Marija Slavkovik
Professor, Department of Information Science and Media Studies
University of Bergen
www4.uib.no/en/find-employees/marija.slavkovik
Short bio
Marija Slavkovik is a Professor in Artificial Intelligence with the Faculty for Social Sciences of the University of Bergen. She is a computer scientists that has been doing research in machine ethics since 2012. Machine ethics studies how moral reasoning can or should be automated. Marija works on formalising ethical collective decision-making. In her work she pulls heavily in her prior research experience in computational social choice. She has held held several seminars, tutorials and graduate courses on AI ethics and has a considerable media presence in Norway as a AI science disseminator. She is a vice chair of the European Association for Artificial Intelligence, and Society track editor of JAIR.
Talk Title: “Good machines must be able to say no - towards normative governance for AI agents”
Imagine delegating to a machine all the work you do not want to do. The great breakthrough of post ChatGPT AI is perhaps in allowing us the ability to delegate tasks to computers in natural language.The level of autonomy of agents describes how much they can do without human intervention. It is an instrument to their usefulness. Autonomy capacity makes the difference between delegation and instruction. The less autonomy an agent has the more detailed and unambiguous the instruction needs to be. Programming classical computers can be seen as very detailed unambiguous instruction. The current state of the art in AI engineering brings us closer to the dream of work delegation, but we are not quite there yet. In this talk we consider that delegation requires competence to not only determine how a task is to be executed, but also which lines should not be crossed. When you ask your partner to book a table for dinner, you do not typically consider that they will hack the restaurant booking system to do it. But AI agents have done this. At the same time, and perhaps counterintuitively, in addition to autonomy we need agents to be able to disobey, not comply, break the rules in order to uphold the values. This talk is not about LLMs but about normative multi-agent systems with consideration for non-compliant robots and pro-social rule breaking agents.