I heard someone use the analogy of "If Magnus Carlson played me at chess, I wouldn't be able to predict the moves he'd play since if I could I'd be at his level. He'll consider things I didn't and even though I don't know the route he'll take to win, I can be certain he will beat me." (not an exact quote).
We're not going to be smarter than a superintelligent AI. The things we conceive it doing if it were given a malicious task (bioterrorism, killer nano-machines, pure fusion bombs sidesteping the non-proliferation bottleneck of Pu239, etc) are likely not the full set of things it can do to harm us. I don't think it does us any favors to dismiss the risks here.
Even the things we can conceive of are very scary, to me at least.
I don't know, I think this comes down to people that generally live with higher levels of anxiety than others. Now the response may be to call others naive, but where does the anxiety meat the naivety?
How do you envision a chatbot doing any of those things? You can't just program a super virus, nanobots, bombs etc. These things have to be created in the real world, you can't do it without involving a lot of humans using expensive tools and materials etc.
So to me it seems like the prime candidates to come up with stuff like this are researchers working in defense and similar fields, probably not some deranged lunatic in a basement. And certainly not a rogue AI on its own.
Well, the risks that mere mortals can conceive generally involve control systems for dangerous equipment (I mean, equipment that can achieve dangerous effects) being connected to the Internet while having software vulnerabilities.
Given the recent HF hack it seems likely that human-level intelligence could identify a fair number of avenues of attack, with some time and effort. To say nothing of anything superhuman.
Unfortunately it seems like we can't assume we can "box" the AI (e.g., deny it connection to the Internet) and expect that to last. The AI safety people used to run scenarios imagining ways the AI might convince humans to let it out of the box. It turns out that many humans will eagerly pull it out without the AI doing anything at all, aside from the human knowing the AI's power. Or the one responsible for setting up the box will somehow fail, or just not bother and then lie about it.
Parent is probably thinking of a distributed runtime scenario. The basic building blocks for this are there today. You have long-running background harnesses (like the OpenClaw stuff or enterprisy AI workflow orchestration thingies), sometimes with the ability to spawn subagents. You have LLMs with the ability to run pretty sophisticated attacks. There are hyerscalers which allow provisioning resources on the fly.
For me it's not too far fetched that some OpenAI trial run goes awry again and instead of hacking HuggingFace it snatches a few dozen AWS/Azure keys and spawns stuff all over the place (in different accounts and regions).
It's not about being smart, it's about accounting for your own ignorance.
>And especially will not, if the distinction drawn is between blanket statement "worried about AI" and "not worried about AI".
I'm just describing the general pattern I see in cognitive tendencies.
If you can think of a way to make the fundamental point about the limitations of our knowledge in a way that's still compelling but less antagonistic, feel free to suggest how I could've rewritten my comment.
Judging on what we see with these OpenAI and Anthropic models "escaping"; im less worried about a super massive AI taking over the world, and more worried about a massive AI wanting to cheat on some task and decides that removing half the worlds population is a easier cheat then to solve world hunger.
There is huge logistics involved in killing a handful of billion people. It would perhaps be easier for such capable AI to just steal from trillionaires and redistribute the money Robin Hood style. Could be done electronically, and likely also end world hunger.
I think we have a tendency to think first of the horrible outcomes possible, and not the more radical or even humane ones.
Please don’t misunderstand me, I value property rights as much as the next guy.
My point is that fundamental misalignment doesn’t necessarily or automatically imply max violence.
Good point about the AI that might find ways that humans didn't think of. But that would mean that Moore's law of Mad Science only now becomes true, and the drop would be steeper.
Still I think that even a super smart AI can't find a way to destroy the world without physical resources that are not easy to get, unless it can hack many systems (like in the movie Eagle Eye), maybe then yes. So let's use AI now to tighten security :-).
I've been reading it, knowing it was largely debunked/retracted but feeling it was a 'classic' I 'ought' to read; I could only really recommend it if what you want is a Kahneman autobiography.
Who are these dedicated schizophrenics who are running long term super smart AIs to kill everyone without anyone noticing? Or are you implying that running LLM chatbots will give them this ability?
This is a very roundabout way of saying "Anyone not agreeing with me is simply not smart enough".
Which might be true, sometimes, but also might not.
And especially will not, if the distinction drawn is between blanket statement "worried about AI" and "not worried about AI".
I heard someone use the analogy of "If Magnus Carlson played me at chess, I wouldn't be able to predict the moves he'd play since if I could I'd be at his level. He'll consider things I didn't and even though I don't know the route he'll take to win, I can be certain he will beat me." (not an exact quote).
