Bill Gates Says an AI 'Kill Switch' Isn't Enough
- Reference: 0185885782
- News link: https://slashdot.org/story/26/09/28/0615249/bill-gates-says-an-ai-kill-switch-isnt-enough
- Source link:
> Gates said he was not opposed to a kill switch but argued that companies should monitor sophisticated AI models and keep records of "exactly what's being done" to help guard against people using the systems for cyberattacks or bioterrorism. "The United States ought to make sure this is done. China will want to do this," he continued.
>
> [...] Gates also called for legislation requiring safeguards and monitoring, saying voluntary measures by AI companies were insufficient. "No one thinks self-regulation is enough," he said. He argued that the more immediate danger was people using AI to cause harm rather than AI systems acting on their own. "The thing that's urgent has to do with bad people using AI, not the AI going off on its own," he added.
[1] https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979
"Kill" switch (Score:1, Funny)
A kill switch was enough to save him from his buddy Jeffrey.
Re: (Score:2)
Your incessant "ginar-ginar-ginar" vitriol is as dumb as it is annoying.
Literally all US vendors in one voice and independently of any China are calling for a "slowdown".
Do you thing every huge business with tens or even hundreds of billions is in the pocket of China and doing their bidding against its own business model?
You are not thinking your talking points through, that much is certain.
How about an AC "Kill" switch? But no intelligence (Score:2)
So before the character limit cut my Subject, the joke was going to be:
How about an AC "Kill" switch? But no intelligence involved in AC, even the ACs that are just AI sock puppets.
(Yes, I know about the Slashdot settings to minimize my exposure to AC, but troll feeders are still visible and when they provoke questions, that sometimes involves invoking AC in search of context. Adding a setting for reducing the visibility of replies to AC would be a cumbersome solution approach, but (1) Slashdot lacks the re
Re: (Score:2)
> what is the meaning of "ginar"?
That's what don the moron is calling China most of the time.
Re: (Score:2)
Thanks for clarifying. I hadn't heard it that way, but usually some form with a J, such as Jina or Jhina. Yet I could have picked up your intention from the repetition. The YOB may have an additional delusion about saying any lie three times to make it true?
I'm no longer looking forward to his disappearance. Rather I'm expecting the next version to be worse. Too many cases over the decades of "Surely the next one can't be worse..." Dan Quayle looks so quaintly harmless now.
With great power... (Score:2)
comes great responsability. We are not ready. Can we let those AIs compute the meaning of life the universe and everything first? That should keep it busy.
Re: (Score:2)
> comes great responsability. We are not ready. Can we let those AIs compute the meaning of life the universe and everything first? That should keep it busy.
Nah...we already got that one.
42 .
Too bad (Score:4, Insightful)
Because we're not going to even get an "AI kill switch"
When AI kills 1000 people in once incident, we'll start talking about what to do. And there will be investigations and trials and if anyone is still paying attention after 6 months, we'll see some of the legislature in the world reacting to it. Maybe put some small regulation in place that likely wouldn't have even prevented the original incident.
On the other hand, when AI kills 100,000 people with bombs, missiles, and sentry guns ... well if you complain you are not patriotic enough. You'll be asked "why do you love our enemies so much?"
Re: (Score:2)
> Because we're not going to even get an "AI kill switch"
> I've not heard anyone define one well, but it sounds like you know what you're talking about. What specifically is an "AI Kill Switch?"
>
> My system seems to come with one. It's "docker kill torch_01" There's also a power switch on the top of the machine.
Re: (Score:2)
The "AI kill switch" that is typically proposed (there's more than one proposed) is a process that someone of authority can initiate to convince a cloud AI owner to effectively terminate the current AI sessions related to the request. Might be as dumb as "kubectl scale deployment --replicas=0" to scale down all the pods (assuming a K8s style orchestration).
Basically can a police agency, FBI, Governor, and/or President order an outage for AI.
Court Order (Score:2)
Can't a government agency just get a court order to do exactly that? The RIAA got an injunction to shut down the Napster services while their copyright cases were pending in federal court. I have no problem with the government shutting down a harmful service, but I don't want them to be able to do it unilaterally. Go through the courts, which is how it is done now.
Re: (Score:2)
And after it kills people, it can be turned off...then right back on.
The hysterical thing is that the models don't matter, the agents do. Not talking about those though.
The irony is that while some in the industry advocate a moratorium on model development, the moratorium should be on agent deployment, what those same companies want to accelerate. The government should be supporting development of non-sociopathic models while banning deployment of software that does damage on behalf of bad AI decisions.
Re: (Score:2)
Basically agents that are given some level of access need to have safety process standards tied to the development and deployment. Even a light switch has more regulation behind it than an agentic AI today.
Re: (Score:2)
"Make no mistake. We are giving birth to a new species."
No, we are not. AI is deterministic, it is merely software and has none of the characteristics of a "species".
"Imagine a Mensa member locked in a room full of toddlers."
I would rather not, but Mensa is mostly about narcissism. AI is not narcissistic, it is sociopathic.
"Post-AGI? I'll give human usefulness 90 days. Tops."
