Philosophical literature on conceptual engineering has identified different kinds of ‘conceptual defects’. For instance, a defect concept may prevent the realization of moral and political values, or it may hinder the acquisition of knowledge and theoretical progress (Cappelen en Plunkett 2020). There are different ‘ameliorative strategies’ to respond to these defects, such as conceptual elimination, conceptual replacement, and conceptual modification. In this presentation I investigate whether ARTIFICIAL INTELLIGENCE can be characterized as a defective concept, and if so, what ameliorative strategy should be pursued in order to ‘fix’ it. I advance five reasons for thinking that the concept of AI is indeed defective: the concept is historically tainted (Cave (2020); it fosters a narrow conception of the good (Beerends en Aydin 2024); it contributes to the trivialization of important human capacities (ibid.); it helps to generate unrealistic expectations about AI (Crawford 2021); it facilitates conceptual appropriation (ibid.). I conclude by contextualizing the claim that AI is a defective concept and by discussing whether the concept of AI should be re-engineered, and if so, how.