Kant and the Designation of AI as Autonomous or Heteronomous

I propose a paper exploring how Kant’s philosophy might advise us about designating current-day 'artificial intelligence' machinery as “autonomous” or even "heteronomous.” This is important, since common parlance has it that AI machines are autonomous, and philosophers of technology often affirm that, e.g., AI-guided weapons have at least limited autonomy. However, when Kant, in the "Groundwork of the Metaphysics of Morals," reflects on how rational beings can set ends for themselves, he determines that such beings display either autonomy or heteronomy according to whether they a) determine their ends according to hypothetical imperatives of the form “if I am to reach X, I must do Y,” or b) determine their ends according to the categorical imperative, which is not conditioned but absolute, and so demands that maxims can be universally actualized and thus be like “laws of nature” for a practical community. Conceding there may be difficulties with this notion (could we not universalize a bad practice, and should we not first find autonomy in reflection on theoretical cognition, as Kantians like Salomon Maimon, Hans Wagner, and Werner Flach have urged?), it seems clear that, in any case, we should not deem AI machines to be autonomous but heteronomous, since they operate exclusively by hypothetical imperatives. However, I then want to explore the perhaps more difficult question of whether AI machines, by Kant’s lights, deserve even the designation “heteronomous.” For we have had the long debate about whether, for Kant, the choice to pursue a hypothetical imperative is also a free choice; and AI machines do not seem to freely choose between categorical and hypothetical imperatives. Perhaps a good solution is that AI machines are heteronomous in a derivative sense: not being end-setting beings themselves, they are tools of end-setting beings, in the Aristotelian sense of “tool” or “machine” elucidated by, e.g., Hans Jonas.