DIMACS Special Focus on Mechanisms and Algorithms to Augment Human Decision Making

The convergence of humans and machines-through machine learning and increasingly intelligent (and autonomous) devices-promises a transformational impact on daily life. Already machines may outperform humans at certain tasks. Nevertheless, complete autonomy remains elusive, and the ability predict future outcomes will continue to benefit from human wisdom for many more years. Beyond accurate predictions, decisions require insight into the preferences or utility functions of the people that they impact. Decision-support tools must go beyond observational data to actively elicit both information and preferences from stakeholders, to reward contributors appropriately, and to combine the inputs in a way that is optimal modulo fairness constraints and practical limits on computational power. A "special focus" (SF) will study tools to augment decision making in individuals and organizations, aiming to vastly improve decision-support systems by leveraging both human and machine intelligence. Tools to be explored as part of this project include (1) mechanisms to elicit complex probabilities and preferences from people, rewarding them appropriately (2) algorithms to combine human judgments and data-driven predictions, and (3) algorithms to aggregate potentially conflicting preferences. The SF will feature workshops exploring ongoing research; a visitor program allowing individuals to dig deeper into issues that arise; and implementation challenges that test out new algorithmic approaches. The project will involve a large number of people in various scientific communities and expose them to new ideas, new problems, and new opportunities for collaboration. These activities encourage diversity of opinions, experiences, ideas, and expertise, while also advancing the goal of involving more women and under-represented minorities in computing. Undergraduates will be exposed to topics of the SF through research experiences each summer. The workshops will explore the nature of work and how it will change in the age of intelligent machines. Presentations will also address current research that has global societal importance in areas such as voting rules, restrictions on autonomous vehicles, and fairness of algorithms used to make decisions in the medical, economic, and other domains. The SF will build on recent advances in computational social choice, crowd-sourced democracy, and crowd-sourced forecasting, including prediction markets and scoring rules. It will investigate how organizations can reduce barriers by rewarding people to enhance inputs to machine intelligence, thus improving predictions and decisions, and it will study how to elicit accurate representation of individual preferences and algorithms to turn elicited votes into organization-level decisions that maximize an objective subject to fundamental axioms and efficient computation. It will emphasize important themes in computer science theory such as complexity and machine learning, as well as issues of social responsibility, fairness of algorithms, aids to decision making, and human-computer interactions that arise in a world with increasingly intelligent machines and ubiquitous data. The SF will explore new ideas for eliciting complex information, including new ideas in privacy-preserving elicitation, market-based elicitation, wagering mechanisms and scoring rules. It will address new directions in algorithmic social choice involving strategy-proof auctions, incentive compatible machine learning, randomized social choice, and iterative voting. There will be discussion of mechanisms and markets to elicit information beyond labels (e.g., features) to improve machine learning models. The SF will address questions of preference aggregation, including trustworthy preference aggregation, aggregation of conditional preferences, and complex rating procedures. Finally, the SF will explore issues of learning from real data and will consider elicitation when one has limited time, bounding parameter values from qualitative and quantitative information about probabilities, and leveraging existing knowledge to inform the elicitation process. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.