Value Learning

Value Learning encompasses approaches where agents or systems discover what matters—the underlying objectives, priorities, and principles—rather than having them explicitly programmed. It bridges problem-solving with alignment, asking: how do entities learn not just how to act, but what to value?

The concept appears across domains. In Machine learning, it refers to techniques where systems infer reward structures from observation or interaction. In philosophy and ethics, it describes how individuals or communities develop moral frameworks through experience, dialogue, and social integration. In organizational contexts, it means cultivating shared norms and priorities.

A central challenge: values learned from limited data may encode hidden biases or fail under novel conditions—echoing concerns in adversarial robustness and validation. Yet value learning also captures something profound about human growth: we don't begin with complete certainty about what deserves our care, attention, and resources. We learn through integration with others and the world.

The field remains young and urgent, especially as autonomous systems become more consequential.

Related

Reinforcement learning, Alignment Problem, Ethical living, Social norms, Language and values, Integration (social)