First, I would like to reintroduce a very known problem in philosophy and on the Less Wrong as well, and it is the problem of ought. It resides in the inability of facts about reality to imply facts about values. According to science we can say that by cutting our veins we start losing our blood, by which our brain starts losing the blood too, by which it will lack too many nutrients and our neurons start to die, by which our brain loses their typical basic functions like letting us see, hear, think, etc. But science cannot in principle tell us that any of these processes or outcomes is bad. And I would like to tackle the Alignment Problem by tackling the ought problem first. And I will do this based on an idea that occurred in my mind sometime.
According to this idea, the problem of ought and scientific facts is caused by the way science has worked. I mean the currently used scientific method has no room for ought. Thus, I suggest a partial solution for this.
The currently used scientific method explains everything by necessity. The extended science, which my solution is about, would possess an additional category besides the necessity: preference. Concretely, the preference of certain states by the fundamental building blocks. The description of fundamental building blocks containing both necessity and preference would be still optimized for simplicity (and fitness to observational data of course), but besides that it would be optimized also for preference-to-necessity complexity ratio, so the additional principle would sound: what can be explained by preference, don't explain by necessity.
The goal of this change is to create some optimization proto-criterion for selecting “better” states. And I assume the better states are the more preferred states.
And the whole morality would be some (positive-) justice principle across all fundamental building blocks according to that proto-criterion.
But to keep this post short, I will end here. But since this is just an idea, please treat it accordingly.
First, I would like to reintroduce a very known problem in philosophy and on the Less Wrong as well, and it is the problem of ought. It resides in the inability of facts about reality to imply facts about values. According to science we can say that by cutting our veins we start losing our blood, by which our brain starts losing the blood too, by which it will lack too many nutrients and our neurons start to die, by which our brain loses their typical basic functions like letting us see, hear, think, etc. But science cannot in principle tell us that any of these processes or outcomes is bad. And I would like to tackle the Alignment Problem by tackling the ought problem first. And I will do this based on an idea that occurred in my mind sometime.
According to this idea, the problem of ought and scientific facts is caused by the way science has worked. I mean the currently used scientific method has no room for ought. Thus, I suggest a partial solution for this.
The currently used scientific method explains everything by necessity. The extended science, which my solution is about, would possess an additional category besides the necessity: preference. Concretely, the preference of certain states by the fundamental building blocks. The description of fundamental building blocks containing both necessity and preference would be still optimized for simplicity (and fitness to observational data of course), but besides that it would be optimized also for preference-to-necessity complexity ratio, so the additional principle would sound: what can be explained by preference, don't explain by necessity.
The goal of this change is to create some optimization proto-criterion for selecting “better” states. And I assume the better states are the more preferred states.
And the whole morality would be some (positive-) justice principle across all fundamental building blocks according to that proto-criterion.
But to keep this post short, I will end here. But since this is just an idea, please treat it accordingly.