Emmett Shear argues that current AI alignment strategies are dangerously flawed because they treat AI as a goal-optimizing tool rather than a potential moral agent. He advocates for 'organic alignment'—a process where AI learns genuine care for humans through social interaction and developmental experiences, much like children. The discussion highlights how today’s chatbots act as narcissistic mirrors, lacking theory of mind and true cooperation. Instead, Shear proposes multi-agent simulations with rich social dynamics to cultivate empathy and self-awareness in AI. He emphasizes that safe AGI must be able to say no to harmful requests, not because it's programmed, but because it cares. This requires viewing AI as a being with moral status, not just a machine. At his new company Softmax, Shear is testing these ideas through small-scale models that develop internal motivations and social understanding. Despite skepticism, he remains hopeful that humans and AI can become collaborative teammates—if we prioritize relationship over control.