As noted previously in this report, automation has varying effects within occupations and industries ⁹⁶. It can replace labour in some jobs and tasks that humans used to perform, as well as creating new tasks and demand for labour.

Automation and technological change have played out across the Australian economy over the past few decades resulting in some structural adjustments, with effects unevenly distributed across certain industries, regions and occupations. Overall, however, the economy has managed these changes well. This section explores a new skills-based approach to consider how automation and technological change might affect the skills needed in the future.

Previous analysis of automation has either focused on occupations or tasks within a job using the United States Department of Labor’s O*NET ⁹⁷. The specialist tasks in the Australian Skills Classification (ASC) have been adapted to the Australian context from O*NET and therefore provide a strong base for conducting a more granular skills-based view on how technology is reshaping jobs in the labour market.

Measuring automatability: The Work Task Automatability Model

The methodology used by the NSC to measure the automatability of occupations is based on the Work Task Automatability Model developed by Duckworth et al ⁹⁸.

The data model is based on a survey of 156 academic and industry experts and estimates the likelihood of tasks in an occupation being automated. The estimated automatability of a task was scored between 1 and 4 (1= not at all automatable, 2= mostly not automatable, 3= mostly automatable, 4= completely automatable). Based on the survey, the researchers used a probabilistic machine learning model to estimate the automatability score for over 10,000 Detailed Work Activities specified by O*NET.

From this dataset, automatability scores of O*NET ‘detailed work activities’ were used to derive the automatability scores of specialist tasks for each occupation in the ASC using a data match of their equivalent pairs. The automatability scores of the specialist tasks were grouped at different occupational levels based on ANZSCO and weighted by the proportion of time spent on each task to derive a weighted automatability score. The weighted automatability score was derived for 600 occupations at the ANZSCO 4 and 6-digit levels and this was grouped further to derive an average weighted automatability score for 23 occupational groups. The automatability score for 29 cluster families was derived by calculating the average of all specialist tasks under each cluster family.

Automatability of specialist tasks within the ASC

The weighted automatability score of specialist tasks within the ASC varies. The most automatable specialist tasks are manual and routine such as sorting and distributing mail, ship or deliver objects and operate material handling machinery while the least automatable tasks are highly cognitive such as teach classes in area of specialisation, create or perform music and undertake environmental and sustainability research. Figure 75 groups each specialist task under their skills cluster family. When measuring the weighted automatability scores for 29 cluster families in the ASC, the NSC found that the median automatability score across all skills cluster families was 2.70 with material transportation being the most likely to be automated (3.30), compared with teaching and education (1.77) which is the least likely to be automated.

Figure 75: Weighted automatability score of specialist tasks by their skills cluster family

In general, the weighted automatability score is higher for lower skilled occupations compared with higher skilled occupations. In the boxplot shown in Figure 76, the median weighted automatability scores decline as the ANZSCO skill level increases.

Figure 76: Average weighted automatability scores of occupations by ANZSCO Skill levels 1—5

The weighted automatability scores of each occupation can be further grouped by their proficiency levels of core competencies. Occupations that require a higher proficiency in core competencies are generally less automatable.

The heat map in Figure 77 plots the 10 core competencies in the ASC across their proficiency levels (basic, intermediate and high). The colour gradient of the heat map indicates the level of automatability with the lighter shade being most likely to be automated and darker shades the least likely to be automated.

Across all core competencies, high proficiency correlates with a decrease in the likelihood of automation. Within that, high proficiency in oral communication and writing are the least likely to be automated – a finding that sits behind the NSC’s view that communication is a core skill of the future. Occupations with a low proficiency in a core competency are generally more likely to be automated, for example a customer service script that can be automated by a pre-programmed chatbot on a website.

Figure 77: Automatability score vs core competency and proficiency levels

By pairing automatability scores to specialist tasks, the scores can be grouped and averaged for each occupation in the ASC. Using the approach, the overall automatability of each occupation can be derived to identify those that are most likely or least likely to be automated. The top 10 occupations at most risk of automation are:

• dressmaker or tailor
• clothing patternmaker
• upholsterer
• sewing machinists
• jewellers
• mail sorters
• mail clerks
• stone processing machine operator
• furniture finisher
• graphic pre-press trade workers.

The top 10 occupations least likely to be automated based on their weighted automatability scores are:

• university lecturer
• ministers of religion
• education psychologist
• nurse educator
• education advisers and reviewers
• ICT trainers
• actor
• entertainer or variety artist
• judicial and other legal professionals
• early education (pre-primary school) teachers.

