AI, Skills and Jobs: Some considerations on the future based on the current evidence
Technological change, employment and skills
The conventional view is that technological change is both employment and skill enhancing. It has, historically, tended to favour employment growth in relatively high skilled jobs. More recently as technologies have become more sophisticated, especially those which encompass a degree of artificial intelligence (AI), concerns have been expressed about the capacity of new technologies to adversely affect the demand for employment and skills.
AI has been around, in one form or another, since the 1950s. In its earliest incarnations it seemed to be concerned with the automation of relatively simple processes by today’s standards. Much more recently the emergence of generative pre-trained transformers (GPTs) such as ChatGPT has focussed attention on, amongst other things, the potential for AI to affect larges swathes of employment. This might mean job loss, changes to the content of existing jobs, or the creation of new jobs. These are concerns which have been raised in time immemorial in relation to any kind of technological change. In the case of AI there is perhaps more concern that AI has the potential to automate tasks previously considered to be beyond the reach of computers and robots. This might mean that the impact of AI on employment and skills will be less benign than previous vintages of technological change. Time will tell.
Presented below are insights based on the current evidence base about the impact of AI on employment and skills to date. In thinking about the future, it draws attention to how the more beneficial aspects of AI for employment and skills might be accentuated.
How AI affects employment and skills demand
In thinking about AI, or any kind of automation for that matter, it is useful to consider three inter-related concepts:
- automation – where machines take over some or all of the tasks undertaken in a job;
- augmentation – where the same machines complement the tasks undertaken by workers such that their productivity increases; and
- task reinstatement – where new tasks or new jobs emerge as a consequence of new technologies being introduced.
To understand how AI might affect the employment one needs to take a step back and look at the historical evidence on the impacts of technological change on labour and skills demand. Autor’s excellent survey provides the interested reader with a succinct summary (Autor, 2022). Empirical data from the USA and Europe reveals that, over time, more conventional forms of technology were able to automate those tasks which followed an explicit set of rules. Those tasks might be relatively complex but because they followed a set of rules, they could be replicated by computer programs.
Evidence from the USA indicated that those jobs at risk of being automated tended to be in the middle of the occupational hierarchy (e.g. skilled trades jobs according to the UK Standard Occupational Classification). In contrast, the tasks undertaken in relatively high skill jobs that required abstract thinking and interpersonal skills, such as those found in managerial and professional occupations, proved more difficult to automate. The tasks they undertook were of a kind which could not be reduced to an explicit set of rules that could be replicated by computers and robots. Less skilled jobs were also seen to be immune to automation with the evidence revealing little or no substitution or augmentation by machines and robots.
Technological change does not take place in a static environment. Technology creates new tasks and new jobs. Accordingly, there is a task reinstatement effect where new jobs or new tasks emerge (Acemoglu and Restrepo, 2018). So long as the reinstatement effect is bigger than the automation one, then labour demand will increase. There is plenty of evidence of new jobs being created. Autor et al (2021) estimate that around 60 per cent of employment in the USA in 2018 was found in jobs that had titles which did not exist in 1940. Technology evidently has the capacity to create new kinds of employment.
Where technological change, such as that represented by AI, induces relatively weak productivity growth the implication is that job creation will be relatively weak. Acemoglu and Restrepo (2020) suggest that this has been the problem affecting (some) western economies over the recent past. The tasks which have been automated to date are ones which have brought about relatively weak productivity growth and with it a relatively weak reinstatement effect.
Evidence from the UK
There are a wide range of estimates of how AI might affect the demand for employment and skills across the globe (Goldman Sachs, 2023; Eloundou et al, 2023). Evidence for the UK suggests around 7% of existing UK jobs could face a high probability of being automated over the next five years, rising to around 18% after 10 years and just under 30% after 20 years (PwC/BEIS, 2021). The same study reveals that AI will have a differential impact on the demand for employment and skills. Net employment gains are expected over the next 20 years in sectors such as health and social work, professional and scientific services, and education, whereas net job losses are forecast in manufacturing, transport and logistics and public administration. The same study indicates that net employment gains will be in managerial and professional jobs whereas net job loss will be in less skilled jobs such as classified to elementary occupations.
