From Robin Hanson

Imagine a billionaire kidnapped a random person and said: 

Here is a camera, banana, bicycle, sundress, and porcupine; you have one day to make a funny video using them all. I’ll upload it to YouTube, and if it gets “enough” views in the first day, I’ll give you $10B. If not, I’ll kill you and your closest twenty relatives. No, I’m not going to tell you my “enough” threshold. Go.

Imagine a billionaire kidnapped a random person and said: 

Here is a camera, banana, bicycle, sundress, and porcupine; you have one day to make a funny video using them all. I’ll upload it to YouTube, and if it gets “enough” views in the first day, I’ll give you $10B. If not, I’ll kill you and your closest twenty relatives. No, I’m not going to tell you my “enough” threshold. Go.

This random person would now have a very strong incentive to produce this video. And yet would most likely fail. Maybe even worse than if you’d just offered them a $1000 reward. The billionaire would be much better off hiring experienced video professions chosen from a competitive market. The lesson: incentives by themselves, even strong ones, can be quite insufficient. 

Consider another example: political sortition. A big problem with democracy is that each voter has on only a tiny chance to be pivotal, and thus has very little direct incentive to decide well. A possible solution to this is to instead pick random juries of voters to decide things. But alas, while I’d like to see this tried, I’m not very optimistic about it. At least at first, I doubt random juries would pick much different from the usual voters. Though maybe after years of experience, society could learn to use them. 

For a third example, consider health insurance indemnities, also known as “critical illness insurance”. Here you get paid cash upon a medical diagnosis of a serious illness, cash that you can either keep or spend on a matching medical treatment, at an insurer-negotiated price. This arrangement is designed to avoid the problem that regular “in-kind” insurance can induce patients to accept to-them-free treatments that are beneficial, but not worth their cost.

Some say indemnities won’t work as docs would take bribes to over-diagnose, but this is already a common form of insurance, and corrupt over-diagnosis is just as serious a problem in the usual case, where the docs who diagnose you are also the docs paid to treat you. Instead, the main problem seems to be that, at least at first, the cash option doesn’t much influence patient treatment choices.

This is suggested by a 2020 study reporting on  

two blinded vignette-based survey experiments with 3,000 respondents, randomized to eight clinical vignettes and three insurance types. … Most or all of the spending due to insurance would occur even under an indemnity.

While patients chose higher-value treatments more often, and accepted treatments more when they were insured, indemnity versus in-kind insurance didn’t make a significant difference to which treatments patients chose. At least in a survey, few pick money instead of the usual treatments. Yes, maybe people would act different in real life, and maybe a society where indemnity was common would learn to choose cash more often. But at least at first, merely giving patients better incentives re once-in-a-lifetime events doesn’t change much.

As a fourth example, imagine that we tried to predict future events via polls that were strongly incentivized via proper scoring rules, but polls which we asked of everyone and where respondents had to immediately answer, and weren’t warned so they could prepare. While such forecasts might be a bit better than un-incentivized but identical polls, they’d be far worse than well-subsidized prediction markets. In such markets, people self-select to become traders, and they can (a) form org teams, (b) take their time to consider answers, and (c) pick what and when they forecast. Such markets work even better when they repeatedly answer similar questions over many years, resulting in the worse teams losing and going away, while the best learn how to win. 

The general lesson here is that instead of just wanting people who serve us to have strong incentives, we more want something like capitalism. That is, it is not enough to “pay for results”, i.e., for the people who sit in roles doing key tasks to get more of what they want if they produce better results. Even getting paid a lot more just isn’t that helpful. We more want self-selection into competitions for sitting in such roles, with orgs and not just individuals able to play. Including for-profit orgs. When many such entities compete under substantial incentives over long periods to sit in similar roles doing similar tasks, the world can learn how to do better. 

For many of our most important tasks, peak performance in our civilization has long looked like this: for-profit orgs sitting in roles where they are paid more cash when they produce better results for customers. Investors compete to find and create good orgs, orgs that are sufficiently free from regulation to explore large spaces of possible products, services, and methods for making and marketing them. Employees and customers have access to sufficiently accurate markers of the costs and gains of alternate offerings to be able to, on average, roughly choose the better options. All of these agents should have substantial incentives, but incentives by themselves are just not enough. 

For example, my proposed political reform is futarchy, wherein speculators are paid for improving our consensus predictions of the consequences of proposed policies, proposal-consideration time-slots are auctioned, and proposers are paid a cut of market-estimated proposal value. Anyone with sufficient capital can make teams to self-select to compete in these roles of proposers and speculators. Voters need only attend to which election candidates support the outcomes (not policies) they want. 

My proposed medical reform is to merge health insurance with life, disability, and pain insurance, and have hands-tied third-parties pay extra premiums for extra benefits. Auctions can select the orgs willing to offer fixed insurance packages for the lowest premiums. Any org with sufficient capital can offer such insurance services. Patients need only judge their relative values for cash, life, disability, and pain. 

My proposed criminal law reform is bounty hunters plus requiring everyone to get a “voucher” who fully insures their crime liability, and who is given great freedom of contract to negotiate client contracts that specify co-liabilty, limits on privacy and movement, and punishments given crime convictions. Sufficient time can be the only requirement to be a bounty hunter, and sufficient capital the only requirement to be a voucher. Clients need only judge their values for cash, privacy, movement, and punishment. 

I hope you get the idea here. Yes, stronger incentives are better than weaker ones, but merely cranking up incentives is often insufficient. We also want a large world of orgs with sufficient capital and regulatory freedom who can self-select into key incentivized competitions. Then over long periods of repeatedly searching for ways to do a similar stream of tasks, orgs can learn to do better, with those who do worse being selected away. To make this all work, ordinary folks need only have a clear enough view of the cost and value of the services they get from such orgs. This, I propose, is the path to service-utopia. 

So, is there a better name for what I’m pointing to here besides “capitalism”?