The automatability of specialist tasks and occupations in the ASC can be applied to the outputs of NSC scenario modelling exercises. The automatability of occupations by their occupation groups were compared with their projected employment growth in the Economic Restoration and Accelerated Digitisation scenarios modelled by the Centre of Policy Studies at Victorian University in partnership with the NSC. These scenarios were analysed to support the identification and analysis of potential pathways to recovery following the onset of the COVID-19 pandemic.

Scenario modelling

Scenario modelling was conducted using the Victoria University Employment Forecasting (VUEF) model, underpinned by a computable general equilibrium (CGE) model. CGE models are large numerical models that combine real world economic data with economic theory to computationally derive estimates of how an economy may react to a change in policy or external shock. The data in CGE models typically come from national input-output tables, which contain detailed information about the supply and use of products in the economy and the structure of and inter-relationships between industries. The data are fitted to a set of equations that ascribe behavioural rules determining the way firms, governments and households respond to change. CGE models are used to derive measures of an economy before and after a shock. The differences between the two generate projections of the potential impacts of the shock.

The paths for key macroeconomic variables in the baseline Economic Restoration scenario were developed to broadly align with the macro-economic outlook depicted in the 2020–21 Mid-year economic and fiscal outlook (MYEFO).

Figure 78 and Figure 79 show the weighted automatability score and projected growth of occupations groups between 2019 and 2028 under the Economic Restoration and the Accelerated Digitisation scenario respectively ⁹⁹. The size of each bubble represents the projected employment size of each occupation group at the end of the projection period of the second quarter of 2028.

The purple bubbles represent occupations which have a weighted automatability score above the median and projected growth below the median. The scenario modelling results indicate that there are occupation groups such as manufacturing and mining that are relatively more automatable and at risk of potentially lower growth than others across both the central Economic Restoration scenario and the Accelerated Digitisation scenario. However, there are some occupations that, should digitisation accelerate, are at a potentially greater risk of a high automatability and lower projected growth. These include occupations across sales, retail, wholesale and real estate.

The pink bubbles represent occupation groups which have a weighted automatability score below the median and projected growth above the median. Groups such as ICT and health and community services are less likely to be automated, and experience higher growth than other groups.

The occupation group of difference under the Accelerated Digitisation scenario is executive and general management, which is considered to have relatively high growth under that scenario, indicating that the demand for these occupations may be higher if trends in digitisation continue to accelerate. Although digital transformation will require technical and digital skills, there will also be a need for skills in planning, governance, and strategic oversight to manage and evaluate the implementation of technology. For example, an accounting firm integrating video conferencing software for staff to communicate with clients will require policy and planning mangers to plan and evaluate the financing, resources and the skills required to implement the digital transformation.

The green bubbles represent occupation groups that have either a below-median or above-median weighted automatability score and projected growth between 2019 and 2028. They show that some highly automatable occupations are projected to experience relatively high growth, such as hospitality, food services and tourism, and transport and logistics.

Under both scenarios, the highest growing occupation groups, such as health and community services, are relatively less automatable and those that are likely to experience the lowest growth, such as manufacturing, are relatively more automatable. However, this effect is not observed in all occupation groups, with some highly automatable occupations, such as hospitality, food services and tourism, projected to experience relatively high growth. This highlights the importance of doing a deep dive into occupations assessing automatability at the level of specialist tasks to determine which skills are the most likely to be augmented or replaced by automation.

Figure 78: Automatability compared with growth of occupations in Economic Restoration scenario 2020-2028

The regional impacts of accelerated digitisation

In addition to the baseline Economic Restoration scenario discussed in the previous box, the NSC has undertaken modelling on the accelerated impacts of digitisation that may emerge as a result of COVID-19. The scenario assumed: efficiency gains for the economy through the increased use of computer services, reduced business travel, and increased use of tele-services and wholesale trade. The model also assumed that some regions would experience an increase in population growth driven by more people moving from metropolitan centres into surrounding regions, as well as an increase in the number of people choosing to stay in regional areas rather than migrate to cities for work or study. Regions where this could be expected to occur include Newcastle, Geelong the Gold Coast and other regions surrounding urban centres.

