What can artificial intelligence not do?

Summary #

Artificial intelligence can do many things better than people. But it cannot be human. AI systems are built on computers which are deterministic. A deterministic system is bound by its instructions. It won’t have the capacity to act independently and make its own choices. This makes it different from humans.

Explanation #

Artificial intelligence (AI) is built on computers. Computers have progressed a lot in the last few decades, but they still have a limitation. They are deterministic. A computer is bound by the instructions built into it. It will always give the same output for a given input. It cannot ignore the instructions and give a different output for a given input. This is a fundamental tenet in computing. No amount of progress in the field can change the deterministic nature of computers. Even advances in machine learning cannot change it. If a computer is built to learn on its own, it will still be bound by the instructions which help it learn. So AI can never have the capacity to act independently and make its own choices (called “agency” in social sciences). AI won’t have agency. But human beings can instinctively sense agency in themselves and others. A human being can sense his ability to behave independently of his biological instructions. All human civilizations use agency as a basic principle of justice and morality.

The lack of agency in AI will be very apparent to people when they interact with it. The AI will seem to lack a “personality”. People will easily identify AI from humans. AI will always be seen as different from people. It will not be accepted as human. It will be treated as a tool. People won’t be able to have any kind of emotional relationship with it. The lack of agency will also preclude AI from being held responsible for its actions. The human makers of AI will be held responsible for its actions.