Invited Talk at AAAI’26 Workshop on Machine Ethics, titled Between Rules and Reasoning: Towards Machine Morality
Date:
Gave a 1h invited talk titled Between Rules and Reasoning: Towards Machine Morality at the AAAI’26 workshop Machine Ethics: from formal methods to emergent machine ethics.
Abstract: Is “learning” sufficient for machine morality, or are we teaching AI to mimic behavior without understanding the stakes? As AI enters increasingly sensitive social spheres, the tension between bottom-up learning and top-down formal constraints has become the central challenge of the field. This talk maps the landscape of AI morality along a continuum—from systems that infer ethics through data to those governed by rigid formal logic. By drawing parallels to the broader history of AI research, we will explore why neither extreme has succeeded in isolation. I will present a concrete framework for “hybrid morality,” demonstrating how formalization can be paired with reinforcement learning to create LLM agents that don’t just follow rules, but internalize moral goals. We conclude with a look at the next frontier, reviewing recent proposals to combine Generative AI with verifiable methods to ensure the systems of tomorrow are as safe as they are capable.