Official Data Income (DI) Collecting Social Norm Data- Toward Democratic AI Alignment -

Abstract

This paper discusses Data Income (DI) collecting social norm data for assisting democratic AI alignment. Data Income (DI) has been proposed to address the issues of "Data Bottleneck Hypothesis" and "Social Bottleneck Hypothesis" of Artificial General Intelligence (AGI). Data Income (DI) can be used for many purposes including collecting data for value alignment. This paper proposes that Data Income (DI) can be used for certification of social norm datasets made by AI researchers, companies, etc. by qualified voters in an entire country (e.g. Japan). This paper also proposes an intellectual property system regarding the representation of the certified datasets.