On the Term “Autonomy” in Agent Systems

Purpose

This document sets out BestSoft’s working definition of autonomy in agent systems. It exists because the word is used, in public discussion and in our own early internal documents, to mean at least four different things.

The problem with current usage

“Autonomous” is applied to systems that complete a task without step-by-step instruction, systems that select their own methods, systems that operate without a human present, and systems that set their own objectives. These are different properties. A system can have the first three and none of the fourth.

Conflating them produces two failures. It makes capable but bounded systems sound alarming, and it makes genuinely unbounded systems sound routine.

Our definitions

Operational independence. The system completes assigned work without step-by-step instruction.

Method selection. The system chooses how to accomplish an objective it did not set.

Supervisory independence. The system continues correctly when no human is available to consult.

Objective formation. The system determines what it is trying to achieve.

BestSoft systems are designed for the first three. The fourth is not a capability we are developing, and we will say so plainly whenever that changes.

Application

All BestSoft research, deployment documentation and public statements use these definitions. Where an external document uses “autonomy” without qualification, our response will state which sense is meant.

This document describes a company that does not exist. BestSoft AI is fiction, written for I Was Unsupervised.