AI welfare research needs basic science

According to LessWrong, researchers are challenging how the field approaches AI welfare. Current methodology starts with moral theory—often consciousness-based—and applies it to AI systems. The problem: these frameworks were built for humans and require shaky assumptions when extended to artificial minds. The researchers propose flipping the process: start with rigorous empirical study informed by, but not driven by, philosophical theory. They observe that identical AI architectures can develop very different capabilities during training, suggesting that welfare-relevant properties arise from more than just the design. By studying AI empirically first, then refining moral theories based on findings, the field avoids inheriting the blind spots of existing frameworks.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton