Alignment in natural language | Marcin Woźniak | EAGxWarsaw 2023

The only non-violent and real-world functioning way to achieve alignment is through discourse. This talk presents a proven model of how a group of people can achieve decisions that have value alignment characteristics using technologically supported discourse and collective intelligence. Lessons learned from the practice of more than 40 projects where such a method has been used in decision-making and public policy-making are discussed. A practical model for the emergence of collectively normative standards through open structured argumentation is also presented. Validity for overcoming multipolar traps and defeating Moloch with small improvements in the decision-making process is demonstrated. Finally, a scaling of this solution to AI-alignment with collective reasoning-based reinforcement learning is proposed.
Find out more about EA Global conferences at: www.eaglobal.org
Learn more about effective altruism at: www.effectivealtruism.org