Recursive Altruism
Consider a planet populated by perfect altruists, where each optimization processes’ utility function is determined by referring to the utility functions of everything around them. When altruist A consider action X, ve judges X’s desirability by looking up X’s desirability to altruist B, to altruist C, and so on. But B’s utility function, in turn, depends on the utility function of A, which depends on the utility function of B, and so on. The calculation collapses in an infinite recursion.
In real human morality, there’s a strong dampening effect on these sorts of recursions. If the utility of X to me is 5, and the utility of X to my neighbor is 7, I might rate X’s total utility anywhere from 5 (total selfishness) to 12 (total utility equivalence). My neighbor, however, isn’t likely to compute that the total utility of X is now 19, even though my valuation of X is 12 and my neighbor’s valuation is 7. Perhaps we treat “direct utility” (= personal desirability) differently from “indirect utility” (= desirability for others). On the other hand, this might not be necessary; if we aren’t perfect altruists, the utility recursion will sum to a finite number anyway through exponential decay. If me and my neighbor are both 25% altruistic, the desirability of X converges to 5 + 1.75 + .3125 … = 7.2, and all the higher recursion levels are assigned negligible utility.
Irregardless, in a perfectly altruistic world, there’s no way to get any finite desirability for any action. Without damping of some sort, any calculation will collapse into an infinite recursion. With damping, any calculation will eventually sum to zero, as the utility function of the perfect altruists will always assign a zero if it is referenced directly. Such a morality, although noble-sounding, is not a stable basis for a civilization.
Peter de Blanc said,
February 19, 2008 @ 11:40 pm
Mike Vassar uses the term “local hedonic function” for the selfish component of one’s utility function. I don’t see any reason why there couldn’t be a planet of altruists with different local hedonic functions but the same utility function.
Tom McCabe said,
February 20, 2008 @ 12:38 am
“Mike Vassar uses the term “local hedonic function” for the selfish component of one’s utility function.”
He should publish more. No point in reinventing the wheel.
“I don’t see any reason why there couldn’t be a planet of altruists with different local hedonic functions but the same utility function.”
That’s perfectly plausible, I was referring to a planet of perfect altruists (= no local hedonic function).
Nick Tarleton said,
February 20, 2008 @ 2:06 pm
If nobody has a local hedonic function, where does the differential desirability that enables altruism come from in the first place? (If everybody only cares about total global utility, how can you actually increase anybody’s utility?)
How about “each optimization process’s utility function is the sum of everyone’s local hedonic function” as a definition of perfect altruism? This seems more ethically and psychologically realistic; the value of a life on the other side of the world shouldn’t depend on the number of altruists in the world. This still suffers from recursion when two people’s local hedonic functions refer to each other, but damping is more acceptable in that case.
Nick Tarleton said,
February 20, 2008 @ 2:42 pm
s/more acceptable/real/
Tom McCabe said,
February 20, 2008 @ 7:33 pm
“If everybody only cares about total global utility, how can you actually increase anybody’s utility?”
This is my question exactly.
“as a definition of perfect altruism?”
No point in haggling over definitions. See http://www.overcomingbias.com/2008/02/disputing-defin.html.
Nick Tarleton said,
February 20, 2008 @ 8:28 pm
Heh, I should have known better. Still, if we’re going to say “altruism” we might as well figure out the most useful way to think of it.