Productivity growth through the lens of AI
Examining the behaviour of innovating firms is a good way to measure the impact of artificial intelligence on productivity and labour, writes Francesco Venturini, associate professor in economics, University of Perugia.
The emergence of generative artificial intelligence is a catalyst for the fourth industrial revolution. The impact of AI as a general-purpose technology is all-encompassing, permeating many sectors of society. But the effective integration of AI necessitates a collaboration between creators and users to foster a culture of innovation among adopters.
Economists and policy-makers have primarily focused on identifying the incentives that drive the adoption of AI and the implications for work organisation and labour. AI plays a pivotal role in enhancing production efficiency, reducing costs and facilitating the exploration of novel methods for producing goods and services. It also enables the expansion of product offer.
Early studies indicate that the implementation of AI increases firm productivity and curtails labour demand, but does not significantly impact wages. This suggests that there may be an unprecedented decoupling between productivity improvements and the labour market.
Understanding the impact of AI
There are two main characteristics that set AI apart from other significant innovations, making economic analysis of this new technology particularly challenging. First, AI has an intangible nature, which distinguishes it from tangible (hardware) innovations such as trains and computers that drove economic progress in the last century.
Second, AI enables machines incorporating these systems to perform cognitive tasks, marking a significant departure from previous generations of disruptive technologies like CAD-CAM, ICT and industrial robots. These earlier technologies primarily aimed to reduce human effort in the workplace, replacing low-skilled workers or those engaged in highly routinised tasks.
To understand the role and effects of AI in the economy, it is important to examine the behaviour of firms that adopt and produce these technologies.
Technology adoption is typically measured through thematic surveys or inferred by analysing a firm’s hiring of AI experts, using information obtained from online job postings. While these sources offer rich information, concerns arise regarding the representativeness of the sampled firms (for the former) and selection bias as they may only cover firms active on the web (for the latter). Furthermore, the information provided by these sources is limited to a single time point, making it challenging to compare data over time and across different countries.
Technology production can be deduced from similar sources of information. The success of innovative activities can also be determined by examining the patents that firms apply for to safeguard their inventions. Patent information is widely accessible and can be analysed using sophisticated big data techniques, which means the content of innovations as well as the characteristics of the firms and inventors involved can be examined.
However, it is important to acknowledge that patent data have certain limitations. First, even if an innovation meets all the requirements for patenting, a firm may choose not to seek patent protection to avoid disclosing their priority algorithm. Furthermore, not all AI systems possess the necessary qualifications for patenting. And even when an AI system is successfully patented, it represents only a fraction of a much larger cluster of innovations that are developed by the firm in collaboration with its customers to meet specific needs.
These factors render the traditional categorisation used in economic literature, which distinguishes between producers and users of new technology, outdated.
Productivity growth and AI
The use of patent data has proven valuable in evaluating the impact of AI on the productivity growth of European firms. It has been instrumental in determining whether AI contributes to the widening inequality between regions that are more or less specialised in their production.
In a study conducted in 2023, a comparison was made between the productivity growth of firms developing AI and companies with similar characteristics not specialised in the development of AI. The findings revealed that AI firms experienced a remarkable 13.7% increase in productivity growth compared to non-AI companies. Furthermore, companies that were falling behind the technological frontier enjoyed an additional premium of 6.4%.
Another 2023 study by Antonio Minniti, Klaus Prettner and myself delves into the impact of AI development on the dynamics of income distribution across European regions from 2000-18. It finds a significant decline in the regional labour share, with an 8% reduction observed for every doubling of the regional stock of AI innovations.
Interestingly, the study reveals that high-skilled workers seem largely unaffected by the development of AI. However, the labour share of medium-skilled individuals experiences a notable decrease of 3%, while low-skilled workers bear the brunt with a staggering 9% decline. This trend may be attributed to a lower sharing of rents associated with AI innovations, which could be smaller than those witnessed for previous technologies. Furthermore, it is plausible that a reallocation of factors has been induced by the rising market share of low-labour intensive firms or a shift in firm labour demand away from certain skill types.