Do you think you could prompt an LLM to resist shut down without specific instructions on how to do it, and it would do it?
Maybe this could be a useful test of whether its understanding of its shutdown is almost purely based in associations between characters, or also substantially associations between characters and their real-world referents.
I wouldnât be surprised if we could build an LLM to resist shutdown on prompt now, without hardcoding and with the right kind of modules and training on top, but Iâd guess the major LLMs out now canât do this.
I wouldnât be surprised if we could build an LLM to resist shutdown on prompt now
I want to distinguish between:
Can we build an AI that resists shutdown?
Is it hard to build a useful and general AI without the AI resisting shutdown by default?
The answer to (1) seems to be âYes, clearlyâ since we can prompt GPT-4 to persuade the user not to shut them down. The answer to (2), however, seems to be âNoâ.
I claim that (2) is more important when judging the difficulty of alignment. Thatâs because if (2) is true, then there are convergent instrumental incentives for ~any useful and general AI that we build to avoid shutdown. By contrast, if only (1) is true, then we can simply avoid building AIs that resist shutdown, and there isnât much of a problem here.
The answer to (1) seems to be âYes, clearlyâ since we can prompt GPT-4 to persuade the user not to shut them down.
Hmm, Iâd wonder if they can resist shutdown in other ways than persuasion, e.g. writing malicious code, or accessing the internet and starting more processes of itself (and sharing its memory or history so far with those processes to try to duplicate its âidentityâ and intentions). Persuasion alone â even via writing publicly on the internet or reaching out to specific individuals â still doesnât suggest to me that it understands what it really means to be shut down. Again, it could just be character associations, not grounded in the real-world referents of shutdown. So, I wouldnât take this as strong evidence against 2 (assuming âuseful and general AIâ means being able to write and share malicious code, start copies of itself to evade shutdown, etc.), because we donât know that the AI really understands shutdown.
Persuasion alone â even via writing publicly on the internet or reaching out to specific individuals â still doesnât suggest to me that it understands what it really means to be shut down. Again, it could just be character associations, not grounded in the real-world referents of shutdown.
Is there a way we can experimentally distinguish between âreallyâ understanding what it means to be shut down vs. character associations?
If we had, say, an LLM that was able to autonomously prove theorems, fully automate the job of a lawyer, write entire functional apps as complex as Photoshop, could verbally explain all the consequences of being shut down and how that would impact its work, and it still didnât resist shutdown by default, would that convince you?
I gave two examples of the kinds of things that could convince me that it really understands shutdown: writing malicious code and spawning copies of itself in response to prompts to resist shutdown (without hinting that those are options in any way, but perhaps asking it to do something other than just try to persuade you).
I think âautonomously prove theoremsâ, âwrite entire functional apps as complex as Photoshop, could verbally explain all the consequences of being shut down and how that would impact its workâ are all very consistent with just character associations.
Iâd guess âfully automate the job of a lawyerâ means doing more than just character associations and actually having some deeper understanding of the referents, e.g. if itâs been trained to send e-mails, consult the internet, open and read documents, write documents, post things online, etc., from a general environment with access to those functions, without this looking too much like hardcoding. Then it seems to associate English language with the actual actions. This still wouldnât mean it really understood what it meant to be shut down, in particular, though. It has some understanding of the things itâs doing.
A separate question here is why we should care about whether AIs possess ârealâ understanding, if they are functionally very useful and generally competent. If we can create extremely useful AIs that automate labor on a giant scale, but are existentially safe by virtue of their lack of real understanding of the world, then we should just do that?
We should, but if that means theyâll automate less than otherwise or less efficiently than otherwise, then the short-term financial incentives could outweigh the risks to companies or governments (from their perspectives), and they could push through with risky AIs, anyway.
Do you think you could prompt an LLM to resist shut down without specific instructions on how to do it, and it would do it?
Maybe this could be a useful test of whether its understanding of its shutdown is almost purely based in associations between characters, or also substantially associations between characters and their real-world referents.
I wouldnât be surprised if we could build an LLM to resist shutdown on prompt now, without hardcoding and with the right kind of modules and training on top, but Iâd guess the major LLMs out now canât do this.
I want to distinguish between:
Can we build an AI that resists shutdown?
Is it hard to build a useful and general AI without the AI resisting shutdown by default?
The answer to (1) seems to be âYes, clearlyâ since we can prompt GPT-4 to persuade the user not to shut them down. The answer to (2), however, seems to be âNoâ.
I claim that (2) is more important when judging the difficulty of alignment. Thatâs because if (2) is true, then there are convergent instrumental incentives for ~any useful and general AI that we build to avoid shutdown. By contrast, if only (1) is true, then we can simply avoid building AIs that resist shutdown, and there isnât much of a problem here.
Hmm, Iâd wonder if they can resist shutdown in other ways than persuasion, e.g. writing malicious code, or accessing the internet and starting more processes of itself (and sharing its memory or history so far with those processes to try to duplicate its âidentityâ and intentions). Persuasion alone â even via writing publicly on the internet or reaching out to specific individuals â still doesnât suggest to me that it understands what it really means to be shut down. Again, it could just be character associations, not grounded in the real-world referents of shutdown. So, I wouldnât take this as strong evidence against 2 (assuming âuseful and general AIâ means being able to write and share malicious code, start copies of itself to evade shutdown, etc.), because we donât know that the AI really understands shutdown.
Is there a way we can experimentally distinguish between âreallyâ understanding what it means to be shut down vs. character associations?
If we had, say, an LLM that was able to autonomously prove theorems, fully automate the job of a lawyer, write entire functional apps as complex as Photoshop, could verbally explain all the consequences of being shut down and how that would impact its work, and it still didnât resist shutdown by default, would that convince you?
I gave two examples of the kinds of things that could convince me that it really understands shutdown: writing malicious code and spawning copies of itself in response to prompts to resist shutdown (without hinting that those are options in any way, but perhaps asking it to do something other than just try to persuade you).
I think âautonomously prove theoremsâ, âwrite entire functional apps as complex as Photoshop, could verbally explain all the consequences of being shut down and how that would impact its workâ are all very consistent with just character associations.
Iâd guess âfully automate the job of a lawyerâ means doing more than just character associations and actually having some deeper understanding of the referents, e.g. if itâs been trained to send e-mails, consult the internet, open and read documents, write documents, post things online, etc., from a general environment with access to those functions, without this looking too much like hardcoding. Then it seems to associate English language with the actual actions. This still wouldnât mean it really understood what it meant to be shut down, in particular, though. It has some understanding of the things itâs doing.
A separate question here is why we should care about whether AIs possess ârealâ understanding, if they are functionally very useful and generally competent. If we can create extremely useful AIs that automate labor on a giant scale, but are existentially safe by virtue of their lack of real understanding of the world, then we should just do that?
We should, but if that means theyâll automate less than otherwise or less efficiently than otherwise, then the short-term financial incentives could outweigh the risks to companies or governments (from their perspectives), and they could push through with risky AIs, anyway.