This is the final post in a series on "AI Benefits." It is cross-posted from my personal blog. For other entries in this series, navigate to the AI Benefits Blog Series Index page.

This post is also discussed on LessWrong.

For comments on this series, I am thankful to Katya Klinova, Max Ghenis, Avital Balwit, Joel Becker, Anton Korinek, and others. Errors are my own.

If you are an expert in a relevant area and would like to help me further explore this topic, please contact me.

AI Benefits Post 5: Outstanding Questions on Governing Benefits

Ideal Consultation and Delegation

Most AI Benefactors probably lack important expertise in creating and distributing AI Benefits.^[1] Thus, to distribute Benefits optimally, they will probably want to enlist the help of or hire external experts to make Benefits decisions.^[2] For simplicity, I will call such a body a “Benefits Group.”

A Benefits Group arrangement can vary in many ways. One obvious choice is whether the Group is internal or external to the Benefits-producing firm. In economics, similar decisions are analyzed in the “theory of the firm.” One can think of various considerations for and against internalizing Benefits decisions. An internal Benefits Group probably makes more sense when the Benefactor organization will be engaged in many, prolonged Benefits decisions and wants their Benefits decisionmaking to advance organization-specific goals. External Groups, on the other hand, will make more sense with one-off Benefits decisions or when an organization’s Benefits strategy is sufficiently generic as to be manageable by a Group with multiple Benefactor-clients. External—and therefore more independent—Groups could also be ethically important as a measure to improve Benefits decisions, since they would hopefully be less susceptible to making Benefits decisions in the firm’s best interests (rather than society’s). This, of course, is an example of a principal–agent problem.

A Benefits Group can have varying degrees of bindingness: at one extreme, the Group has complete control over the Benefits process; at the other pole, the Group could merely provide nonbinding advice. Intermediate arrangements could be giving the Group one of many votes on Benefits decisions or making the Group’s recommendations binding unless overridden.

Group composition is also important. An ideal Group will probably contain (or at minimum, have access to) experts in the various areas identified throughout this post. If, as discussed above, some distribution decisions will be made at the national level, national representatives or knowledgeable individuals will also be important.

The Group’s mandate should generally be maximizing benefits as defined above, but work may be needed to specify that goal more precisely than I have done here. Additionally, Group accountability to ensure fidelity to its mandate will be important. Accountability for philanthropic organizations has traditionally been quite weak and dependent on enforcement by state attorneys general. Experimentation with novel philanthropic accountability mechanisms might be appropriate.

Many of these parameters could be varied as conditions change. For example, it seems intuitively appropriate that a Benefactor with a larger pool of resources should have stronger external oversight than a small-scale Benefactor. It might also make sense for a Benefactor to devolve increasing amounts of decisionmaking authority to a Group over time to allow for initial unbound experimentation.

I expect that answering this question will require insights from economics, political science, law, and management studies.

Democratic Input

Some Benefits decisions should be subject to explicit democratic input. However, this is a nontrivial task.

First, as with delegation decisions, Benefactors must decide when to democratize Benefits decisions. Similar considerations will be at play here.

There is also the question of how to democratize. If democratization entails delegating decisions to a democratically elected body, then democratization is just a special case of delegation.

Since Benefits work is egalitarian and universal, if Benefits decisions are simply delegated, all people should have an input into the Benefits process. However, there is currently no such global democratic mechanism. Global governance institutions like the United Nations are far from egalitarian, as state votes are not proportional to population and some states have more influence than others. Deferring to the government of the nation in which the Benefactor resides is unlikely to be satisfactory since national governments spend the overwhelming majority of their resources domestically.

Where national or subnational governments are reasonably democratic, those governments might reasonably serve as representatives for their constituents in a Benefits allocation process. However, many governments are imperfectly democratic or entirely nondemocratic. The citizens of those nations still deserve consideration, but deferring to their government to speak on their behalf seems unsound.

Democratization need not be based on delegation. A Benefactor could in theory construct its own democratic input mechanism, such as through online polling. However, internet access, identity verification, and language issues limit the viability of this approach. Workarounds like randomly selected focus groups or random ballots could partially overcome these problems. “Private” democratization like this also raises questions of political legitimacy: namely, is it politically legitimate for a private Benefactor to construct its own democratic accountability mechanisms?

A final approach would be to choose representatives from populations who do not have existing democratic representatives and for whom private democratization is impossible or infeasible. While such chosen representatives can help outsiders gain insight into a population, they cannot fully replace democratically selected representatives.

I expect that forming a good democratization plan will require insights from political philosophy, political science, public choice, law, and international relations.

Evaluations

Various actors will want to evaluate Benefits. Benefactors themselves will hopefully want to know that they are doing good work. Evaluations could also be useful in monitoring and selecting between different possible delegates. Finally, society as a whole has a right to verify that Benefactors are actually benefitting society.^[3] Benefactors should probably update their Benefits plans in light of lessons learned from evaluations.

Designing good evaluations procedures will probably require insights from empirical economics, statistics, law, and public policy.

External and Internal Governance

Benefactors will probably want to remain attentive to the ways that external governance—i.e., governance of the communities in which recipients reside—could affect their plans. Insecure property rights, poorly functioning markets, corruption, and poor infrastructure could all make it difficult for beneficiaries to turn whatever Benefits they receive into sustained improvements in quality of life.

At the same time, large-scale Benefactors might inevitably have to construct their own internal governance processes. As the American federal government (and particularly the welfare state) expanded, so too did its internal bureaucracy for governing the distribution of those benefits. A large-scale AI Benefactor might have to do the same and take on governance tasks like assessing qualification for Benefits, resolving disputes about Benefits, verifying the identity of possible beneficiaries, and providing technical support to beneficiaries. The ability to build scalable internal governance of Benefits distribution could be a major determinant of success for Benefactors.

Good Benefits governance will probably require insights from law, political science, and development economics.

Expert Input

If you are an expert in one of the identified fields and are interested in contributing to this project, I would love to hear from you!