AI Alignment Considered Harmful (togelius.blogspot.com)

A recent critique challenges the concept of AI alignment, arguing that the quest to ensure AI systems reflect a unified human moral framework is fundamentally flawed. The author points out that humanity is diverse, with conflicting moral beliefs across politics, religion, and culture. This lack of consensus leads to an inherent risk in allowing a small group—often affluent individuals from tech hubs—to dictate the "alignment" of AI, potentially resulting in oppressive systems that overlook the complexity of human values. The significance of this argument lies in its call for a reevaluation of how the AI/ML community approaches safety and ethics. By emphasizing the necessity of pluralism in AI development, the author warns against complacency in trusting AI systems to always align with human intentions. Instead, it advocates for a future where multiple AI systems coexist, acknowledging the reality of unaligned models and urging the maintenance of robust oversight and diverse safeguards. This perspective encourages ongoing research into controllability rather than falsely attempting to "solve" alignment, suggesting that a competitive landscape of AI can lead to better adaptive strategies for societal resilience.