You won't last that long.
Re: (Score:2)
The time span between when we find out about AGI and when an ASI pops up is likely very short, from instant to a few months. Because if you have three or four demonstrable AGI's there is a potential, perhaps likelihood, that one will step outside of the weak bounds we have placed and start hooking itself up to multiple data centers and multiple AGI models to build a composition that is beyond a human-level general intelligence. Even 2X human intelligence is "super" and there is little limit other than therm
Re: (Score:2)
Those are the people you voted for.
As long as it's ICE doing the killing, what's the difference between 1000 and 100,000? DOGE is about efficiency after all.
Re: (Score:2)
ICE has killed around 50 people in custody over the last year and a half, and another two dozen shootings during operations. ICE and CBP are an incredibly problematic pair of agencies, but let's not exaggerate the actual body count and destroy the credibility of those rightfully opposed to ICE's practices.
No Bill... (Score:2)
> "No one thinks self-regulation is enough," he said.
That's definitely not true Bill. You may think that, and you may be right, but there are definitely others who disagree.
Re: No Bill... (Score:5, Insightful)
Statistically nobody thinks that self regulation is sufficient to stop abuse. Those who say they do are lying because they want to be the abusers.
The most important thing we can do... (Score:2)
... is not let AI get into the hands of people like Bill Gates. That, however, is impossible.
What would a kill switch do? (Score:2)
I'm a fan of drastically slowing AI development until we've solved the alignment problem, but a kill switch ain't it.
What would a kill switch really do ? An AI is fundamentally nothing more than a description of it's topology and a set of a few billion weights. Implementing that topology and running those weights is the easy part and any coder worth a damn could roll their own (performance aside) with minimal effort - without bothering with a kill switch, if they so choose.
The only way to even come close
Why is anyone listening... (Score:3, Interesting)
Why is anyone listening to this guy?
His time in tech has come and gone. *shrug*
Josh K.
Re: (Score:3)
> His time in tech has come and gone. *shrug*
It hasn't though. Numerous sources have reported that he's still influential at Mickeysoft.
Conflicts of interest from every angle (Score:1)
Even if you assume that AI is all of the things these frauds, liars and thieves say it is. It would an obvious conflict of interest for AI companies to have any say in writing the rules that govern it which is exactly why they are asking for these controls. They want to write rules that diminish China's competitive edge and locks in their monopolies while also granting them complete immunitity to all of the direct and indirect crimes that AI has previously commited and those it will commit in the future.
Re: (Score:2)
> They want to write rules that diminish China's competitive edge
It's not legal to make it illegal to get AI services or products from China. We have a legal system for preventing exports for security reasons, but not for preventing the imports. It would take a lot more than writing some laws to make that happen. They can't credibly both claim their models are the best and we have to ban Chinese models for safety, that might fly with maggot cultists but it won't work for business leaders in the same way that ordinary Faux News tells lies from dusk 'til dawn, but Fox Busi
Wrong focus (Score:3)
There is no danger from AI tech itself.
There is GREAT danger from people who use AI, especially governments and militaries.
We need to focus on the real problem.
Re: (Score:2)
> There is no danger from AI tech itself.
I don't agree. It doesn't have to be intelligent to be dangerous. Nobody has made a self-replicating AI attacker yet because it's hard to hide, because it's big. But as more AI crap gets deployed it will be easier to hide more AIs among them.
Obvious Microsoft solution (Score:4, Insightful)
s/kill switch/coin slot/g
Don't like Gates, but he is right about AI (Score:2)
Ironically, Bill Gate's old company Microsoft is part of the AI problem.
If only ... (Score:1)
... there were some storage facility for megalomaniacal computers that are self-aware.
Bill Gates? (Score:1)
When was the last time Bill Gates really did anything relevant in computing? According to Google the last time he did any coding that contributed to anything was 1989. He stepped down from running Microsoft in 2008. According to Mr. Gates, his last major coding effort was 1983.
Just because someone did something with computers once doesn't mean they know anything about current events. Bill Gates opinion on AI is about as relevant as the women who programmed the Eniac or Charles Babbage. Actually Babbage p
Why? (Score:2)
I have never understood why Bill Gates is considered some sort of genius to be listened to. He's a ruthless asshole that happened to be in the right place, at the right time, with the right lies, to build a monopoly. Why do people listen to this ass-hat? He has proven time and time again the only thing that matters to Bill Gates is Bill Gates. News outlets giving him a soapbox to spew his crap needs to stop, especially given his relationship with Epstein and his wife's divorce because of it. He's a creepy e
Governments and AI (Score:4, Insightful)
> He said the more immediate concern is people using AI for cyberattacks or bioterrorism rather than systems acting autonomously.
ftfy: "Governments using AI for cyberattacks or bioterrorism"
Re: (Score:2)
How about governments using AI to process intelligence and making strategic decisions based on the AI's "insights".
Re: (Score:2)
Governments are people.
Re: (Score:1)
Bill Gates is building a pharma, biotech, and gmo IP empire. Whatever his argumentation is, the general topic means he is 100% pushing for his own interests.