It is not simply about job gains and losses. There is an interest in knowing about the jobs which are likely to be affected in some way by AI. This may mean that the content of jobs is altered but not necessarily levels of employment. An indication of AI’s potential impact on jobs and skills can be obtained from the AI Occupational Exposure (AIOE) score developed by the DfE Unit for Future Skills (DfE, 2023). Table 1 shows the top 10 occupations most exposed and least exposed to AI with a distinction between all AI applications and large language models. While a degree of caution is required when interpretating the rankings it is apparent that AI has the potential to affect relatively high skilled jobs much more than low skilled ones.
Table 1: Occupations most exposed to AI
Exposure to all AI applications | Exposure to large language modelling | |
---|---|---|
1 | Management consultants and business analysts | Telephone salespersons |
2 | Financial managers and directors | Solicitors |
3 | Charted and certified accountants | Psychologists |
4 | Psychologists | Further education teaching professionals |
5 | Purchasing managers and directors | Market and street traders and assistants |
6 | Actuaries, economists and statisticians | Legal professionals n.e.c. |
7 | Business and financial project management professionals | Credit controllers |
8 | Finance and investment analysts and advisers | Human resource administration occupations |
9 | Legal professionals n.e.c. | Public relations professionals |
10 | Business and related associate professionals n.e.c. | Management consultant and business analysts |
Source: Table 1, p.12, DfE (2023)
Table 2: Occupations least exposed to AI
Exposure to all AI applications | Exposure to large language modelling | |
---|---|---|
1 | Sports players | Fork-lift truck drivers |
2 | Roofers, roof tilers and slaters | Roofers, roof tilers and slaters |
3 | Elementary construction occupations | Steel erectors |
4 | Plasterers | Vehicle valeters and cleaners |
5 | Steel erectors | Elementary construction occupations |
6 | Vehicle valeters and cleaners | Plasterers |
7 | Hospital porters | Metal plate workers, and riveters |
8 | Cleaners and domestics | Vehicle paint technicians |
9 | Floorers and wall tilers | Floorers and wall tilers |
10 | Metal plate workers, and riveters | Mobile machine drivers and operatives n.e.c. |
Source: Table 2, p.13, DfE (2023)
There is, of course, a degree of uncertainty about how jobs are affected. As noted above, if there is to be a reinstatement effect the impact on productivity needs to be substantial otherwise there is a risk that (some) jobs are simply replaced by automation without the concomitant creation of new jobs.
Will AI take over?
Does this mean that world of work is about to be taken over by machines and robots? There is every reason to be believe that AI, like previous generations of technological change, will continue to automate some tasks, augment or complement the skills of existing ones, and create new tasks and jobs. There is a degree of strategic choice about how AI is used. Acemoglu and Restrepo (2020) suggest that government and other societal stakeholders have a say in how AI is developed and used so that the potential for AI to raise productivity levels is increased to a level that generates strong reinstatement employment effects.
Even if the impact of AI on employment and skills might be less than is sometimes surmised, there are likely to be winners and losers. The evidence on more conventional forms of technological change amply demonstrates this to be the case. There is clearly a need to identify those at risk of losing their jobs because of AI and put in place, in good time, the requisite re-skilling and up-skilling to assist them enter the new kinds of job that AI will generate.
References
Acemoglu, D. and Restrepo, P. (2020). “The wrong kind of AI? Artificial intelligence and the future of labour demand.” Cambridge Journal of Regions, Economy and Society 2020, 13, 25–35
Acemoglu, D. and Restrepo, P. (2018). “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment.” American Economic Review
Autor, D. (2022) The Labor Market Impacts of Technological Change: From unbridled enthusiasm to qualified optimism to vast uncertainty. Washington: NBER Working Paper 30074
Autor, D., Salomons, A., and Seegmiller, B. (2021). “New Frontiers: The Origin and Content of New Work, 1940 – 2018.” MIT Working Paper, July.
DfE (Department for Education) (2023) The impact of AI on UK jobs and training. London: Department for Education
Eloundou, T., Manning, S., and Mishkin, P. (2023) GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. OpenAI Working Paper
Goldman Sachs (2023) Generative AI could raise global GDP by 7%.
PWC / BEIS (2021) The Potential Impact of Artificial Intelligence on UK Employment and the Demand for Skills. London: Department for Business, Energy and Industrial Strategy (BEIS) Research Report Number: 2021/042