Recommendations for Future Development of Artificial Agents [Commentary

Abstract

A set of technologies, loosely referred to as “artificial agents,” is becoming more pervasive and more powerful in the current computing landscape. All artificial agents are built on a computational foundation. Some are purely computational, e.g., Internet bots, search engines, and others are physically embodied entities with computational decision-making components, e.g., robots, unmanned aerial vehicles (UAVs), and autonomous cars. The noteworthy feature of artificial agents - the feature that leads to the artificial agent label - is their capacity to operate autonomously. At some level or to some degree, artificial agents operate independently from the humans who design and deploy them. They are agents in the sense that we deploy them to perform tasks on our behalf and often these tasks involve learning and decision-making. Since humans previously performed many of these tasks, we mark the difference, that is, the machine performance of these tasks, by referring to them as “artificial.