PhD thesis defence by Yeliz Doker
The unpredictable, dynamic, and potentially limitless nature of Artificial General Intelligence (AGI) characterises it as a 'wicked problem' that fundamentally resists traditional, static governance approaches. The core argument of this thesis is that effective AGI governance must therefore rely on a hybrid architecture that combines mandatory external legal controls with complementary, internal, learning-based ethical reasoning.
The thesis consists of three interconnected parts. Part I establishes the conceptual foundation by defining AGI through five specific Properties and diagnosing its wicked nature by applying ten characteristics of Rittel and Webber's wicked problems theory. This demonstrates that addressing the challenge of AGI cannot be accomplished with a 'business as usual' approach. Part II, in turn, critically evaluates the European Union Artificial Intelligence Act against this wicked problem framework. While the Regulation provides fundamental constraints for Artificial Narrow Intelligence (ANI), its inherent logic of treating AI primarily as a 'product' is structurally strained due to AGI's 'unknown unknowns' nature and the radical uncertainties it would bring if realised.
To fill this governance gap, Part III explicitly reframes the challenge from 'taming the product' to 'guiding the learner,' establishing a new architecture for internal ethical learning. This part proposes the WREL model, a novel hybrid approach to machine ethics that integrates David Kolb's Experiential Learning (EL) for the bottom-up development of ethical judgments with John Rawls's Wide Reflective Equilibrium (WRE) for top-down normative constraint and justification. The WREL model is not designed to replace law, but rather as a necessary second layer of governance intended to enable AGI systems to responsibly develop their ethical reasoning in dynamic environments. The thesis concludes that a resilient AGI governance strategy must develop internal systems capable of producing justified ethical judgments under uncertainty, complementing external legal oversight.
Register