If we extend the concept of making beliefs pay rent to structures in computer memory, then AIs could better choose which structures are more valuable than they cost when many objects are shared in an acyclic network. Each object at the bottom could cost 1, and any objects pointing at x equally share the cost of x plus 1 for themself. If beliefs are stored in these memory structures, then a belief would be evicted when its objective cost exceeds some measure of its value, and total value would be in units of memory available. When some beliefs are evicted, those they depended on would become more expensive to others who depend on them because the number of beliefs sharing the cost decreases. On the other hand, if many beliefs depend on a certain structure in memory, those many sharing the cost each pay less.
If we extend the concept of making beliefs pay rent to structures in computer memory, then AIs could better choose which structures are more valuable than they cost when many objects are shared in an acyclic network. Each object at the bottom could cost 1, and any objects pointing at x equally share the cost of x plus 1 for themself. If beliefs are stored in these memory structures, then a belief would be evicted when its objective cost exceeds some measure of its value, and total value would be in units of memory available. When some beliefs are evicted, those they depended on would become more expensive to others who depend on them because the number of beliefs sharing the cost decreases. On the other hand, if many beliefs depend on a certain structure in memory, those many sharing the cost each pay less.