Society Protect: Developing A Voting Mechanism for Blockchain-based AI Alignment

This paper proposes a novel anonymous voting mechanism for Artificial Intelligence (AI) alignment rules and conducts an associated simulation. The aim is to provide a voting mechanism that is protected from malicious agents to ensure that trustworthy humans remain in control of AI and have a say in its regulation. The discussed mechanism enables a majority of uncoordinated and trustworthy human agents to agree on a legitimate alternative to an AI rule poll, while outvoting a malicious alternative that is chosen by a smaller but coordinated set of malicious agents. For this purpose, the use of blockchain technology in combination with a novel voting mechanism is proposed and voting outcomes are simulated. The proposed mechanism and well-considered choices of human agents justified by social identity theory help to minimize the impact of malicious agents on the system. The simulation results in this study, building upon theoretical concepts derived from the literature, enable well-founded assumptions regarding the system’s robustness.