Voice-based virtual assistants design and European legislation: the interpretation of subliminality, manipulation and deception

This position paper, which is a work in progress of my PhD research, highlights the terminological unclarity of article 5(1)(a) of the AI Act regarding the terms subliminal, manipulative, and deceptive techniques, which the article itself prohibits. Since it is unclear how these concepts should be interpreted, it is difficult to determine when AI systems should be considered subliminal, manipulative, or deceptive. In order to prevent deployers and providers from developing, deploying, or commercializing such systems, it is necessary to specify the meaning of these expressions. To do so, the European regulatory framework on digital services (DSA), data (Data Act), and consumer protection will be analyzed, both at the European level (Unfair Commercial Practices Directive) and Italian level (Legislative Decree No. 145/2007 and Legislative Decree No. 146/2007). The context in which this analysis will be conducted is that of voice-based virtual assistants, in order to understand whether and how these increasingly used software can deceive or manipulate consumers.