A framework for expected capability sets

Overview

• The paper addresses decision-making problems with multiple objectives and uncertain states of the world.
• Inspired by the capability approach, the focus is on cases where a policymaker chooses an action that leads to a set of choices for citizens, even when no preferential information is available to weigh the criteria.
• The paper proposes two procedures to effectively support decision-making in this context, by merging the potential set of choices for each state of the world and accounting for their respective likelihoods.
• The proposed procedures satisfy several fundamental and desirable properties.

Plain English Explanation

The paper addresses complex decision-making problems where there are multiple goals to consider and uncertainty about the future state of the world. Imagine a policymaker who needs to choose an action that will then give citizens a set of options to choose from. Even if the policymaker doesn't have clear information about how important the different goals are, they can still make a decision by considering the likelihood of each possible future state.

The paper introduces two new methods to help the policymaker in this situation. These methods combine the set of options available to citizens for each possible future state, taking into account how likely each state is to occur. The end result is a recommended action that satisfies several important properties, like ensuring the final set of options is well-balanced.

Technical Explanation

The paper is inspired by the capability approach to decision-making, which focuses on the choices available to individuals rather than just the outcomes. In the context studied here, a policymaker must choose an "act" that, when combined with a "state of the world," leads to a set of choices for citizens.

The key challenge is that the policymaker lacks preferential information to weigh the importance of the different criteria. However, they are able to obtain likelihoods for the different possible states of the world. To support decision-making in this context, the paper proposes two procedures that merge the potential set of choices for each state, taking into account their respective likelihoods.

The first procedure involves aggregating the sets of choices for each state using a "union" operation, weighted by the state likelihoods. The second procedure uses a "convex hull" operation to create a balanced set of choices that covers the range of possibilities across states.

Both of these procedures are shown to satisfy several fundamental properties, such as ensuring the final set of choices is non-dominated (i.e., no option is clearly superior to another) and that the likelihood-weighted average of the sets of choices for each state is contained within the final set.

Critical Analysis

The paper provides a thoughtful framework for supporting decision-making under uncertainty and multiple objectives, without requiring precise preferential information. The proposed procedures have appealing theoretical properties and could be a useful addition to the toolkit for ML-supported decision-making.

That said, the paper does not address some potential practical challenges. For example, it's not clear how the policymaker would obtain accurate likelihood estimates for the different states of the world, especially in complex real-world scenarios. There may also be challenges in effectively communicating the resulting set of choices to citizens in a way that enables meaningful decision-making.

Additionally, the paper does not consider the potential for strategic manipulation of the decision-making process by interested parties. Further research may be needed to understand how these procedures could be made more robust to strategic behavior.

Overall, the paper presents a promising decision-theoretic framework for supporting complex decision-making under uncertainty. However, additional work may be needed to address some of the practical and strategic challenges that could arise in real-world applications.

Conclusion

This paper introduces an innovative approach to decision-aiding in situations with multiple objectives and uncertain states of the world. By proposing two procedures that merge the potential choices for citizens while accounting for the likelihood of different states, the authors offer a framework to support policymakers in making well-balanced decisions even without precise preferential information.

While the theoretical properties of the proposed methods are compelling, there may be practical and strategic challenges to consider in applying this approach. Nonetheless, the paper contributes an important step towards expanding the toolkit for ML-supported decision-making, with potential applications in a wide range of policy and organizational contexts.