Building a Kill Switch for the Terminator: A Case for Slow AI

Abstract

Students in my grad CS ethics course recently led class discussion in an examination of the intersection of vulnerable populations and AI. They decided that they wanted to focus on the elderly. It was a great presentation. They raised issues specific to different cultures across the globe, and they reminded the class that vulnerable populations have vulnerable data too. At one point, one of the presenters instructed the class to remember that the point of bringing technology into the homes and lives of elderly humans is to *increase* their agency, not to restrict it. They reminded us that as we age, stereotypes and family fears will conspire to restrict our choices, under the guise of keeping us safe. However, if AI can be harnessed via various tools and devices to enable us to maintain our agency, to stay in our homes longer as we age, to stay longer in control of our decision-making, then that would be a truly good thing. Imagining a role for AI that enables humans to have more, do more, be more and at their will, is clearly what many young people see as a version of AI development that they would be proud to drive. If my ethics class has led them to this awareness, it has done so by keeping human agency in the center.