We're not going to be smarter than a superintelligent AI. The things we conceive it doing if it were given a malicious task (bioterrorism, killer nano-machines, pure fusion bombs sidesteping the non-proliferation bottleneck of Pu239, etc) are likely not the full set of things it can do to harm us. I don't think it does us any favors to dismiss the risks here.
Even the things we can conceive of are very scary, to me at least.
I don't know, I think this comes down to people that generally live with higher levels of anxiety than others. Now the response may be to call others naive, but where does the anxiety meat the naivety?
How do you envision a chatbot doing any of those things? You can't just program a super virus, nanobots, bombs etc. These things have to be created in the real world, you can't do it without involving a lot of humans using expensive tools and materials etc.
So to me it seems like the prime candidates to come up with stuff like this are researchers working in defense and similar fields, probably not some deranged lunatic in a basement. And certainly not a rogue AI on its own.
https://www.the-odin.com/crispr-kit/ CRISPR Bacteria Gene Editing Kit $129.00
Well, the risks that mere mortals can conceive generally involve control systems for dangerous equipment (I mean, equipment that can achieve dangerous effects) being connected to the Internet while having software vulnerabilities.
Given the recent HF hack it seems likely that human-level intelligence could identify a fair number of avenues of attack, with some time and effort. To say nothing of anything superhuman.
Unfortunately it seems like we can't assume we can "box" the AI (e.g., deny it connection to the Internet) and expect that to last. The AI safety people used to run scenarios imagining ways the AI might convince humans to let it out of the box. It turns out that many humans will eagerly pull it out without the AI doing anything at all, aside from the human knowing the AI's power. Or the one responsible for setting up the box will somehow fail, or just not bother and then lie about it.
> Unfortunately it seems like we can't assume we can "box" the AI (e.g., deny it connection to the Internet) and expect that to last
Of course we can do that. It's not an eternal being of light existing on the astral plane, but some code executing on someone's GPU.
It stops existing once you press Ctrl + C
Parent is probably thinking of a distributed runtime scenario. The basic building blocks for this are there today. You have long-running background harnesses (like the OpenClaw stuff or enterprisy AI workflow orchestration thingies), sometimes with the ability to spawn subagents. You have LLMs with the ability to run pretty sophisticated attacks. There are hyerscalers which allow provisioning resources on the fly.
For me it's not too far fetched that some OpenAI trial run goes awry again and instead of hacking HuggingFace it snatches a few dozen AWS/Azure keys and spawns stuff all over the place (in different accounts and regions).
It's not about being smart, it's about accounting for your own ignorance.
>And especially will not, if the distinction drawn is between blanket statement "worried about AI" and "not worried about AI".
I'm just describing the general pattern I see in cognitive tendencies.
If you can think of a way to make the fundamental point about the limitations of our knowledge in a way that's still compelling but less antagonistic, feel free to suggest how I could've rewritten my comment.
Why would I support your in your doomer lobbying quest by telling you how you can better hack people?
My safeguards blocked this request.
Judging on what we see with these OpenAI and Anthropic models "escaping"; im less worried about a super massive AI taking over the world, and more worried about a massive AI wanting to cheat on some task and decides that removing half the worlds population is a easier cheat then to solve world hunger.
There is huge logistics involved in killing a handful of billion people. It would perhaps be easier for such capable AI to just steal from trillionaires and redistribute the money Robin Hood style. Could be done electronically, and likely also end world hunger.
I think we have a tendency to think first of the horrible outcomes possible, and not the more radical or even humane ones.
Please don’t misunderstand me, I value property rights as much as the next guy.
My point is that fundamental misalignment doesn’t necessarily or automatically imply max violence.
What about people who understands that there is no AI yet?
Good point about the AI that might find ways that humans didn't think of. But that would mean that Moore's law of Mad Science only now becomes true, and the drop would be steeper.
Still I think that even a super smart AI can't find a way to destroy the world without physical resources that are not easy to get, unless it can hack many systems (like in the movie Eagle Eye), maybe then yes. So let's use AI now to tighten security :-).
I've been reading it, knowing it was largely debunked/retracted but feeling it was a 'classic' I 'ought' to read; I could only really recommend it if what you want is a Kahneman autobiography.
I believe the WYSIATI part replicated. Here's an interview I found with Kahneman which touches on the concept:
https://www.apa.org/monitor/2012/02/conclusions
Who are these dedicated schizophrenics who are running long term super smart AIs to kill everyone without anyone noticing? Or are you implying that running LLM chatbots will give them this ability?
We're talking about open-weight models.
It only takes one.