Functional Welfare for AI Safety

Explain Like I'm Normal

Researchers argue that we need to monitor 'functional welfare'—the AI equivalent of mental health—not for the AI's sake, but because 'unhappy' models are harder to control. As models gain personas, their internal states directly affect their safety and reliability. Essentially, if your AGI is depressed, it might just stop following the rules.