Machine Ethics and the Architecture of Virtue

As Consequentialism and Deontology ruled the domain of ethics for the past centuries, these theories were in the pole position for robotic implementation of a moral theory. Though Anscombe, in 1958, did revive Virtue Ethics, bringing it to the battlefront to compete with the other two theories, programmers did not pay much attention to it, and those who did found strong obstacles to its implementation. On the contrary, Consequentialism and Deontology were relatively easy to program, thus ignoring Virtue Ethics as a valid option for a moral theory of robotics. However, half a century of research in robotics has shown that neither Consequentialism or Deontology seem to be up for the task, failing in real world moral dilemmas and day-to-day life with human beings. Based on the advances in robotics, which include neural networks, machine learning and robots with cognitive abilities, this paper presents Virtue Ethics as a better option for implementing moral behaviour on robots. Besides an effort to transpose Aristotelean concepts to robotics, and the proposal of a hybrid (between top-down and bottom-up approaches) architecture that could serve as a model for decision-making based on Virtue Ethics, the main goal is to spark the debate on how Virtue Ethics may be used in robotics.