Emmett on Interesting Essays

I completely agree that values are constructed from self-observation in much the manner you describe, but that doesn't make them unreal. In fact, your values may be more stable and universal than many of your other attributes, they are the invariants produced by your dynamics. If you zoom in enough on any object, you'll find that it is a periodic attractor of dynamics from parts that might not even be aware it exists. The lipids and proteins of a cell don't really belong to that cell, they are sometimes there and sometimes not. A cell is a bunch of biochemicals "performing cell"...yet I think it is mistaken to say that cells are in any way unreal or even less real than their constituent biochemicals. The cell is more enduring, more robust, more solid than its parts are. Similarly, values are more enduring, more robust, more solid than most of our beliefs, because they are belief attractors. Nonetheless I think your view on value-functions is correct. If a value is a predictable attractor in goal-space, trying to construct "the correct set" of values is clearly a mistake.