New algorithm trains AI to avoid bad behaviors

Robots, self-driving cars and other intelligent machines could become better-behaved thanks to a new way to help machine learning designers build AI applications with safeguards against specific, undesirable outcomes such as racial and gender bias.