SUPPORTER: America needs to start talking to China to come up with a bilateral agreement to pause AI. The agreement would need to be transparent, mutually enforceable, and…
OPPONENT: We can’t unilaterally pause AI! China would destroy us!
SUPPORTER: As I said, we need to start negotiating a bilateral agreement so that both sides will…
OPPONENT: You fool! Don’t you know that while we unilaterally pause AI, China will be racing ahead and using their lead to erode our fundamental rights and freedoms? How could you be so naive!
SUPPORTER: Look, I promise this is about negotiating for a mutual pause. We don’t think a unilateral pause would work any more than you would. But we think that if we negotiate…
OPPONENT: And while we unilaterally pause, do you think China will just be twiddling their thumbs, doing nothing? Obviously not! This is about ceding the future to our rivals!
SUPPORTER: I get the feeling you’re not listening to me.
OPPONENT: Just like China won’t listen to us when we ask them nicely not to destroy us with the advanced AI they developed while we unilaterally paused like chumps!
SUPPORTER: Okay, let’s back up. Is your problem that you don’t think China would agree to a pause in negotiations? Because we’ve actually had some pretty successful low-level discussions with Chinese scientists. And they’re losing the race, so their incentive to pause is stronger than ours. Xi has expressed some concern about the risks of AI and the importance of alignment - nothing super-strong, but more than our government has done. We agree it’s not obvious that China would agree to pause, but we think we should get the offer out there, and maybe work on a preliminary framework that we could use to pause later, if we got a warning shot and both of our governments became more amenable.
OPPONENT: No, my problem is that you want to unilaterally pause, while China rushes forward! That’s dangerously close to treason!
SUPPORTER: Or is your problem that you don’t trust China to stick to an agreement, once signed? Because we agree that an agreement has to be mutually transparent and enforceable. We have some ideas for how we could have a light-touch approach to monitoring Chinese data centers - of course, they would get to monitor ours in the same way - and actually the math mostly works out and we think it would be less intrusive than other things that have worked in the past, like nuclear monitoring.
OPPONENT: You foolishly think that if America paused, everything would be fine. But there’s a flaw in your utopian high modernist plan - our enemies won’t pause!
SUPPORTER: Or is your problem that you think AI will deliver lots of benefits, so it would be foolish to pause? I agree the benefits of AI would be great, and I think there are ways we could try to maximize those benefits even during a pause. For example, we and China could try to build the infrastructure for a pause, put a mutual red line in place for activating the pause, and then have green lines in place for what sorts of control schemes we would need to see before winding down the pause and continuing to advance. It wouldn’t be a total stop on AI improvement so much as an attempt to do it in a monitored way, with the US government, Chinese government, and scientific community all having input. I know it’s reasonable to worry that such a graduated strategy could devolve into a more extremist Luddite approach, but there are steps we could take to make that less likely.
OPPONENT: I feel like you’re not listening to me at all! The problem is that while you frolic in your hippie-dippie flower world of unilateral pauses, China races ahead to the prize!
SUPPORTER: Or is your problem that you’re worried about the economic consequences of getting rid of existing chatbots? Because a pause would just mean that China and ourselves slow down training new AIs. Inference - running the kinds of AI that people use now - could keep going ahead as planned in both countries.
OPPONENT: But what about China? While we pause training, they would train faster than ever!
SUPPORTER: I’m getting exasperated here. There are lots of reasons to be worried about an AI pause - starting with the possibility that China wouldn’t agree to it, or that they might agree but then secretly defect against us by trying to get around the agreement. I’m excited about debating those concerns with you. But it seems like we can’t get past you asserting that I want a unilateral pause, which just isn’t true. Almost nobody wants a unilateral pause! Pause AI, the biggest activist group in this area, says:
We are primarily asking for an international pause, enforced by a treaty . . . such a treaty also needs to be signed by China.
Eliezer Yudkowsky, the most famous pause proponent, writes in his book that:
The goal is not to have your country unilaterally cease AI research and fall behind. It is to have enough major powers express willingness to halt the suicide race, worldwide, that your home country will not be placed at a disadvantage if you agree to stop climbing the AI escalation ladder.
David Krueger, the keynote speaker at the recent round of AI pause protests, said:
It's actually quite simple: [First,] company leaders agree to a conditional pause, [then] US and China agree to a conditional pause, [then] international pause. Notice how no step here involves "US unilaterally pauses"
…and added that “I concentrate on America [because] China has shown more interest in slowing down and regulating.” If you think someone is demanding a unilateral pause, I think you have a responsibility to say who it is you’re talking about. If you can really find someone like this, I’ll criticize them just as hard as you are.
OPPONENT: You think you’re getting exasperated? I don’t see you responding to my key point, which is that if we institute a unilateral pause like you’re suggesting, China will beat us, and we’ll lose all our freedoms and have to learn Chinese and draw a thousand squiggly characters every time we want to communicate! And all because you were too stupid to realize that it doesn’t make sense for only one side in a race to pause and hope for the best!
SUPPORTER: Forget it. This debate is over.
OPPONENT: See, it’s just like you to unilaterally declare this debate over! You don’t realize that even if you want to pause the debate, I can just keep speaking! Exactly what I would expect from a gullible fool who want to cede the AI race to China by pausing unilaterally! What you don’t realize is that while we pause, Chairman Xi will be … will be … (*faintly, barely audible*) Hey, who cut my mic?
Astral Codex Ten.
Интересное...
- ...Луи Ли Ронг, скульптор
- ...Хулиновости: Иран побеждает Трампа, удар по эро-чат-ботам, ставки на войну
- ...Все дебаты о приостановке Искусственного Интеллекта
0 коммент.:
Отправить комментарий
Ваш комментарий появится в блоге после проверки администратором