Algorithmic Fairness

Recently, new constraints such as the transparency, fairness and interpretability have been imposed on machine learning (ML) algorithms and these constraints will likely have a significant impact on the future algorithm design. Since ML algorithms are being increasingly used as an important aid for decision-makers, these new constraints will also have a significant impact on decision-making process. We propose to develop models for decision-making that will utilize these new constraints and adapt them to complex socio-technical systems. Specifically, we plan to

• (A) design algorithms that will explicitly include the monotonicity constraint, and to
• (B) develop algorithms for multi-agent information aggregation. To evaluate the quality of our models, we will utilize new theoretical insights in the areas of problem solving, expert identification and optimal team design. The project objectives consist of the following tasks:
  • (1) Modeling utilities of human decision-makers in multi-agent settings;
  • (2) Decomposing the bias structure in decision-making algorithms;
  • (3) Identifying experts and optimal compositions of teams based on human and algorithm voting;
  • (4) Participatory budgeting based on human and algorithm voting.