Shrinking working-age populations and declining productivity growth are creating significant challenges for economists, policymakers, and managers across the industrialised economies. New technological innovations and their wide adoption in society will be essential to address these challenges. Artificial intelligence (AI) is possibly the most promising technology currently in development but also one that creates considerable new challenges.
AI is a technology that significantly reduces the cost of predictions (Agrawal et al. 2018). Thus, anywhere predictions are used will be transformed by AI – job tasks, occupations, and the way business are run in general. AI is expected to impact economic growth, employment, and innovation throughout the economy, making it imperative to understand how and where AI is being used.
Demand for AI skills across the economy
The scarcity of data on the diffusion of AI has been a primary reason why so little is known about the current state of AI in the real economy. Only recently, a few studies managed to open this black box and give us some early insights into how much and for which purposes companies use AI. Earlier studies documented a fast spread of specific AI techniques – deep learning and machine learning – across computer science subfields (Klinger et al. 2018) and across industries and research jobs (Goldfarb et al. 2019). A few papers focused on the impact of AI on transforming job tasks (Felten et al. 2019, Webb 2019) and found that AI technology mostly affects high-skilled occupations.
Our recent analysis (Alekseeva et al. 2020) adds to the existing evidence by showing AI diffusion in the overall economy and across all industry sectors and occupation groups in the US. To track the diffusion of AI in the US, we used data from the online job postings database provided by Burning Glass Technologies, a job-market analytics company. These data contain nearly the universe of job vacancies posted online by firms in the US over the time period 2010–2019 and give us detailed information on vacancies that request AI skills.
In Alekseeva et al. (2019), we document that demand for AI skills, measured as the number of posted vacancies as a share of all online job vacancies, has been rapidly increasing over the analysed time period, and the growth has accelerated since 2015. Between 2010 and 2019, the absolute number of job postings looking for AI skills grew by a factor of ten in absolute numbers and by a factor of four as a proportion of total job postings (Figure 1, panel a). This trend lies in marked contrast with that for other computer-related skills (Figure 1, panel b). For instance, the demand for other computer skills, while much larger, was remarkably stable over time as a proportion of all vacancies and only grew with the general hiring growth in a good economy.
Figure 1 Growth in demand for (a) AI and (b) computer skills in online job postings
(a) Demand for AI skills
(b) Demand for computer skills
The demand for AI skills has exploded across all industry sectors and occupations. Predictably, the demand is highest in the IT industry and computer occupations but is not limited to them (Figure 2). AI skills are also in high demand in professional services and finance. Even administrative services, manufacturing, and agriculture showed significant growth in the relative demand for AI skills (Figure2, panel a).
Figure 2 Growth in demand for AI skills across (a) industries and (b) occupations
(a) Demand for AI skills across industries
(b) Demand for AI skills across occupations
By occupation, the demand for AI specialists – while highest among computer and mathematical occupations – is increasing in a broad range of occupations. Architecture, life and social sciences, management, legal, and business occupations all experienced significant growth in the proportion of vacancies demanding AI skills (Figure 2, panel b). This demand for AI across a range of industries and occupations shows that AI is indeed a technology with potential application in many spheres of economic activity.
Effect of AI in labour markets
The increasing diffusion of AI across the economy raises real questions about how this technology will affect the demand for other skills. Concerns that new technologies can displace human labour have existed for a long time, but they have reached a new level with recent advances in AI technology and its surpassing human capability in a range of cognitive tasks.
Early evidence, however, generally offers a positive view. Felten et al. (2019) showed that occupations affected by developments in AI experience a positive wage growth and AI does not reduce overall employment in these occupations. Webb (2019) finds that AI technologies have the potential to decrease wage inequality. Alderucci et al. (2019) show that innovations in AI are associated with growing employment in the innovating firm, though also with increasing within-firm wage inequality.
In Alekseeva et al. (2019), we show that jobs requiring AI skills command a wage premium vis-à-vis jobs requiring other skills. On average, jobs requiring AI skills offer an 11% wage premium compared to similar jobs that do not require knowledge of AI. This wage premium, however, varies widely across industries and occupations (Figure 3).
Figure 3 Estimated wage premium for AI skills by (a) industry and (b) occupation
(a) Estimated wage premium for AI skills by industry
(b) Estimated wage premium for AI skills by occupation
The largest premium for AI skills is offered by vacancies in the administrative and support services sector. In this sector, a job posting demanding AI skills offers an almost 18% higher salary than a vacancy with no demand for AI. The IT, as well as finance and insurance industries, offer a premium of around 11%. Manufacturing offers a 9% premium and professional services an 8% premium.
A similar analysis of the wage differentials across occupations shows that AI skills are also valued differently in different occupation groups. Managerial occupations offer the highest premium – a job vacancy requesting AI skills offers, on average, wages that are almost 11% higher than a position requesting a comparable portfolio of other skills. This is notably higher than the approximately 9% wage premium in the computer and mathematical occupations most often associated with AI. Life, physical, and social science as well as business and financial-operations occupations offer close to a 7.5% premium, and architecture and engineering a 5% premium.
The significantly higher premium for AI skills in management compared to other occupations suggests that the real shortage is of managers who are able to create and capture value with AI. If AI indeed is a general-purpose technology, then the real productivity gains will only come slowly over time and only as a consequence of deeper changes in how businesses operate. Hence, the evidence suggests that AI is at least as much a managerial challenge as it is a technological challenge.
What happens with wages offered to the other employees in firms that hire AI specialists? One the one hand, adoption of AI (and thus demand for AI skills) can require investment in high-skilled non-AI tasks and in employees who will complement AI technology. Therefore, non-AI salaries could be higher in firms with vacancies requiring AI skills. On the other hand, AI can automate some well-paid jobs and thus firms will need to invest in skilled human capital less, leading to lower non-AI salaries.
Our study finds that firms with a higher proportion of AI vacancies tend to also pay higher wages to non-AI employees. This complements the findings of Acemoğlu et al. (2019) that firms demanding AI skills change the composition of occupations within the firms by hiring less for jobs at risk of replacement by AI.
Given the importance of AI, more research is needed to inform effective government policies and managerial practices. Evidence of a fast diffusion of AI in the economy and its already apparent impact on labour markets show how fast governments and companies must act to keep abreast of such changes.
Acemoglu, D, D Autor, J Hazell and P Restrepo (2019), “AI and jobs: Evidence from online vacancies”.
Agrawal, A, J Gans and A Goldfarb (2018), Prediction machines: The simple economics of artificial intelligence, Brighton MA: Harvard Business Press.
Alderucci, D, L Branstetter, E Hovy, A Runge and N Zolas, (2020), “Quantifying the impact of AI on productivity and labor demand: Evidence from US census microdata”, American Economic Association Annual Meeting, 3 January.
Alekseeva, L, J Azar, M Gine, S Samila and B Taska (2020), “The demand for AI skills in the labor market”, CEPR Discussion Paper DP14320.
Felten, E W, M Raj and RSeamans (2019), “The occupational impact of artificial intelligence: Labor, skills, and polarization”, working paper, SSRN 3368605.
Goldfarb, A, B Taska and F Teodoridis (2019), “Could machine learning be a general-purpose technology? Evidence from online job postings”, working paper, SSRN 3468822.
Klinger, J, J C Mateos-Garcia and K Stathoulopoulos (2018), “Deep learning, deep change? Mapping the development of the artificial intelligence general purpose technology”, working paper, SSRN 3233463.
Webb, M (2019), “The impact of artificial intelligence on the labor market”, working paper, SSRN 3482150.