Also, any solution to the alignment problem must suffer from is-ought confusion when presented in plain language rather than extensively worked out theoretical equations with extensive empirical verification.
Which part would you have me remove, the plain language, the extensively worked out theoretical equations, or my list of open problems that I hope people will use to help me assemble extensive empirical verification of my work?
On reflection, I suspect that I’m struggling with the is-ought problem in the entire project. Physics is “is” and ethics is “ought”, and I’m very skeptical that “ethicophysics” is actually either, let alone a bridge between the two.
If you consult my recent shortform, I lay out a more measured, skeptical description of the project. Basically, ethicophysics constitutes a globally computable Schelling Point, such that it can be used as a protocol between different RL agents that believe in “oughts” to achieve Pareto-optimal outcomes. As long as the largest coalition agrees to prefer Jesus to Hitler, I think (and I need to do far more to back this up) defectors can be effectively reined in, the same way that Bitcoin works because the majority of the computers hooked up to it don’t want to destroy faith in the Bitcoin protocol.
I suspect we have a disagreement about whether the “worked out theoretical equations” suffer from is-ought any less than the plain language version. And if they are that fundamentally different, why should anyone think the equations CAN be explained in plain language.
I am currently unwilling to put in the work to figure out what the equations are actually describing. If it’s not the same (though with more rigor) as the plain language claims, that seriously devalues the work.
Check out my post entitled “Enkrateia” in my sequence. This is a plain language account of a safe model-based reinforcement learner using established academic language and frameworks.
Also, any solution to the alignment problem must suffer from is-ought confusion when presented in plain language rather than extensively worked out theoretical equations with extensive empirical verification.
Which part would you have me remove, the plain language, the extensively worked out theoretical equations, or my list of open problems that I hope people will use to help me assemble extensive empirical verification of my work?
On reflection, I suspect that I’m struggling with the is-ought problem in the entire project. Physics is “is” and ethics is “ought”, and I’m very skeptical that “ethicophysics” is actually either, let alone a bridge between the two.
That’s fair (strong up/agree vote).
If you consult my recent shortform, I lay out a more measured, skeptical description of the project. Basically, ethicophysics constitutes a globally computable Schelling Point, such that it can be used as a protocol between different RL agents that believe in “oughts” to achieve Pareto-optimal outcomes. As long as the largest coalition agrees to prefer Jesus to Hitler, I think (and I need to do far more to back this up) defectors can be effectively reined in, the same way that Bitcoin works because the majority of the computers hooked up to it don’t want to destroy faith in the Bitcoin protocol.
I suspect we have a disagreement about whether the “worked out theoretical equations” suffer from is-ought any less than the plain language version. And if they are that fundamentally different, why should anyone think the equations CAN be explained in plain language.
I am currently unwilling to put in the work to figure out what the equations are actually describing. If it’s not the same (though with more rigor) as the plain language claims, that seriously devalues the work.
Check out my post entitled “Enkrateia” in my sequence. This is a plain language account of a safe model-based reinforcement learner using established academic language and frameworks.