Trading off accuracy and explainability in AI decision-making: findings from 2 citizens’ juries