Normative Metacognition in AI by Vincent Mueller
Join us online for this talk by Professor Vincent Mueller, Humboldt Professor at Friedrich-Alexander Universität Erlangen-Nürnberg, Philosophy and Ethics of AI Director, Centre for Philosophy and AI Research.
This talk will not be recorded – be sure to join live!
Webinar ID: 160 295 8342
Classic theory of rational choice claims that a rational agent should choose the action with the maximal expected utility (MEU). However, actual agents, be they natural or artificial, have to deal with limited cognitive resources for finding what that action is (processing time, storage, available information). In addition to rational choice given particular goals, humans agents also have the ability for normative reflection, in particular to reflect on their goals, on what should be MEU. Furthermore, they have the ability for metacognition, i.e. to watch their own cognitive processes and decide to change them, e.g. to bring more resources to work, to stop optimising the decision, or to start reflecting on the value of the goals they had assumed thus far – I call this latter ability “normative metacognition”. My talk is about the role of normative metacognition in human cognition and in AI, including the relation to frame problems, to “value alignment” and to what constitutes “having a goal”.
Dr. Mueller’s talk is part of a speaker series co-presented by the Johns Hopkins Institute for Assured Autonomy (IAA) and the Berman Institute of Bioethics, featuring national and international scholars presenting new research and development at the intersection of autonomy, ethics, and governance.