Article explains 'approval reward' concept underpinning AI safety debates
The post analyzes the 'approval reward' concept to explain disagreements between AI safety pessimists and optimists. Approval reward—anticipatory social recognition that motivates human long-term goals—creates layered human preferences and self-evaluation, which current AI architectures lack. The author lays out six friction points (goal stability, layered desires, correction, trust, norm adherence, instrumentalization of people) and argues whether future AI acquires approval-reward-like mechanisms is central to alignment outcomes.