Guglielmo Tamburrini (PhD 1987, Columbia University) is Philosophy of Science and Technology Professor at Universita’ di Napoli Federico II, Italy. His main research interests concern the methodology of robotics, AI and the cognitive neurosciences, in addition to Ethical, legal, socio-economic (ELSE) issues arising in the context of human-computer and human-robot interactions. Acted as coordinator of the first EC project on the ethics of robotics (CA ETHICBOTS, 2005-2008, VI FP). Visiting Scholar in 2009-10 at ZIF (Zentrum für Interdisziplinäre Forschung, Universität Bielefeld, Germany), in 2014 he was awarded the Giulio Preti International Prize by the Regional Parliament of Tuscany (Italy) for his research and teaching activities on ethical and social implications of ICT and robotic technologies.
Open-ended learning robots: the interplay of epistemic and ethical issues
Distinctive ethical issues concerning learning robots arise from familiar limitations affecting our capability to predict exactly and explain their behavior. It is well-known that these epistemic limitations bear on the ethical problem of fairly ascribing retrospective responsibilities (for harmful behaviors of learning robots) as well as prospective responsibilities (for the decision to field some learning robots). By a comparative analysis of various machine learning methods – notably including non-incremental, standard on-line/incremental and open-ended learning – it is argued here that the epistemic predicaments generally affecting human observers of learning robots are significantly enhanced in open-ended learning. Indeed, in this latter case one has to develop predictive and explanatory tools without knowing a priori which categories and behaviors the robot has been continuously learning in its open-ended learning environment. Importantly, to tackle these difficulties we suggest some viable strategies in the context of neural network approaches, based on similarity measures to identify bounds on the input-output response distance. Moreover, this epistemological analysis is brought to bear on robot ethics, insofar as it enables one to identify in a principled way tasks that would be ethically problematic to let open-ended learning robots to perform.