I’ve seen a lot of work on voting systems, and on utility maximization, but very few direct comparisons. But I think that often we can prioritize systems that favor one or the other, and clearly our research efforts are limited between the two, so it seems useful to compare.
Voting systems act very different to utility maximization. There’s a big host of literature on ideal voting rules, and it’s generally quite different to that of utility maximization.
Proposals like quadratic voting are clearly in the voting category, while Futarchy is much more in the utility maximization category. In US Democracy, some decisions are quite clearly made via voting, and others are made roughly using utility maximization (see any cost-benefit analysis, or the way that new legislation is created). Note that by utility I don’t mean welfare, but whatever it is that the population cares about.
My impression is that the main difference is that voting systems don’t need to assume any level of honesty. Utility maximization schemes need to have competent estimates of utility, but individuals are motivated to try to lie about the numbers so that the later calculations come out in their favor. So if we can’t identify any good way of obtaining the ground truth, but people can correctly estimate their own interests, then voting methods might be our only option.
If we can obtain good estimates of utility, then utility optimization methods are normally more ideal. There’s a good reason why the government rarely asks for public votes on things when it doesn’t need to. Voting methods essentially leave out a lot of valuable information, and they often require a lot of overhead.
I think the main challenges for the utility maximization side are to identify trustworthy and inexpensive techniques for eliciting utility, and to find ways of integrating voting that get the main upsides without the downsides. I assume this means something like, “we generally want as little voting as possible, but it should be used to help decide the utility maximization scheme.” For example, one way that this could be done, is that multiple agents could vote on a particular scheme of maximizing utility between them. Perhaps they all donate 10% of their resources to a particular agency that attempts to maximize utility under certain conditions.
Voting systems vs. utility maximization
I’ve seen a lot of work on voting systems, and on utility maximization, but very few direct comparisons. But I think that often we can prioritize systems that favor one or the other, and clearly our research efforts are limited between the two, so it seems useful to compare.
Voting systems act very different to utility maximization. There’s a big host of literature on ideal voting rules, and it’s generally quite different to that of utility maximization.
Proposals like quadratic voting are clearly in the voting category, while Futarchy is much more in the utility maximization category. In US Democracy, some decisions are quite clearly made via voting, and others are made roughly using utility maximization (see any cost-benefit analysis, or the way that new legislation is created). Note that by utility I don’t mean welfare, but whatever it is that the population cares about.
My impression is that the main difference is that voting systems don’t need to assume any level of honesty. Utility maximization schemes need to have competent estimates of utility, but individuals are motivated to try to lie about the numbers so that the later calculations come out in their favor. So if we can’t identify any good way of obtaining the ground truth, but people can correctly estimate their own interests, then voting methods might be our only option.
If we can obtain good estimates of utility, then utility optimization methods are normally more ideal. There’s a good reason why the government rarely asks for public votes on things when it doesn’t need to. Voting methods essentially leave out a lot of valuable information, and they often require a lot of overhead.
I think the main challenges for the utility maximization side are to identify trustworthy and inexpensive techniques for eliciting utility, and to find ways of integrating voting that get the main upsides without the downsides. I assume this means something like, “we generally want as little voting as possible, but it should be used to help decide the utility maximization scheme.” For example, one way that this could be done, is that multiple agents could vote on a particular scheme of maximizing utility between them. Perhaps they all donate 10% of their resources to a particular agency that attempts to maximize utility under certain conditions.