cycle is monthly? Then businesses can't adapt. You've only just heard of a thing, and are working to understand how to use it effectively for your work or your life, and then...boom...here's another version that's better, and with a superior interface. And then another three weeks later. And then another five days after that.
Why engage or commit to an approach, if one week later your competitors might take advantage of something better? Why restructure your entire workforce to utilize a new functionality, only to have that functionality change three times before you've even had the second meeting of your restructuring taskforce?
As cries rise to slow down the pace of AI development, most of those cries are justifiably about safety. But I do wonder if some of the motivation comes from realizing that hyperaccelerative recursive self improvement shatters our entire economic system.
Posts
