Björn Lundgren

The current privacy regime depends on self-management through informed consent, which has been broadly criticized (see, e.g. Lundgren 2020c; Nissenbaum 2011; Solove 2013). For example, according to...

In the literature, there is a concern that if AI and robotic systems become super-intelligent, we will lose control over those systems. The loss of control, some worry, may yield bad or even catastrophic outcomes. Some think that we can avoid this problem even in the absence of control, simply by ensuring that the systems are aligned with our human values (what is known as “value alignment”). Here it is argued that value alignment will not succeed without sufficient control.

So-called “deepfakes” (i.e., highly believable but fabricated media) are infamous for their potential political application. However, they can also contain false information about individuals, which raises the question whether deepfakes can violate a moral right to privacy. This question is directly related to the often ignored but still highly contentious issue of whether the spreading or use of false or fake information can violate a moral right to privacy. While such queries normally turn on how we conceptualize a moral right to privacy, they are answered here based on premises that any conception of the moral right to privacy must adhere to.

Social AI systems have the potential to outperform humans in certain ways—raising the question of whether we should have AI systems (including robots) replace humans in the labor market (i.e., not only in non-social contexts). This article will first put forward three sets of pro tanto reasons why social AI systems should not be implemented, then it will consider three possible situations in which these reasons may be overridden, arguing that the use of social AI is only justified in a very circumscribed set of cases. Moreover, in situations in which replacement is justified, the AI systems’ social and informational capabilities should be limited. And finally, where social AI systems are utilized, they should be made to comply with certain social codes.

The AI Act advances a risk-based approach to the legal regulation of AI systems in the European Union. While we support this development, we argue that adequate AI governance requires paying attention to the broader implications of AI systems on the socio-technical landscape in which they are designed, developed, and used. In addition to risk-based impact assessments, this involves coming to terms with the socially disruptive implications of AI, which should be governed and guided in a dynamic ecosystem of regulation, law, ethics, and evolving human practice. In this paper, we outline a ‘social disruptiveness-based’ approach to AI governance aimed at addressing disruptions by AI that are not easily captured by legal regulation, but that are nonetheless of great societal and ethical concern. We argue that integrating the AI Act risk-based approach with a social disruptiveness-based approach can offer a more nuanced understanding of the dimensions of impact of AI systems on society at large, thus enhancing the governance of AI and other socially disruptive technologies.