AI overlords made headlines over the past year, talking about how their business model could lead to the end of life on earth. Fun! Exciting!
After the usual hyperventilating, many experts on issues like nuclear safety or biological engineering wondered how this extinction event could occur. Seemed very unlikely. One writer I recall differentiating between existential and catastrophic threats. The idea that AI is an existential threat seems overblown and melodramatic, but the idea that AI poses some sort of catastrophic threat seems distressingly real.
The main issue - as Krugman notes - is from a proliferation of hacking that makes life fundamentally different. Best case scenario is that everything becomes stickier and more labor intensive as AI actually unwinds some of the digitization of the modern world. Over the past 40 years, countless forms of commerce and other means of interacting have moved into the digital world. Much of the code undergirding that world is very old. AI - it seems - is very much adept at exploiting that ancient code.
Best case scenario is therefore a slowing down of economic activity as we cope with the various threats from hacks. Every transaction becomes suspect, and the world of two-factor authentication is therefore not really enough. The process of enshitification is when websites start out easy to use, but then are designed to be harder to use until you pay them for premium access. It's the proliferation of paywalls and other hurdles to access in order to maximize returns.
AI has the potential explode enshitification without actually gaining anything from it in terms of ease of access or additional content. Every website will become a pain in the ass to access.
The worst case scenario is the erasure or hostage taking of critical realms of the digital world. Imagine, for instance, major brokerage firms suddenly having all of their accounts just erased or taken over by hackers. AI makes hacking easier, it's not that AI itself is doing the hacking (necessarily). We worry about someone using AI to bioengineer a supervirus that kills everyone, but far more likely is just a rampant proliferation of hacking.
We are already seeing some aspects of this in electoral interreference in places like Brazil, but so far people have been able to outlast the bots and other issues. How long will that last? That seems more or less the oligarchs placing their thumbs on the scale, but how soon until AI fueled mis and disinformation overwhelm the voting public? How long before AI's risk become so profound we wind up unraveling the digital revolution?