re “anarchy”, you’ll have a hard time convincing me that lack of an ai monarch is bad. use of a word that describes lack of central authority to describe safety failure seems to me to imply a fundamental misunderstanding of what it is we need to protect, which is everyone’s individual agency over their own shape.
safety needs to be a solution to an incentive problem; if your solution to safety is to not give individuals the means of intelligent computation, then you aren’t able to solve safety at all. we need to be thinking about safety from a multi-agent systems perspective because I think at this point we can conclude that AI safety is just a new interspecies version of the same conflicts of intention that lead to all human and animal fights and wars. The difficulty is how to ensure that, in a world where everyone can imagine other people in compromising positions and then share that imagination, we can protect each other’s dignity. What content filter would you choose to apply to your own view of the world in order to respect people who don’t want certain images of them to be seen? can we make good tools for automatically filtering out content by your own choice, without relying on the incredibly fragile authority of platforms?
I would propose that what we need to be optimizing for is to end up in a pro-social anarchy rather than a destructive anarchy. authority cannot last; So authority needs to be aiming to build structures that will preserve each individual’s rights at least as well as the authorities did, but keeping in mind that the network of co-protective agents need to be able to operate entirely open source.
I would suggest checking out open source game theory, a subset of game theory research that focuses specifically on what you can do when you can verify that you have both read each other’s minds perfectly in a game theory situation. it’s not a full solution to the safety problem and it doesn’t plug directly in to current generation high noise models, but it’s relevant to the fact that it is not possible to ban AI as long as people have computers; we need to aim for a world where publishing new capability developments that will be cloned open source will change the game such that the price of anarchy decreases to 1.
I think it is clear that we are not advocating for centralized authority. All of the three points in “takeaways” lead into this. The questions you asked in the second paragraph are ones that we discuss in the “what do we want section” with some additional stuff in the EA forum version of the post.
Without falling into the trap of debating over definitions. Anarchy can be used colloquially to refer to chaos in general and it was intended here to mean a lack of barriers to misuse in the regulatory and dev ecosystems—not a lack of someone’s monopoly on something. If you are in favor of people not monopolizing capabilities, I’m sure you would agree with out third “takeaways” point.
The “what do we want” section is about solutions that don’t involve banning anything. We don’t advocate for banning anything. The “banning” comments are strawpersoning.
re “anarchy”, you’ll have a hard time convincing me that lack of an ai monarch is bad. use of a word that describes lack of central authority to describe safety failure seems to me to imply a fundamental misunderstanding of what it is we need to protect, which is everyone’s individual agency over their own shape.
safety needs to be a solution to an incentive problem; if your solution to safety is to not give individuals the means of intelligent computation, then you aren’t able to solve safety at all. we need to be thinking about safety from a multi-agent systems perspective because I think at this point we can conclude that AI safety is just a new interspecies version of the same conflicts of intention that lead to all human and animal fights and wars. The difficulty is how to ensure that, in a world where everyone can imagine other people in compromising positions and then share that imagination, we can protect each other’s dignity. What content filter would you choose to apply to your own view of the world in order to respect people who don’t want certain images of them to be seen? can we make good tools for automatically filtering out content by your own choice, without relying on the incredibly fragile authority of platforms?
I would propose that what we need to be optimizing for is to end up in a pro-social anarchy rather than a destructive anarchy. authority cannot last; So authority needs to be aiming to build structures that will preserve each individual’s rights at least as well as the authorities did, but keeping in mind that the network of co-protective agents need to be able to operate entirely open source.
I would suggest checking out open source game theory, a subset of game theory research that focuses specifically on what you can do when you can verify that you have both read each other’s minds perfectly in a game theory situation. it’s not a full solution to the safety problem and it doesn’t plug directly in to current generation high noise models, but it’s relevant to the fact that it is not possible to ban AI as long as people have computers; we need to aim for a world where publishing new capability developments that will be cloned open source will change the game such that the price of anarchy decreases to 1.
Did you read the post?
I think it is clear that we are not advocating for centralized authority. All of the three points in “takeaways” lead into this. The questions you asked in the second paragraph are ones that we discuss in the “what do we want section” with some additional stuff in the EA forum version of the post.
Without falling into the trap of debating over definitions. Anarchy can be used colloquially to refer to chaos in general and it was intended here to mean a lack of barriers to misuse in the regulatory and dev ecosystems—not a lack of someone’s monopoly on something. If you are in favor of people not monopolizing capabilities, I’m sure you would agree with out third “takeaways” point.
The “what do we want” section is about solutions that don’t involve banning anything. We don’t advocate for banning anything. The “banning” comments are strawpersoning.
fair enough. I read much of the post but primarily skimmed it like a paper, it seems I missed some of the point. Sorry to write an unhelpful comment!