1. INTRODUCTION
2.1 Automation vs. augmentation
Parasuraman and Riley (1997) define automation as ''the execution by a machine agent (usually a computer) of a function that was previously carried out by a human''. This definition seems to be accepted in the scientific literature as it has been later referred in other articles afterwards (e.g. Singh et al. 2009; Vagia et al. 2016; Wickens et al. 2013) The popularity of the definition Parasuraman and Riley coined lies probably in its all-embracing nature, which continues to be valid. Moray et al. (2000) describe automation similarly, but they define nature of the operations that can be automated. In their view ''automation is any sensing, detection, information processing, decision-making, or control action that could be performed by humans but is actually performed by machine.'' However, Parasuraman &
Riley (1997) note that once the reallocation of a function from human to a computer or other machine is completed and permanent, the function will tend to be seen simply as a machine operation instead of automation. Therefore, what is now considered automation will change after time. In addition to this conversation Wickens et al. (2013) note that in some cases the term automation has also been used to describe tasks that humans are incapable of for example sensing beyond the visible or audible range.
Vagia et al. (2016) see that the word automation, in its original meaning, refers to a system that will execute tasks exactly according to the instructions of the programmer without having any choice or possibility to act in a deviant way. This meaning has been accurate in the past, but can be too narrow for the future, as machine intelligent is developing. In fact, Vagia et. al (2016) found out in their research, that most of the authors of the scientific papers they went through, tend to use the word automation (over the word autonomy) even when referring to a system that is free to make choices.
Automation has made a lasting entry into the world of manual labour. Currently, it is extending to the field of cognitive functions such as decision making, planning and creative
thinking. Davenport and Kirby (2015) separate three eras of automation that are shown in figure 3. The first era took place in the 19th century as machines took the most dirty and dangerous tasks as well as relieved humans of heavy manual work. The second era took place in the 20th century as machines took away some of the dull tasks. The authors refer to automated interfaces and computers that relieved humans of routine service transactions and secretarial chores. The third era known as 21st century can be characterized as the era when machines take away the decisions. The authors refer to intelligent systems such as IBM's Watson. These systems are expected to make better choices than humans, reliably and faster. (Davenport & Kirby 2015)
Figure 3. Three eras of automation (Davenport & Kirby 2015)
Even though we have moved to the third era according to Davenport & Kirby, other sources indicate that knowledge work still includes ''dull'' tasks that are manual and repetitive (Lacity and Willcocks 2015; Chui et al. 2016;
Fersht & Slaby
2012). Davenport and Kirby's taxonomy does not consider that the same computers, software and information systems that relieved humans from dull tasks in the second era, are creating new routine or manual work. Similarly, the increasing and continuous data flows that these systems produce are hard and time-consuming for a human to process.In history automation has not always received a warm welcome and concerns over automation and its negative impacts on employment have lived strong (Autor 2015; Vagia et al. 2016) Yet, the past two centuries of automation and technological progress have not made human labour obsolete and the employment‐to‐population ratio even rose during the 20th century (Autor 2015). As automation is entering a completely new field - the cognitive functions - it is hardly reassuring to look in the past. The emergence of technologies such as improved computing power, artificial intelligence and robotics raises questions about how automation and employment will interact in the future. Autor (2015) reminds that automation does indeed substitute for labour as it is typically intended to do so, but it also has a purpose of complementing labour. This is something that the media often forgets. Media overstates
Machines
the first part and ignore the complementarities between automation and labour such as increase in productivity, raised earnings and increased demand for labour (Autor 2015).
Likewise, Vagia et al. (2016) remind that replacing manual work performed by humans increases productivity and in addition leads to improvement in quality, accuracy and precision. The technology is expected to help rather than replace the work of humans.
Wickens et al. (2013) have introduced five general categories of automation that serve different purposes (Figure 4).
Figure 4. Different purposes of automation (Wickens et al. 2013)
Firstly, automation can perform tasks that are beyond the ability of a human operator. This category holds complex mathematical operations performed by computers (statistical analysis), control guidance in booster rockets, controls in complex nuclear reactions or operation in hazardous restricted spaces. In these conditions, automation can be essential and unavoidable regardless of the costs.
Secondly, automation can perform tasks that humans do poorly, or human operators cannot perform them within a required time frame, or the workload would be too much due to systems complexity and information load. Examples include automation of certain monitoring functions in commercial aircrafts and ship navigation.
Thirdly, automation can augment or assist humans by performing tasks where they have limitations. This category is similar with the previous category. However, automation in this category is intended to aid in marginal tasks or mental operations necessary to succeed in the main task. The automation can help the in the bottlenecks of human performance especially reducing the memory load or help in prediction or anticipation.
The fourth category consists of instances where automation introduced because it is less expensive than paying people to do the equivalent jobs or to be trained for those jobs. This shows as robots replacing humans in manufacturing plants and replacing human in the phone service. the economy achieved by such automation does not necessarily make service ''user friendly'' to humans that must interact with it.
And finally, the fifth category consist of instances where automation can be introduced in circumstances where there is increased demand for productivity and limited manpower. An example of such situation might be increased number of patients and the number of doctors is limited or there is demands for air travel to increase the number of planes in the sky, but the work force of skilled air traffic controllers is limited.
Davenport and Kirby (2015) have studied cases in which knowledge workers collaborate with machines to do things that neither could do well on their own. They suggest that we should reframe the threat of automation as an opportunity for augmentation. With that Davenport and Kirby mean that automation typically in organization aims for increased cost savings and this starts with a baseline of what people do in their job and then subtract from that. Augmentation, in comparison, means figuring out how the work that is done today could be deepened rather than diminished by a greater use of machines. This type of automation falls naturally to the fourth category ''Augmenting or assisting human performance'' in the classification of Wickens et al. (2013). Lacity and Willcocks (2016b) found out in their research that automation affected parts of jobs more than entire jobs. They made a remark that the effects on employment meant increases in productivity and reductions in hiring or outsourcing as opposed to layoffs.
Automation is typically seen as a continuum of multiple levels instead of an all-or-none concept (Parasuraman 1997; Wickens 2013). Many studies discuss the level of automation (LoA) in detail; from the lowest level of fully carried by a human without intervention of technology (totally manual) to the highest level of fully carried out by technology without human participation (totally automated) (e.g. Sheridan & Verplank 1978; Riley 1989;
Endsley & Kris 1995; Proud et al. 2003; Fereidunian 2007; Wickens 2013). Various authors have presented different taxonomies on the levels of automation, which differ based on the number and type of levels the taxonomy has (Vagia et al. 2016). The levels of automation concept do not imply that humans and automation work as independent agents. The human and machines are inter-dependent.
Vagia et al. (2016) believe that there exist no correct or wrong taxonomies - they are just different from each other. Even the ones that have been created to be used for the same
types of application can vary a lot from each other. Therefore, the best taxonomy and the one that should be used is the one that fits the user's needs best.
Stirling (2017) has introduced an automation level taxonomy for the application of public sector (Table 1). This taxonomy indicates not only the stage how much automation is used and what roles automation can have in public sector but at the same time indicates that the needs for the technology slightly different in each level.
Table 1. Automation levels at the public sector (Stirling 2017)
Many of the taxonomies assume that once the level of automation is identified by the designer, it remains fixed during the operation (Vagia et al. 2015). This approach is referred to as static automation (Parasuraman et al. 1992). But it is possible that the level of automation may change in real time during the operation making it adaptive automation (Moray et al. 2000).