There's a lot of stuff in Bruce Schneier's article “AI Doesn't Have to Go Rogue to Break the World” that has me thinking.
Back in October I helped run an internal AI day event for my company, and a colleague and I decided to pose a question about how to make a peanut butter and jelly sandwich. The lesson was this: you have the makings of a peanut butter and jelly sandwich, and then you ask people to supply prompts—instructions on how to make one. It's fun, it's silly, and it is a perfect example of how the article's premise is enacted.
Welcome, dear reader, we're so excited to have you here. Please consider subscribing to our newsletter to get Unenlightened Generalist content in your inbox regularly!
Well, that's Bruce's initial point in a nutshell. AI frequently takes instructions and makes of them interactions that are totally unexpected—but for most of us, the types of unexpected directions that it chooses to take will be completely invisible. When I watch it write a piece of code, half the time I don't know what it's doing unless I sit there and figure it out or work with another AI to pick it apart. The same would be true of some of these complex social loopholes.
Hacking the Bureaucracy
I've had great luck using AI to support my research and to aid my elderly disabled mother, who has to go through the same rigmarole of bureaucratic nonsense just to get the most basic and meager financial and health care support.
Without AI, this would have been functionally impossible for me to do alone. With AI, I'm able to push through complex legal documents and utterly horrific, looped systems of disdain for the elderly and the poor, and I actually managed to make her life a little better. This is AI doing the kind of hacking that I think is perfectly wonderfully fine.
The Same Hack, With More Resources
Now, the point that the article's making is that a lot of the big companies—maybe all of them—as well as other evil actors will be doing the exact same kind of hacking, but with far more resources at their disposal; they can hack our social systems to further their own scummy goals. But of course, to that I say, this just makes it more important than ever that we focus on building support for systems that we ourselves control, for balance of power on multiple levels. This would obviously be social and political and legal, but it also comes down to access to this technology.
If we have a system that is broken, then it is our duty to exploit those breaks in order to survive. That's the goal of life. That's what we must do. An AI can make that possible.
References
In response to the original article.