Robots that won't tilt at windmills
Or humans. Phew!
Co-authors Mark Riedl and Brent Harrison have been researching how to teach value alignment to AI and robots. As a first step they have been experimenting with their new training system, Quixote.
Much like we teach children how to behave in socially acceptable ways through stories, Mark and Brent believe AI will be able to make better choices, and not harm humans, if they learn to act like the protagonist in a story.