We begin our discussion of “autonomy” with its Western meaning for the human individual: “to be autonomous is to govern oneself, to be directed by considerations, desires, conditions, and characteristics that are not simply imposed externally upon one.” Autonomy is “the capacity to impose upon ourselves, by virtue of our practical identities, obligations to act.”1 Similarly, extending autonomy to machines is a partial release from external control that comes with obligations to act. That’s the easy part.
Until the last decade, machines with no human in the loop had very limited repertoires of actions they could take, turning on the pump when they detected the water was rising. From that set of inherent constraints came reliability and understandability. As is obvious, we are transiting an inflection point where machines are gaining trained reasoning capacity that can allow problem-solving without a human in the loop. Even the training can be self-administered: the autonomy of self-modification.
READ THE FULL ARTICLE HERE