Is Artificial Intelligence Coming for Your Job?

Artificial intelligence (AI) potentially generates big changes to the way we live and work. In a recent study we simulate the uptake of AI-enabled automation software in engineering and manufacturing. It shows an S-shaped adoption path – slow at the beginning and accelerating as a critical mass is reached. The digital transformation during the Covid-19 crisis implies that the process is gathering pace. The transformation will not reduce the need for people in jobs, but it will change what people do at work and the composition of skills, exacerbating the need for upgrading of skills as well as closing the digital gap when building back from the Covid-19 crisis.

AI-enabled automation is not widely used in firms    

In the late 1980s Robert Solow famously quipped that he could see computers everywhere except in the productivity statistics. Brynjolfsson and Hitt (2000) later explained the lack of productivity impact by the fact that new technology requires changes in the way production is organized and a different set of human skills to exploit the new technology to its full potential. Indeed, productivity growth did pick up in the late 1990s and early 2000s.

Today AI is ubiquitous in our daily lives, but it is not used much for automating tasks in the workplace. As was the case for computers in the 1980s, AI-enabled automation software does not simply do the same things the same way as workers did them before. Instead, the adoption of AI-enabled technology requires completely new ways of organizing work.

Reports on the use of AI in firms differ widely. The most comprehensive surveys in Sweden and the United States find that only between 5% and 10% of firms overall, and less than one percent of Swedish manufacturing firms use at least one technology classified as AI.[1] , AI will contribute to productivity and sustainability – and disruptions – only when it is widely adopted.

In Sweden, the survey also asked for the reason why firms do not use AI. Most firms did not know, suggesting that they had not even thought about it. Those that did identify a reason, pointed to costs and lack of skills as the most important obstacles.

Engineering services are enablers of technology uptake

Engineering services such as process and product design and sensor integration play an important role in bringing innovations to practical use in firms. For AI-driven automation software, engineers automate some of the tasks and the client support that they themselves used to do. At the same time, they take up new roles as developers and maintainers of AI-enabled automation software. They also continue to provide customer support of a different nature. AI thus requires a synchronized change in business model both for engineering firms and their clients.

Although engineering is among the occupations most exposed to AI, it is also among the occupations exhibiting the fastest rate of job creation and employment growth.[2]

It is hard to find data for a traditional empirical analysis of the uptake of AI and its driving forces. We do, however, know that technology adoption is a process involving the interaction between engineering and manufacturing firms – where the parties change the way they work both individually and together. With qualitative information on the mechanisms and processes of technology adoption in hand, we built an agent-based model and simulated technology uptake.

What is an Agent Based Model (ABM)? “An agent-based model is a computerized simulation of a number of decision-makers (agents) and institutions, which interact through prescribed rules.” [Farmer and Foley, 2009, p.685]. The model entails decision making rules for each type of agent, adaptive processes or learning, an interaction typology among the agents, and an environment in which they interact. In our case the agents are engineering firms and manufacturing firms. Engineering firms work with manufacturers face-to-face on site – or they develop AI-enabled software that performs similar tasks and license it to the manufacturers. Using AI-enabled automation software implies a switch to a more skills-intensive production technology. The benefit of using an ABM for this type of analysis is that it simulates the path to the final outcome; not only the starting point and the end point. Furthermore, ABMs require solid information on how agents behave, but do not necessarily require a lot of data. Finally, with ABMs one can simulate different scenarios with different assumptions on the environment in which the agents operate, making it suitable for policy analysis.

New technology creates hurdles

The baseline simulation results (Figure 1) show turbulence in the engineering sector similar to what happened during the dot.com bubble in the late 1990s. In the early stage of technology adoption, technology firms get ahead of potential users. They see new opportunities opening up and rush to create new applications ahead of competitors. However, potential users are hesitant as the existing way of doing things works just fine, the benefits of new technology are uncertain while the cost may be significant. Most early developers of new applications therefore fail.

Figure 1. Development and use of AI – a timeline

Source: Kyvik Nordås and Klügl (2021)

The early manufacturing adopters are the largest and most productive firms, who have the capacity to skill up and reorganize production around the new technology. Over time, the benefits of using AI-enabled automation software become more evident, the applications mature, and better support services become available for reluctant potential adopters. In our baseline scenario it takes almost twenty years for an existing AI innovation to become fully adopted. What holds back the uptake is the cost of changing business model and shortage of relevant skills.

Implications for policy

Our study reminds us that the rate of adoption of technology is gradual and may take decades between innovation and universal adoption. Technology uptake is particularly slow when complementary investments in skills, machinery and equipment, reorganization of production and technological alignment with suppliers and customers are required.

Even so, technology uptake is S-shaped and accelerates sharply after a critical mass is obtained. So far, the uptake of AI in firms outside the tech sector has been on the flat part of the S-curve. But one might speculate that the digital transformation that has taken place during the Covid-19 crisis may have moved us closer to the sharply rising part of the S-curve sooner than what would otherwise have been the case.

Policy makers therefore need to take action to reap the benefit of AI uptake in the workplace while minimizing social costs. For this, innovation policy is not enough. Our study shows that adoption of AI requires substantial skills upgrading on the user side as well as changes in the way production is organized. Therefore, innovation policy needs to involve tolerance for failure of early innovators and adopters and be complemented by education and training policies.

Our study also shows that the early successful adopters are the largest and most productive firms, while small and medium sized enterprises (SMEs) are laggards. This is inevitable since early adoption requires significant resources. However, access to cloud computing as well as a good regulatory regime for handling large amounts of data that could be sensitive from a privacy, cyber security and intellectual property protection perspective would help SMEs reap the benefits of AI-enabled technology.

Access to cloud computing and other technical and legal infrastructure is even more of a challenge for firms in developing countries. Investments in broadband internet and digital skills are essential for the adoption of AI-driven technology, which would support inclusive, sustainable growth.

 

References

Brynjolfsson, E., & Hitt, L. M. (2000). Beyond computation: Information technology, organizational transformation and business performance. Journal of Economic Perspectives, 14(4), 23-48.

Farmer, J.D. and D. Foley (2009). The Economy Needs Agent-Based Modelling. Nature, 460(7256), 685-686.

Kyvik Nordås, H. and F. Klügl (2021) Drivers of Automation and Consequences for Jobs in Engineering Services: An Agent-Based Modelling Approach. Frontiers in Robotics and AI (8). DOI=10.3389/frobt.2021.637125 https://www.frontiersin.org/article/10.3389/frobt.2021.637125.

SCB (2020) Artifisiell Intelligens i Sverige. https://www.scb.se/contentassets/048c2c293c404f3e899e91b844b6b9c2/artificiell-intelligens-i-sverige-2019_slutrapport.pdf

Zolas, N., Kroff, Z., Brynjolfsson, E., McElheran, K., Beede, D. N., Buffington, C., Goldschlag, N., Foster, L & Dinlersoz, E. (2021). Advanced Technologies Adoption and Use by US Firms: Evidence from the Annual Business Survey (No. w28290). National Bureau of Economic Research.

 

[1] See Zolas et al. (2021) and Statistics Sweden (2020).

[2] See for instance Eurostat on employment by occupation.