From Automation to Adaptation: Jobs at Risk of Exposure to AI in the UK
Artificial intelligence (AI) is advancing at pace. In its wake the world of work is being transformed. With the emergence of an AI application such as Chat-GPT, jobs previously thought immune to automation now appear to be at risk. Or are they? Because there is a great deal of hype surrounding AI and its impact on jobs, there is a need to take a step back and consider the evidence. Analysis based on a new, innovative measure of the risks posed to various kinds of jobs in the UK by AI reveals that those occupations most exposed to technologies such as Chat-GPT have shown no abrupt behavioural changes six months after the tools’ initial release.
Although the fear of AI-induced job displacement is understandable, the current evidence from the UK suggests that AI tools such as Chat-GPT are not yet leading to job losses. The initial findings from the United Kingdom (UK) indicate that, following the launch of Chat-GPT, there have been no sizable changes in the labour market trends, particularly for jobs deemed susceptible to these type of AI tools. That said, the rapid pace of development in GPTs (Chat Generative Pre-Trained Transformers) means that there may yet be transformations in the labour market.
What AI tools are we talking?
The integration of natural language processing and machine learning has given rise to sophisticated new chatbots such as Chat-GPT, Bard and Jasper to name a few. These AI-powered chatbots effectively interact with users, interpret their queries, and deliver tailored responses. They are also generative. You could ask for a country and western song delivered in the style of an opera singer and it will produce a possibly passable facsimile. All this raises questions about the potential displacement of human workers.
The swift adoption of AI tools such as Chat-GPT is expected to contribute to significant transformations in the workplace (Cedefop, 2023). But how can evidence, to prove or deny such assertions, be produced in a timely manner such that policy makers are suitably forewarned? Relying on traditional methods of information gathering, such as surveys of employers or workers, takes too long. There is a danger that the evidence is dated even before it is published (Cárdenas, 2020; Eurofound, 2021). And it tends to look backwards: what happened yesterday rather than what is likely to happen tomorrow. There is a solution to hand.
If handled suitably, real-time online data provides the means to promptly assess labour market trends. Online job advertisements (OJAs) offer rapid access to labour market information. This means that data can be accessed as soon as it is posted to provide up-to-date information on the current state of the labour market.
Measuring AI’s impact on jobs in the UK
When Chat-GPT launched a few months ago, researchers from Open AI published a paper analysing the possible impact of large language models or LLMs on the United States’ labour market (Eloundou et al., 2023). A list was compiled of the occupations with the greatest exposure to GPTs. This was defined with reference to a reduction in the time needed to complete a task by at least 50 per cent as a result of a GPT.
Drawing upon the advertisement data collected by IER since 2019 it is possible to provide an estimate of the impact of GPTs on employment in the UK.
Jobs at risk
The paper by Eloundou and colleagues identified 15 occupations – classifed according to the US standard occupational classification (US SOC) – particularly susceptible to the impact of GPTs. These can be translated into their UK SOC equivalent using Cascot software. They are:
- Survey Researchers – Business and related research professionals (UK SOC code 2434)
- Animal Scientists – Biological scientists (2112)
- Climate Change Policy Analysts – Social and humanities scientists (2115)
- Blockchain Engineers – Programmers and software development professional (2134)
- Web and Digital Interface Designers – Web design professionals (2141)
- Financial Quantitative Analysts – Finance and investment analysts and advisers (2422)
- Tax Preparers – Taxation experts (2423)
- Mathematicians – Actuaries, economists and statisticians (2433)
- News Analysts, Reporters, and Journalists – Newspaper and periodical journalists and reporters (2492)
- Public Relations Specialist – Public relations professionals (2493)
- Proofreaders and Copy Markers – Authors, writers and translators (3412)
- Accountants and Auditors – Book-keepers, payroll managers and wages clerks (4122)
- Correspondence Clerks – Records clerks and assistants (4131)
- Clinical Data Managers – Database administrators and web content technicians (3133)
- Court Reporters and Simultaneous Captioners – Typists and related keyboard occupations (4217).
Figure 1 shows the share of jobs which are potentially sensitive to the widespread use of GPTs derived from UK online job advertisements. The figure includes two vertical lines representing key points in the UK’s recent labour market history:
- April 2020 marks the beginning of pandemic restrictions; and
- December 2022 represents the launch month of Chat-GPT and other AI tools.
Over the four-year period depicted in the graph, there has been a downward trend in the overall share of jobs falling into this category. Between February 2019 and February 2021, most occupations sensitive to AI experienced a significant decline, dropping from 10.5% to 8.0%. It’s noteworthy that in the subsequent four months, there was a rapid resurgence, with the percentage surpassing 10.5% by June 2021. Following this peak, the distribution of these occupations reverted to its previous pattern.
For the time being there is little evidence, based on online job advertisement data, that new forms of AI such as GPTs, have had substantial impact on occupational employment. The demand for people to work in jobs susceptible to GPTs appears to be holding up for the time being. But this may well change. After all, six months is a short time period over which to identify the impact of GPTs on employment. Nevertheless, it provides a benchmark against which to compare future change.
Final Thoughts: GPTs’ Current and Future Impact on Jobs
With sophisticated chatbots such as Chat-GPT and other similar AI tools arising from the convergence of natural language processing and machine learning, there is concern about the potential displacement of human workers. An indicative list of occupations or job roles that are susceptible to being replaced by LLMs has been presented. This serves as a baseline indicator of the potential impact of GPTs on the future demand for jobs and skills. To date, those working in occupations most exposed to these technologies have not shown any significant dip in demand for their labour. This implies that, for now in their early use, these tools may be enhancing productivity by reducing task completion time, rather than replacing jobs altogether. It is, however, important to keep monitoring the impact of GPTs on labour and skill demand. Further information will be provided on the LMI for All portal in due course about how demand for people to work in jobs susceptible to GPTs is changing.
Cárdenas, J. (2020) ‘A Web-Based Approach to Measure Skill Mismatchesand Skills Profiles for a Developing Country: The Case of Colombia’, Journal of Chemical Information and Modeling, 110(9), pp. 1689–1699. Available at: https://www.jstor.org/stable/j.ctv1g6q8dv.
Cedefop (2023) Ready, steady, go! Available at: https://www.cedefop.europa.eu/en/publications/9179.
Eloundou, T. et al. (2023) ‘GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models’. Available at: http://arxiv.org/abs/2303.10130.
Eurofound (2021) Employment and labour markets Tackling labour shortages in EU Member States. Edited by P. O. of the E. Union. Luxembourg. Available at: https://www.eurofound.europa.eu/publications/report/2021/tackling-labour-shortages-in-eu-member-states.
 “Generative Pre-trained Transformer”. It is an acronym that describes the underlying language model used in the conversation. GPT is a transformer-based language model architecture that has been pre-trained on large amounts of text to learn language patterns and structures.