Social Environment Design (SED) is a theoretical framework for the use of AI for automated policy-making. At the same time, SED is designed to be bootstrapped by making piecemeal advancements that inform future work and have positive social impact. The SED Contributor Framework is our approach to organize and solicit projects that make those piecemeal advancements. Here we present the SED Contributor Framework, which outlines the domains of work to be done, how you can contribute through your own projects, and with specific examples and an ongoing request for proposals. This is a living document and wiki of sorts that we will update as progress is made and that we intend for others in this space to enrich with their own knowledge.
The SED Contributor Framework parallels the theoretical framework, including three major iterative steps:
In the sections below, we go into more detail on each of the major steps, their minor constituents, and ways to contribute to each.
The first major need in SED is to figure out what we are trying to achieve. In the long-term, SED promises to optimize global social welfare - to make sure that all people and the environment on which we depend are healthy, happy, and sustainable. Until then, however, advancements will necessarily focus on more narrow measures of social impact and for relatively small groups. The challenge then is to learn what we care about as individuals and how that is aggregated to a group-level objective.
The World State gives context to participants, agents, and the Principal to understand what is at stake and how it can be changed. The specific format of the World State will depend on the scenario being optimized and the technology used to do the optimization. In early iterations of SED, this will most likely be a natural language description of the social environment in which agents are acting and the states of each agent. Over time the World State could evolve to be more mathematically expressive and therefore more conducive to optimization.
Human participation is vital to make sure solutions to SED scenarios are relevant and that humanity continues to be in control of its own future. Participation may come in many forms and at various points in the SED process. For example, it will be especially important early on to have beneficiaries of SED solutions build a database of human values and human behavior through value elicitation, voting, and intervention in simulation. As we amass accurate models of humans, it may be possible for new participants to be less intensively surveyed, since we will be able to interpolate their behavior from existing data.
There will be three primary forms of human input to SED