Automation framework – Levels of Automation and Cynefin

Blog / Simon Vajda / July 18, 2018

As technology becomes more prevalent in everyday life, from smartphones to “cloud” services, the biggest question is how does this technology help address the needs and wants of everyone? One way is through automation – or the removal of the human from a process, freeing up human potential to create, live and experience.

Automation, as will be shown later, can be categorised into a few broad categories which align nicely with a framework for problem solving called Cynefin (/ˈkʌnɪvɪn/ KUN-iv-in). By pairing a solution, such as Automation, with a problem solving framework, such as Cynefin, we can start to think of any problem (e.g a repetitive task) and shape a potential solution for it (e.g a simple task can be fully automated). But more on this later!

Levels of Automation

Automation, the technology by which a process or procedure is performed with human assistance [1], is normally applied to manufacturing but as technology is becoming more pervasive in day to day lives, we start to see automation in more and more things. An incredibly simple idea to think about automation is the ways in which you can call someone on a phone. You can either ask your voice assistant to call someone for you, or you can click a ‘call’ icon next to the person’s name in a contacts tool, or you can manually type in the person’s phone number and hit dial. This demonstrates, at an extremely simple level, that automation is a scale and has multiple levels of human involvement.

This scale of automation is a recognised phenomena called Levels of Automation (LoA) and has existed for a while [3]. A recent review of Levels of Automation in Literature [2] highlights that there is a big variance in how people refer to and identify LoA. LoA have appeared in a variety of different applications including; avionics, teleoperation systems, remote control operations, advanced manufacturing, real time control (air traffic control and piloting), spacelift teleoperation, unmanned aerial vehicles, and power distribution [2].

A number of differing scales of Levels of Automation exist for a wide variety of applications (as listed previously). Vagia et al [2] have compiled and presented these levels into a figure (Figure 1 [2]). The level of granularity that exists between a fully manual solution and an autonomous/fully automated system can be seen in the distinction between levels as identified within literature. These are presented in Figure 1. Across all applications where LoA are mentioned there are 19 distinctive levels identified. Typically a LoA will contain no more than 10 levels.

0 – Manual 10 – Executes with Human Approval
1 – Data Acquisition 11 – Decision Support
2 – Telepresence 12 – Executes If No Human Veto
3 – Manual Control with Intelligent Assistance 13 – Executes and Informs Human
4 – Remotely Operated 14 – Partner
5 – Computer Offers Decision 15 – Informs Human If Asked
6 – Narrows Down Selection 16 – Informs human If Decides
7 – Director/Agent Control 17 – Supervisor
8 – Share Control 18 – Autonomous System/Fully Automated
9 – Suggests One Alternative
Fig.1 – Levels of Automation from across literature [2].

The scale between varying LoA presented above covers a wide variety of Human Computer Interaction but for understanding Automation and LoA the above is overwhelming. Therefore a generalised LoA proposed by Vagia et al [2]  exists. The proposed LoA contains 8 levels and is presented in Figure 2. This generalised approach covers the variance found within the LoA within literature.

0 – Manual Control
1 – Decision Proposal Stage
2 – Human Decision Select Stage
3 – Computer Decision Select Stage
4 – Computer Execution and Human Information Stage
5 – Computer Execution and On Call Human Information Stage
6 – Computer Execution and Voluntarily Information Stage
7 – Autonomous Control Stage
Fig.2 – Generalised Levels of Automation [2].

This generalised approach, while being extremely beneficial for thinking about LoA, can be adjusted as required for a given application. It can also be abstracted further into 4 fundamental levels:

  • Fully Manual – The user is in full control of the process.
  • Decision Support System – In the simplest case, the system offers decisions for the user to select for execution. In the most advanced case, the system selects a decision and seeks confirmation from the user.
  • Human in the Loop – In the simplest case, the system executes and informs the user. In the most advanced case,  the system executes and decides if to inform the user.
  • Full Automation – The system is in full control of the process. 

can be seen from the generalised LoA, there are bounding levels – Fully Manual on one end and Autonomous System/Full Automation on the other. The in-between levels can be quite clearly split into a more manual approach where the system provides decision making support and then the remaining levels where a system has a higher level of autonomy and the human operator is involved in some capacity. Even though we lose some granularity and control by merging a number of levels together, it is useful to group into simple levels because we can match these to another framework.


In a previous blog post Cynefin – The Decision Makers Framework for Software Engineering, I introduced a well regarded thought process used by leaders and decision makers. Cynefin proposes that all situations can be thought of as belonging to a domain; Simple, Complicated, Complex and Chaotic. Each domain has a strategy and set of guidelines that can be used to deal with that situation.

Fig.3 – Cynefin Framework [4].

In our simplified LoA we have 4 levels. Fully Manual, Decision Support System, Human in the Loop and Full Automation. In Cynefin we have 4 domains, Simple, Complicated, Complex and Chaotic. A simple problem (such as repetitive manufacturing task), is an incredibly easy thing to automate. A chaotic problem, (such as responding to a disaster or warfare), is an incredibly chaotic problem due to the nondeterministic nature of the literal battlefield. There may be procedures outlined, but in the moment and heat of the situation no system could act and respond. Therefore, as Cynefin mentions, Chaotic situations are the domain of novel practices and are solved by acting first. At this point, the connections between Cynefin and the Simplified LoA are obvious.

Connecting Automation and Problem Solving

This blog post has introduced a way of understanding Automation, as Levels of Automation, as seen within literature [2]. This post has also examined the key concepts on Cynefin, a leader’s framework for thinking. Levels of Automation (LoA) can be generalised and presented as 4 main levels; Fully Manual, Decision Support System, Human in the Loop and Full Automation. By presenting a generalised LoA there is an argument that can be made that this LoA and Cynefin align. This alignment is presented in Figure 4.

Mapping between Cynefin and Level of Automation
Simple Full Automation
Complicated Human in The Loop
Complex Decision Support System
Chaotic Manual
Fig.4 – Mapping between Cynefin and a generalised Level of Automation.

There are a number of insights such a mapping can introduce. These insights can be beneficial for problem identification and the strategies from Cynefin can be used for informing a solution direction. A discussion regarding the benefits and insights of this mapping will be presented in future blog post.


  1. Groover, Mikell (2014). Fundamentals of Modern Manufacturing: Materials, Processes, and Systems.
  2. Vagia, M., Transeth, A. A., & Fjerdingen, S. A. (2016). A literature review on the levels of automation during the years. What are the different taxonomies that have been proposed? Applied Ergonomics.
  3. Sheridan, T. B., & Verplank, W. L. (1978). Human and computer control of undersea teleoperators. Massachusetts Inst. of Tech.; Man-machine Systems Lab.; Cambridge, MA, United States.
  4. Berger, Jennifer Garvey; Johnston, Keith (2015). Simple Habits for Complex Times. Stanford, CA: Stanford University Press, 237, n. 7.

Header image courtesy of Liane Metzler

Thanks to Tanya Frank, Shannon Pace, James Gardner and Antonio Giardina for proofreading and providing suggestions.