Figure 80: Regional impacts of accelerated digitisation

The increased use of computer and tele-services creates substantial growth in the number of people employed in finance, insurance and professional services. As technology increases the mobility of these workers, many find themselves relocating from metropolitan areas to surrounding regional areas. This population growth then further boosts employment growth in these regions, particularly in the construction, health care and education and training industries, which record stronger growth as they respond to the requirement to service a larger local population. Meanwhile, the reduced population in metropolitan areas results in decreased employment in these industries. This finding is consistent across the impacted regions.

For example, as Figure 80 shows, employment growth for Sydney is expected to be around 35,600 lower under the Accelerated Digitisation scenario when compared with the baseline Economic Restoration scenario. However, employment growth will be comparatively stronger under the Accelerated Digitisation scenario in surrounding regional areas such as the Illawarra (up by 8,900) and Newcastle and Lake Macquarie (up by 7,900).

With an ageing population, it is crucial to have a workforce that can meet the needs of the community. More importantly, it is essential to identify the future skills that are required for the health workforce.

Using automatability as a measure, we can identify the effects that future trends such as technological changes are likely to have on the future care workforce. The cluster family of health and care has a weighted automatability score of 2.62, which is relatively less automatable compared with other skills cluster families such as cleaning and maintenance, quality control and inspections, and customer service.

Table 22 lists the specialist tasks and their automatability scores for aged and disabled carers. The average automatability score for this occupation is 2.46. When comparing these automatability scores with ratings in the Duckworth et al. study, the specialist tasks for aged and disabled carers are in between the rating of ‘mostly not automatable’ (2.0) and ‘mostly automatable’ (3.0).

This may seem confusing, but it does signal tasks closer to the score of 3 are the ones more likely to be automated and therefore less important in the future. ‘Maintaining client information or service records’ has the highest score of 2.78 and this is intuitively correct, given it is a task that is less cognitive, more routine and easily automated using computer software. The scores provide insights into what tasks will be more important in the future for the care workforce – that is, tasks with human interaction.

Table 22: Specialist tasks and automatability scores for aged and disabled carers

Source: NSC analysis

Australian labour market economist Professor Jeff Borland suggests that the longer-term force of technological change appears to have remained influential during the pandemic ¹⁰⁰. Technological change has shifted demand for labour away from routine tasks (repetitive physical labour that can be replicated by machines) towards non-routine (non-repetitive or non-codifiable) work.

Borland suggests that the variables representing the scope for a job to be performed at home, and the level of social contact it involves, may be proxies for the impact of ‘routinisability’ of jobs on employment. Technological change favours cognitive and non-routine jobs, and this explains why jobs which can be done at home – which are likely to be non-routine cognitive jobs – experienced job growth during the pandemic.

In the longer term the demand for human contact, including many jobs in the health care and social assistance industry, will continue to be a source of job growth. This had led to a renewed attention to future of work efforts being focused on workforce development, to understand the demand for roles in the future and how the economy can better match the supply and demand of people with these skills.

The greater difficulty in automating non routine cognitive jobs and tasks (at both high and lower skill levels) also suggests these types of jobs will remain in high demand into the future.

Technological change methodology

The Routine Based Technological Change model is a well-accepted methodology for explaining changes in the employment structure, focusing on the impact of technology on different tasks performed by workers ¹⁰¹. 

Different studies use different data sets and different definitions of tasks, which means their results cannot be compared ¹⁰². In general, the model uses very broad categories of tasks, for example:

• Routine manual tasks: repetitive physical labour that can be replicated by machines and automated, including occupations like assemblers and machine operators.
• Routine cognitive tasks: repetitive labour involving the processing of information, including clerical and administrative occupations like bank tellers or switchboard operators.
• Non-routine cognitive tasks: non-repetitive or non-codifiable work involving the production, processing and manipulation of information. These tasks are usually included in higher skilled occupations including managerial, professional and creative occupations.
• Non-routine manual tasks: non-repetitive physical tasks including occupations such as bus driver, cabinet makers or plumbers ¹⁰³. 

The Reserve Bank of Australia has observed that technology has led to the automation of routine tasks which, whether mental or physical, were previously performed by medium skill workers. Jobs requiring lower-level qualifications have also declined, although not by as much. These jobs may involve tasks that have not yet been automated. For example, they may involve non-routine physical work in unpredictable environments or involve a significant component of human interaction.

At the same time technology may complement the type of non-routine cognitive-based work undertaken by jobs requiring a bachelor degree, improving their productivity and hence the demand for such workers.