Where automation ends and autonomy begins

As the marine industry increases its use of automated processes, and AI matures, the time is ripe for a clearer understanding of when an automation system becomes an autonomous one.

Navigation on autopilot

Automation systems rely on clearly defined logic, mathematical models and algorithms to guide their actions. An example is a ship’s autopilot system tasked with controlling course according to a predetermined setpoint – such as true heading at 90 degrees. The autopilot senses the vessel’s heading using a gyrocompass, analyzes the deviation between actual and desired heading and adjusts the rudder angle accordingly.

Mathematical models and algorithms can be designed to enable a conventional automation system to handle very complex tasks, but the technology is only as sophisticated as the input it receives – it cannot ‘think’ for itself. If the actual situation calls for a deviation from the plan, the system is missing these inputs as well as the capability to analyze those, and therefore is unable to respond appropriately. As a result, human intervention is required to bring the operation to a successful conclusion.

For instance, if the vessel is on course to collide with another ship, the ability to understand the risk and adjust course in line with the rules of the road relies on the operator’s perception, understanding and decision-making capabilities. So, while a standard automation system can sense, analyze and act based on existing input, it lacks the ability to recognize and comprehend unexpected threats and present solutions to deviate from those inputs and mitigate the risks associated with the situation.

An autonomous system, by contrast, simulates these human faculties within the scope of a particular operation. An autonomous navigation system would be able to perceive another ship, interpret the threat and change course to prevent a collision in a safe and efficient manner. However, in the foreseeable future even an autonomous system would require a human in the loop to complement the operation in a collaborative manner. In fact, the combination of human capabilities and experience, and autonomous technology can do a better job together than any one of the two could do alone.

Automated auxiliaries

Autonomous and automation systems are not only used for vessel navigation; the technology is also deployed in auxiliary machinery such as cranes. An advanced automated (but not autonomous) crane can adapt its behavior depending on factors such as the length of a rope or the mass of a load. The effects of these variables are relatively simple and can be described with mathematical modeling.

Again, however, if the automated crane was presented with an unforeseen obstacle preventing it from hoisting the load, it would rely on the intervention of a human operator. An autonomous crane, on the other hand, could ‘perceive’ the item, recognize that it presented an obstruction and find an alternative path by which to move the load. Again, the human supervisor would be on hand to intervene if necessary.

The distinction between automation and autonomy, then, lies in the presence, or absence, of the decision-making cycle that includes perception, understanding and problem-solving. In short, a system can be considered autonomous where technology steps in to handle multi-sensory perception and interpretation of the current situation based on previous experience or learned concepts to apply spontaneous problem-solving.

It is worthwhile to note, however, that autonomy is not about an ‘all or nothing’ approach. A system that begins to complete or execute any aspect of the ‘perception – understanding – problem-solving’ loop can be considered to have aspects of autonomy. This is due to the fact that such a system is able to partially or fully execute tasks which, by contrast, would be only done by human in an automation system.

The far end of the spectrum

Although autonomous systems can mimic certain cognitive processes and take action towards a favorable outcome, today their abilities are limited to specific tasks – like navigating a ship or controlling industrial machinery. A fully autonomous system, regardless of the context of its application, would apply humanlike creativity, judgment, learning and knowledge to solve any number or any type of problems in any context.

Today, ‘generic AI’ does not exist, and it is likely to be a decade – at least – before the technology with the ability to learn any type of application emerges without context-specific tailoring. And even then, humans will remain key for supervising the vast majority of shipboard activities. Yet, even in its current stage of maturity, autonomy is a valuable tool which is easing the physical and mental burden on shipmaster and crew and which – already – is changing the roles of those working on board ship for the better, safer and more efficient operations. And even when the level of automation increases, we will always need competent crew working alongside the technology.