The obscure politics of artificial intelligence: a Marxian socio-technical critique of the AI alignment problem thesis

Abstract

There is a growing feeling that artificial intelligence (AI) is getting out of control. Many AI experts worldwide stress that great care must be taken on the so-called alignment problem, broadly understood as the challenge of developing AIs whose actions are in line with human values and goals. The story goes that ever more powerful AI systems are escaping human control and might soon operate in a manner that is no longer guided by human purposes. This is what we call the AI-out-of-control discourse which, in this paper, we critically examine and debunk. Drawing on complementary insights from political theory, socio-technical studies and Marxian political economy, we critique the supposed animistic and autonomous nature of AI, and the myth of the uncontrollability of AI. The problem is not that humanity has lost control over AI, but that only a minority of powerful stakeholders are controlling its creation and diffusion, through politically undemocratic processes of decision-making. In these terms, we reframe the alignment problem thesis with an emphasis on citizen engagement and public political participation. We shed light on the existing politics of AI and contemplate alternative political expressions whereby citizens steer AI development or stop it in the first place.