Autonomy in AI Systems: Rationalizing the Fears

Abstract

The news, popular culture, and legislatures are concerned with the recent reemergence of artificial intelligence (AI) technology. Some of the fear is fueled by the terminology including artificial intelligence, machine learning, deep learning, and superintelligence that have specific meaning within the technology community, but can be misunderstood by the general public or by other fields of inquiry. Because of this, the fears are not well linked to what the technology actually does. The type of AI technology such as neural networks or decision trees does little to clarify the conversation. However, considering where an AI system exhibits autonomy better highlights what the systems capabilities are and what may be rationally feared from such capabilities. This paper develops a typology of autonomous functions within AI systems and why they matter.