The Unbiased Arbiter: Teaching Robots to Be Fair Moderators
Published:
Recently, I read “Superintelligence: Paths, Dangers, Strategies” by Nick Bostrom. It argues that superintelligence, if created, would be difficult to control and that it could take over the world to accomplish its goals. In this short article, I want to add my perspective to make superintelligence a collaborator for humans. In my view, we should train robots to become unbiased learners and moderators. My vision was rooted in the belief that just as genuine humans can become unbiased over time by embracing their true selves, robots could be taught to act as fair moderators by considering multiple perspectives before taking any action. I propose a multifaceted approach to achieve this, involving learning through conflict, embedding ethical guidelines, and transparency.
What do I mean by learning through conflict (LTC)? What motivates me? The basic idea of LTC is that when a human faces contradictory facts, they gradually become a judge and stand alone to make unbiased and independent decisions. We plan to enable an artificial intelligence (AI) agent to conduct LTC by training with contradictory data. First, the agent is inputted and trained with contradictory data comprising current data and a divergent perspective(s) of the data. Then, by adopting some strategies, early convergence to the data or its perspective(s) is prevented. The main advantage of this method is the unbiased output by learning the underlying features of both data and its perspective(s) and making the AI agent become a judge capable of independent decisions.
The objective of learning through conflict is developing ethical norms embedding ethical guidelines into AI systems, and making moderation as a middle value for them. Also, transparency in AI decision-making processes fosters trust and facilitates the identification and mitigation of biases.
