What is autonomy?
Let us go through a thought experiment: if some team members work independently, can their cumulative effort yield the same performance as if they had worked as a team? Mostly, this is not the case. Indeed, smaller parts together need to be combined to solve the problem in real life. For instance, every worker on a building site can place their few bricks, but if they are not placed together there will never be a house.
Now, let us replace one of the team members with an “autonomous agent.” If the autonomous agent and the other team members work independently, can their cumulative effort yield the same performance as the team’s effort? As before, no!
Autonomous systems are in fact not autonomous: the goal of AI systems is never to work completely on their own, but in conjunction of other components and possibly with humans in between.
Some other myths of autonomy include:
Another example: an unmanned aerial vehicle (UAV).
UAVs typically have the goal of eliminating the pilot from a manned aerial vehicle, i.e. take off on its own, fly a course and land on its own. But then why do we still have control stations? This is for safety measures, economic reasons, and more. So, essentially you do not completely remove the human from the system, because the high level goal is not just to fly independently.
With all AI systems that we want to build, the goal is not to make systems that work completely on their own. There are always humans involved in the overarching social and technical context. This is what we know as interdependence. The boundary of interdependence may shift, but there will always be interdependence!
The first and most important point to raise is that autonomy does not mean replacement. The goal of AI systems is never to replace humans, but rather to build intelligent systems to work with humans.
Actually, replacement is rarely what actually happens with the addition of technology. Technology changes the nature of the work. Replacement is not the only, or even the best, way to view technology:
Additionally, autonomy is not just about task allocation but also system integration (integrate towards common goal). The goal is to work interdependently, not independently.
Finally, interdependence has critical implications that permeate all design elements (including algorithms and interfaces). There needs to be a good understanding of every part of the system to work together efficiently.
Interdependence describes the set of complementary relationships that two or more parties rely on to manage required (hard) or opportunistic (soft) dependencies in joint activity.
It is not limited to compensate for some missing ability – or hard constraints, but also involves normal supportive behaviour associated with teamwork – or soft constraints.