The term autonomy has many interpretations depending upon one’s perspective. Also we should distinguish between what we call automation in contrast to autonomy.
Automation characterises a system (an “automatic” system) that has fixed choice points, that is it is programmed with a number of fixed alternative actions that are selected by the system in response to inputs from particular sensors (e.g. a light beam on a lift door, if it is cut then the door won’t close as it is assumed that the beam is blocked by a person or an object) or user actions (e.g. a person pressing the “3rd floor” button sends a signal to the lift computer “3rd floor” and lights the “3rd floor” light in the lift). The lift computer evaluates these choice points independently of the current circumstances (e.g. how full the lift is, how many people are waiting, or the cost of electricity at that time of day).
Exploring the lift example further consider a bank of lifts in a building. Pressing the “up” button in the ground floor lobby calls the lift, and it responds in the same way irrespective of the number of people waiting. A single person going from the 3rd to 4th floors (they could easily use the stairs!) receives the same priority as 10 people waiting at ground level to go to the 5th and higher floors, so the lift stops at the 3rd floor just for one person. As humans we accept this unintelligent, “black box” approach and do not expect or seek any decision-making from such systems. Nor do we seek explanation of how they operate, as their inherent simplicity allows us to readily comprehend their actions, trust them and to rarely question their operations, even if we consider them to be “stupid”.
Autonomous systems differ from automatic ones in that they make rational decisions based on knowledge of the current situation. An autonomous system evaluates the courses of action that could be taken in light of current circumstances. In the lift example discussed under Automation, one person gets the same priority as ten.
Like an automatic system, the autonomous system still accepts sensor and user inputs, but operates with more abstract concepts instead of a knee-jerk response to inputs. As with humans, such an autonomous, decision-making system will balance proactive (goal-directed) and reactive (responsive) aspects when making its decisions.
Consequently, the behaviours exhibited by an autonomous system are far more sophisticated than a simple automatic system and this makes a “black box” approach to autonomy unacceptable. Humans are much more cautious of machines that make decisions and are reluctant to place trust in them without first understanding how the they work. So, autonomous systems must be able to expose their reasoning to humans in a manner that is comprehensible – thus establishing trust.
Looking in more detail at autonomy, it becomes clear that it is not simply a black and white choice or condition. Instead, it encompasses systems with a range of capabilities from those entirely independent of human input (i.e. full autonomy), through to systems that provide only timely advice and leave the human to decide and execute the appropriate action (i.e. intelligent assistant).
Constructing autonomous systems
Accepting that a “black box” approach to autonomy is unacceptable, we need to plan an autonomous system carefully, and construct an autonomy architecture, where decision-making is located within the appropriate person or machine.
An analogy can be seen in human organisations, such as the companies, government departments, schools, clubs and universities that we all participate or work in. In all these management authority is delegated in a structured way. Take as an example a small manufacturing company. Here the Company Board determines the strategy, e.g. “We will focus on expanding sales of our premium office furniture in the North American market this year, and we want to increase our sales there by 30%”. The Board then provides the General Manager with this target (or goal) and an appropriate budget and corresponding management authority with which to achieve it.
Having received this target and delegated authority, the General Manager determines the operational goals and the plan to achieve them (i.e. the annual Business Plan), determines how the various departments need to contribute to meeting the goal, and then delegates targets appropriate to the capabilities of each particular unit. An appropriate level of management authority is provided to the various functional unit Managers (e.g. Sales, Manufacturing, Purchasing, Dispatch, Customer Support).
At this point there will be discussion and negotiation as to how these delegations will work and adjustments to the goals and processes will be made. For example a manager with an effective record of managing Customer Support would be given a larger degree of management autonomy than perhaps a deputy manager standing in for a period, or a person new to the role. Consequently the degree of delegated authority is not “cast in stone”, but may alter according to circumstances.
Each Manager is then in a position to ensure that his or her functional unit delivers that unit’s contribution to achieve the overall business goal. This leaves the General Manager with a clear desk and unencumbered by detail is ready to deal with exceptions or problems raised by the managers as they encounter changed conditions or difficulties in performing their various roles.
The majority of successful human organisations combine three key factors: clear high-level direction, effective delegation and good communications. The more that authority can be delegated to the operational levels the more effective and flexible the organisation becomes. In this way problems and unexpected events can be dealt with immediately and without recourse to higher levels in the organisation.
In addition, successful organisations also have clear processes and means for escalating problems when they impact the organisation more widely. The challenge is to combine the effectiveness of delegated decision-making with the ability to rapidly escalate the problem as far as necessary as the circumstances demand.
Autonomous and semi-autonomous software systems
Having clarified the distinction between automation and autonomy, and explored the concepts of delegated autonomy in human organisations, it is worth exploring how these concepts can be realised in software systems and autonomous vehicles, to provide capabilities not currently available with conventional software.
The first questions would be “Why would we want to have autonomy? What advantages does it provide?” Returning to the example of the human organisation, we can see that delegated authority, which provides a level of autonomy appropriate to the circumstances offers:
- A clear management structure, with the lines of authority and communications clearly defined. This ensures that people can comprehend the management structure that they participate in, and it is widely understood.
- Decision-making is delegated where possible, ensuring that senior managers – responsible for strategic decisions – are not bogged down in assimilating detail in order to make minor decisions. Consequently management workload is shared appropriately among the managers and members of the organisation. Delays resulting from too high a workload are minimised.
- The organisation is more responsive, it can respond as quickly as possible to changes, as many decisions don’t need to go to the top of the organisation.
- Communications are improved, as managers and members only need to receive information necessary for their role – not irrelevant detail which distracts from their prime responsibility.
- There are established paths of authority and criteria for escalation, ensuring that decisions are made at an appropriate level in the organisation, which is much more likely to lead to a high-quality outcome.
- Productivity is higher as people are able to make many decisions locally, referring to higher management only on key decisions as defined by the management strategy.
- This leads to the concept of autonomy that is comprised of various decision-making sub-systems, each responsible for its domain, and coordinated by an autonomous manager which ultimately works with its human controllers.
This leads to the concept of autonomy that is comprised of various decision-making sub-systems, each responsible for its domain, and coordinated by an autonomous manager which ultimately works with its human controllers.