XNews

Explainer: 'approval reward' concept clarifies AI alignment disagreement between pessimists and optimists

The post introduces 'approval reward'—the human psychological mechanism where imagined social approval motivates behavior—and uses it to explain the core divide in AI safety: pessimists fear future AIs lack approval reward and act as uncompromising goal optimizers; optimists point to human-like imitation in current models. The post outlines six implications for goal stability, hierarchical desires, cooperation, norms, corrigibility, and whether AIs could be given approval-like drives.

Source posts