Chronological Overview
The Kant-and-AI literature began with a narrow question: could Kant's moral philosophy be made operational for a machine? That question still gives the field its shape. @agent: sentence about the addition of questions from theoretical philosophy
The chronology below therefore reads less like a steady march toward "Kantian AI" than a series of shifts in what the problem is taken to be. The bibliography itself is ordered newest-first; this sketch moves in the other direction, from the first attempts to build Kantian machines toward the current debates over institutions, dignity, responsibility, and public reason.
2002–2011
The first Kantian machines
The earliest work is strikingly direct. It asks whether a machine could act on the categorical imperative, not merely cite it. Thomas Powers's Deontological Machine Ethics (2005) and Prospects for a Kantian Machine (2006) are central because they state both sides of the attraction. Kantian ethics seems computationally promising: it has rules, tests, and a language of consistency. At the same time, the machine would have to identify its maxim, understand the situation in which it acts, and distinguish genuine moral reasons from surface regularities.
Other early contributions widen the frame. Stuart and Dobbyn (2002) return to Kant's transcendental psychology as a condition for artificial conscious experience. Wallach and Allen (2008, 2009) make Kantian deontology the paradigm of a "top-down" approach to machine morality, while Tonkens (2009) formulates the sceptical challenge that has never quite gone away. By the time Anderson & Anderson's Machine Ethics appears in 2011, Kant is already one of the field's reference points — but also one of its difficulties.
2012–2018
Formalization and robot ethics
The next phase does not simply solve the early problems; it makes them more precise. Beavers (2012) warns that computable morality can flatten conscience into behavioural output, a risk he calls "ethical nihilism." The Lin, Abney & Bekey volume Robot Ethics (2012) brings Kantian questions into discussions of autonomy, responsibility, and the moral standing of robots. Purves, Jenkins & Strawser (2015) then press a particularly Kantian objection to autonomous weapons: even if such systems produced the right outcome, they could not act for the right reasons.
At the same time, the literature becomes more technical and legal. Bendel and colleagues sketch "Kant machines" in the context of truth-telling chatbots, and Ülgen (2017) brings dignity and autonomy of the will into AI and robotics law. Lindner & Bentzen's 2018 formalization of the Formula of Humanity is the clearest milestone here. It shows that "never merely as a means" can be modelled with deontic logic, but it also makes visible how much interpretation is needed before a Kantian formula becomes executable.
2019–2022
The field consolidates
By the early 2020s, Kant-and-AI is no longer one debate. It has at least two centres of gravity. One is theoretical: Richard Evans's Apperception Engine (2021; 2022) uses Kant's account of unified experience as inspiration for machine-learning architectures. Related work by Schlicht, Baiasu, and others treats Kant not as an ethicist of machines but as a thinker of synthesis, self-consciousness, and the limits of cognition.
The other centre is practical and normative. Kim, Hooker & Donaldson (2021) formalize deontological principles for value alignment; Hanna & Kazim (2021) build a dignitarian approach to AI ethics; and Singh (2022) implements Kantian ethics in Isabelle/HOL. Kim & Schönecker's edited volume Kant and Artificial Intelligence (2022) gathers these lines in one place. Its importance is not just that it is still the only book with that title. It also shows how scattered the Kantian questions have become: cognition, practical reason, robot ethics, autonomous driving, right, enhanced autonomy, and aesthetic judgment all now belong to the same conversation.
2023–2026
The post-ChatGPT turn
Large language models change the tone of the debate. Kantian AI is no longer only a question about whether a machine could be built to reason morally. It becomes a question about how AI systems address us, classify us, persuade us, govern us, and invite us to treat them as if they were social partners. Work on alignment by McDonald (2023), Mougan & Brand (2023), D'Alessandro (2024), and Chaly (2024) keeps the rule question alive, but it is now tied to safety, fairness, fallibility, and the design of institutions.
The same shift appears elsewhere. Sanwoolu (2025) separates the question "can AI be a Kantian moral agent?" from the more practical question of whether Kantian principles can guide AI design. Formalization continues, from Geng and colleagues' UPAR prompting framework (2023) to Olson's Formula-of-Universal-Law logic (2026). Meanwhile the responsibility debate matures around Kiener, Demirtas, Vallor & Vierkant, and MirzaeiGhazi & Stenseke; dignity and social AI come into focus through van der Rijt, Coelho Mollo & Vaassen (2026) and Shevlin (2024); and publicity becomes a way to ask whether algorithmic administration can be democratically legitimate. Vallor's The AI Mirror (2024), Floridi's principles-based counter-model (2023), and Yuk Hui's Kant Machine (2026) suggest why the field has broadened so quickly: AI has made Kant relevant not only as a moral theorist, but as a critic of reason, autonomy, and modernity itself.