Cassandras voicing about the perils of AI to democracy abound. It exists though a real danger that technology stiffens the increasing gap between controllers and controlled. Black box AI algorithms owned by the controller class can shape public perception reinforcing the asymmetry of power. There is furthermore a problem of misaligned incentives, political actors have a strong incentive of using machine learning to win elections, which is no match with the keenness to use this same technology to efficiently learn about their constituents’ problems. And still, AI represents a unique opportunity for citizens’ empowerment. Connectivity, transparency, and accountability are capabilities supported by design in the digital realm. AI will both strengthen and weaken democratic institutions. History teaches us that more important than innovation is what we decide to do with it. Will democracy by 2030 be outwardly thriving (elections keep taking place) but inwardly failed (votes being shaped by influencers and platform owners)? The answer will depend on personal responsibility and social institutions that protect a shared sense of community and the common good. Debating to be for or against AI is at this point futile, AI is here to stay, and artificial agents are already taking decisions for us. Contrary to the popular belief of one-fits-all algorithms that would oversimplify personal attributes, machine learning algorithms require randomness, the same randomness on which democracy may thrive. Too little or too much randomness is a marker of instability in complex systems, and a democratic use of AI represents an opportunity to reboot democracy.