Institutions

AI allows—and may require—different ways of organizing society. Schools are already reimagining how they spend time, and how they assess students. But change will come to other sectors as well. We will need not only technical innovation, but also social innovation developing the new institutions that will govern how we interact with each other, and with machines.

Questions in this area

What new institutions are needed for a world of machine intelligence?

What we're asking

What new institutions would help society benefit more from AI? Can these be provided by existing organizations or will new organizations be needed? We are both analyzing and building.

How can we align automated systems with human values?

What we've found

Machine learning algorithms commonly face competing objectives. We have worked on tools to allow algorithms to trade off multiple objective functions. Theoretical work suggests how predictive algorithms may trade off private and public objectives when predictions are noisy. We also have audited PROGRESA, Mexico's landmark cash transfer program to reveal its implicit priorities on different types of people and different outcomes.

What we're asking

Who, and which outcomes, should automated systems prioritize? How should the public inform these decisions? How can we audit whether a system implements them?

All Institutions research →