Panelot's goal is to facilitate the selection of citizens' panels, which bring together randomly selected people from all walks of life to discuss key questions and deliver policy recommendations. These panels should be representative of the population in terms of features like gender, age, ethnicity, and education. At the same time, Panelot aims to select a panel in a way that gives all volunteers a fair chance to participate.
Panelot receives as input the composition of the pool of volunteers, the desired panel size, and quotas meant to ensure a representative outcome (for example, in a panel of 100 people the number of women might be required to be between 47 and 53). Panelot outputs a list of quota-compliant panels, each with an assigned probability. The list of panels and their probabilities are chosen to be leximin optimal: the output first maximizes the selection probability of any volunteer, then maximizes the second lowest selection probability, then maximizes the third lowest, and so on. Since the number of quota-compliant panels is astronomically large, this optimization problem is extremely difficult and requires finding the right panels to make selection probabilities fair. Panelot solves it through state-of-the-art optimization techniques. You can learn more about this algorithm in our paper.