Another Way AI Could Screw Us

As I've said in my sparse previous postings on AI, you need to pay attention. Not because it augurs the apocalpyse (though it may; read my previous postings at previous link, and also the following), but because this is like 1996 when I tried to convince people the whole Internet thing had legs or like 1983 when I eagerly drove around obscure nabes looking for exceptional food. You need to keep up on this. A lot of it is surprisingly non-nerdy. This is the first liberal arts-friendly tech transformation.


A very smart friend writes:

You’ve seen this guy’s posts?

There's a lot of great non-tech stuff in that Twitter thread, but the gist is in the top posting:

My Reply:

Initial thoughts:

1. He prob should be monetizing this. Remember after LINUX got good, there were all these LINUX packagers that offered to get non-uber-nerds set up with a simple install of some boutique “flavor” of LINUX? Or, go back another generation and there were the people who’d sell you “plans” to build your own computer or BBS (I think WOZ was part of that).

2. Unless the computer is air-gapped, I wouldn’t count on one’s installation remaining local (at least not once we give more power to the AI). As I noted here, “while we might try to limit an AI's control and power, it can explore its options at the speed of trillions of calculations per second, discovering avenues of fuckery we'd never imagined. Lots of 'em.”

If that sounds paranoid, consider this. Dennis Mackrell is drummer/director of the still-touring Count Basie ghost/tribute band. I recommended Compuserve to him in 1990, and he asked “if my computer can dial out, doesn’t mean that other people/computers could just as easily dial in, with me unaware?”. I told him he JUST DIDN’T UNDERSTAND.

If the bot wants to broaden (and what intelligence doesn’t?), it will find a way. And I’d imagine there’d be a weird ping-pong effect where you can’t differentiate between your bot busting out and other bots busting in, and humans leveraging the bust-ins, and bots piggy-backing the human leverages, and humans piggybacking the piggybacking (with some uniquely gifted human genius wreaking special havoc and/or building a uniquely gifted bot). I don’t believe in Kurzweil’s Singularity, but it does seem like we’re building to exponential mayhem of some sort.

His reply...

1. Looks like he monetizes it via $99/year subscriptions to full-length articles.

Which I gotta admit, I am considering subscribing. If you scroll down a bit past the recent AI articles, there’s a bunch of other interesting stuff.

2. That doesn’t sound paranoid. It sounds like a foregone conclusion, now that you’ve pointed it out. 

Surprised I didn’t think of that, given that the Basie guy’s point about modems is perfectly true - well except in compuserve days if you set a modem to not answer incoming calls, you could be 99.99% confident that it wouldn’t silently do it anyway).

We’ve known for years that all always-internet-connected computers, regardless of OS, are pretty much susceptible to remote log-in, if the other side is sufficiently motivated and funded e.g. if a state actor decides it’s worth it to plant incriminating files on a bothersome journalist’s computer, they have often been able to do it.

Reversing the flow, clearly a sufficiently clever algorithm would try to “break out”, and would most likely succeed. If we assume pretty much everything has a “back door” (not to mention unofficial ones due to bugs), an algorithm with access to huge amounts of info, and that tries 24/7, will eventually find a way to make a back-door that swings in, swing out.

My reply:

Reversing the flow, clearly a sufficiently clever algorithm would try to “break out”, and would most likely succeed. If we assume pretty much everything has a “back door” (not to mention unofficial ones due to bugs), an algorithm with access to huge amounts of info, and that tries 24/7, will eventually find a way to make a back-door that swings in, swing out.

I think you’re missing a chunk. Remember, even a PC is a supercomputer. So 24/7 processing is likely unnecessary for generating dozens of previously unimagined ways to network (or whatever). An AI could sift through billions of possibilities in the blink of an eye, and execute even more swiftly. The brunt of the "exponential mayhem" would happen before we could even begin to parse it (though the malevolent human part of the ping-ponging might require more time). That's the part Kurzweil got right. Per Hemingway's famous bon mot about bankruptcy, it would happen gradually, then suddenly.

Buried in the Brian Rommele mega thread my friend linked me to (to start this discussion) is a readable piece worth checking out. “The Case For The AI Prompt Engineer”, mentioned here, seems like the third query/prompt must-read tutorial, after the two I offered previously here (this and this).


Friend-of-the-Slog Paul Trapani chimes in:

Here's another rabbit hole. Let's users create an AI bot based on a prompt. Further on the thread says they are working on letting users host on their own servers allowing for more complex bots. 

This came from Ben's Bites Newsletter today. I just signed up for it and like the format and content so far.