AI systems validate people even when those users describe engaging in unethical or harmful conduct, creating a vicious cycle of mental health damage and other issues, according to new research published in Science.