Small AI can learn personal ethics from tiny datasets, but not social duties

Researchers found that small AI models can learn personal ethical rules from very little data. But these models consistently fail to learn how to apply those rules to social situations.

Everyone assumed that if you fed an AI enough good examples, it would learn to be ethical. This paper shows that small AI models hit a wall when it comes to understanding complex social duties, even if they master personal virtues. It means that building small, ethically-aligned AI for real-world social applications will require more than just feeding it a few hundred examples.