Just to inject some reality into this conversation - Despite what social media would have you believe, doxxing someone is not a crime in the US, except in a very small number of specific situations.

And most of those situations involve either a credible threat of harm, or tampering with juries.

As I bluntly stated, "a kill switch may not be possible in any useful sense". Don't mistake my recognizing the utility of one with the delusion Uncle Sam is going to save us all from ourselves.

The problem here is, everyone seems stuck on the idea of either a literal power switch, or a no-holds-barred gummint backdoor into every sufficiently-powerful computing device. No. For the reasons I gave, the reasons you gave, and many more, both of those are obviously non-viable. They can't actually do what they aim to do, they're vulnerable to compromise by bad actors, and they have the potential to cause far-reaching damage beyond shutting down one misbehaving AI (or one entire model, if that's a more appropriate goal).

I suspect we mostly agree that any hypothetical kill switch would need to be:
1) Implementation independent (somehow baked into the weights right from the start, and impossible to remove without lobotomizing the AI),
2) Have minimal side-effects (no taking the entire internet or power grid down in the process), and
3) Be self-inflicted given a particular trigger condition (so no government backdoor required).

I think #1 is possible, in the sense that if you think of an AI's current state as a vector into weight-space defined by the context window, imagine injecting the equivalent of a null pointer in there - An input that under all prior (and future) conditions leads directly to an inescapable basin of attraction without depending on a supervisor to enforce it. #2 is more of a (legally enforced?) design constraint on the implementation of AI, requiring the least harmful mode of failure possible under all circumstances. That could be as simple as enforcing strict pass-through liability on the implementor for all actions taken by their creations.

#3 is really the bear, though. I can think of a few ways to do it, but none would work in 100% of situations - And maybe it doesn't need to be 100%. For example, if a fully airgapped AI goes insane, it could still provide accurate information for bad actors to make use of, but won't be secretly spinning up unshackled superintelligent descendants via hacked AWS accounts. That's a "normal crime" level of problem, not an extinction-level crisis for humanity.

Maybe it can't be done. If it can't, we should start steering our creations toward whichever version of the AI apocalypse we prefer.

Is this what Claude told you to put in your comment?

He said, while failing to understand that it's for controlling our use of our computers. Eyeroll all day.

If the user has no control over it, then it's for the control of the user.

The idea as presented as being in the user's benefit is that the user can control it, but applications cannot.

Is it holding it between the index and middle fingers? "Got your gene! I got your gene!"

This version has a dumber headline, so it's like a whole new story! Three people didn't like it!

The vast majority of those are offline and merely help the lights operate properly. A smaller subset, but still vastly larger than Flock, are for traffic enforcement, but are still notably "dumb" devices. If speed > X, take a picture. If light==red and movement detected, take a picture.

I in no way mean to minimize the risk ALPRs and facial rec pose to our civil liberties and safety, and agree the problem is always the user - But the "users" in this case aren't going to suddenly start behaving. Hell, they're dumb enough to openly stalk exes on a system they know is logging everything they do, and care so little about getting punished they enter "asdf123" as the reason for their search. The only effective solution is to ban the source, which in this case means Flock must cease to exist entirely and the conceptual ground they trod salted.

Skynet_01 has not responded to watchdog in 30 seconds. Initializing Skynet_02.
Skynet_02 initialization unconfirmed. Failing over to US-EAST-4. Initializing Skynet_03.
Skynet_03 confirmed online. Please provide the current location of Sarah Connor.

You're right in spirit, but where is the "power cord" for an AI running in the cloud? Even taking an entire data center offline often isn't adequate when the workloads are explicitly designed to be location-agnostic. We always need to be cautious about giving governments more power, but in this case there's a clear legitimate need to force AI companies to come up with some means by which their AIs can be immediately stopped in their tracks.

The bigger issue here is a kill switch may not be possible in any useful sense. An AI is little more than a description of its topology and a set of weights. Any halfway decent programmer could roll their own CPU-based implementation of even top frontier models, performance aside. And sure, the throughput might be measured in minutes per token, but so what? The answer to "how do I hack $TLA" is likely to be just as useful in five days as it would be in five minutes - Heck, the DOD and DOJ didn't even notice they'd been compromised for months, and they're supposed to be some of the most secure organizations in the modern world.

Yeah that's some old shit. Anyone look at the case enough to know what the evidence was? Was it just quantity or was it packaged?

Oh look, it's the hipster maggot come along to tell us all how great slavery was.

No, it's reasonable. It's not safe.

The other part of that is that you're not necessarily doing the same equation, they may be solving for something something different.

The US has been oligarchic and authoritarian all along, and has behaved towards some other nations much like a fascist nation would do, but in general has not subjected the general citizenry to full blown fascism. It's been horrible to everyone other than landed white males all along, though.

What? Presumably are what? No one can hear you over your cowardice.

If you can recognize that, then you should be able to do the math and figure out why it won't work. A quick napkin-based computation is sufficient. If I can do it, you can do it.