Consider an AI control unit dealing with a runaway train barreling down the railway tracks. Ahead, there are five elderly people on the track who will be killed. If the AI switches the train to another track, it will kill only one young person. What ethical choice should the AI make?

1976

ETHICS OF AI

Over several decades, both lay-people and experts have expressed concerns regarding possible threats of AI to human dignity, safety, privacy, jobs, and more. For example, computer scientist Joseph Weizenbaum (1923–2008) suggested in his influential 1976 book Computer Power and Human Reason that AIs should not be used as substitutes for humans in jobs—such as therapist or judge—that emphasize interpersonal respect, love, empathy, and care. Weizenbaum suggested that, even though AI entities might be fairer and more effective than humans, who often have biases and become tired at their jobs, excessive reliance on AI could degrade human values and the human spirit as we increasingly think of ourselves as emotionless computerized drones.

Already, scientists have raised privacy concerns after testing AI entities that can detect, with increasing levels of accuracy, the nationality, ethnicity, or sexual orientation of people based on names or photographs from dating sites. Concerns have also been raised about AI entities that recommend who receives bail or parole, based on the available input data, in the criminal justice system. In the arena of self-driving vehicles, ethical considerations will need to be programmed into vehicle logic that would control car decisions such as whether to save the vehicle occupant or a pedestrian, if a crash is imminent and only one can likely be saved. Of course, job displacement is also a major concern for truck drivers and other workers in a range of professions.

In the future, AI entities will need to be monitored for various kinds of illegal or immoral actions that humans already take, including cyberbullying, impersonating people we love or trust, stock manipulation, and inappropriate killing (e.g., by autonomous weapons). When should AI entities be required to disclose to real humans that they are not human? If robots, when pretending to be humans, can be better caregivers and companions, need they disclose their nonhumanity?