Numerous reports have highlighted the impact of generative artificial intelligence(GenAIS) systems on productivity. In fact, many players are now convinced of the benefits of generative AIS. However, behind the hopes that these technologies raise, and the apparent evidence of these gains, lies a complexity that questions this presupposition that seems to be taking root in people's minds.

According to a Harvard study published in September 2023, GenAIS could indeed deliver a 40% increase in productivity, but - and this is an important point - only for highly qualified profiles, and only if the GenAIS, in this case GPT-4, is used for tasks within its capabilities, and the people concerned are trained in its use. This increase in productivity also applies to creative tasks related to innovation. By contrast, using GPT-4 to solve business problems leads to a 23% drop in performance.

Another study conducted in April 2023 (revised in November of the same year) by MIT and Stanford, shows that the productivity of the customer service employees of a digital services company (based in the Philippines), had jumped by 14% on average thanks to generative AI. It also highlights that the least qualified employees were the main beneficiaries with a 34% increase in their productivity. The study concludes that while access to GenAIS boosts productivity, the increase is nonetheless highly heterogeneous among employees.

Finally, in January 2024, Barclays Research and IBM Institute for Business Value, published a study asserting that there are "good reasons to be optimistic about the potential of gen AI to drive growth", provided an enabling environment is created in terms of policies and regulations at the level of the company itself, an industry sector or regulators. Furthermore, the paper points out that these technologies could potentially generate productivity gains if workers were to move into service sectors. Generally, speaking, potential gains are conditional on many factors such as policies, standards, costs, employee skills, time management, or the use of AIS generated "as a complement to work rather than a substitute".

As we can see, behind the widely-held claim that AIS-generated productivity increases lie a vast and complex subject that requires in-depth reflection. It is crucial to get away from the ambient noise, so that private and public structures alike do not become embroiled in policies that could prove dangerous.

The studies cited above, and many others, do not assert the factuality of the productivity gains to be expected, but envisage them in terms of potentials subject to particular conditions which it is essential to consider carefully. To approach the subject in too general a manner is to ignore the conditions necessary for these potential gains, and thus to place the debate at a level of abstraction unsuited to the specificities of companies.

There are a number of lessons to be learned from these studies, therefore, if we are to be able to consider serenely and seriously the productivity gains associated with GenAIS.

·      First and foremost, productivity cannot be reduced to its time-saving dimension. Productivity is a complex concept, which certainly includes a time component, but also the resources available/necessary, the quality of production as much as the quantity produced, the ratio between resources committed and results obtained, comparative work to measure the real gain generated, performance analysis and many other factors. To assess potential productivity gains, it is therefore essential to start from a solid definition of productivity.

·      The potential productivity gains of GenAIS are not homogeneous across all users, business sectors, countries, company functions, use cases, tasks to be performed, or GenAIS themselves. A GenAIS used for tasks for which it is not specifically adapted, or by untrained users, could harm performance and therefore productivity.

·      We do not have enough hindsight to settle the debate on the potential productivity gains attributed to GenAIS, especially if the question is posed in absolute terms. The potential is therefore hypothetical at this stage, and is based on elements that are often scattered and inconsistent with each other.

·      To speak of productivity gains from GenAIS is to place the debate at too high a level of abstraction, preventing a pertinent understanding and analysis of the subject.

·      The potential productivity gains of GenAIS must not obscure the side-effects on skills, on the mental workload of certain categories of professionals, the loss of creativity and therefore the impact on innovation, induced costs, or career development.

In the final analysis, it would seem premature to assert that GenAIS will lead to productivity gains. Enthusiasm for these tools and their potential should not preclude more in-depth reflection on the positive and negative impacts of these systems on company results, but also on the workers themselves.

A lack of reflection could lead to major side-effects, which could ultimately produce the opposite effect to the one hoped for. By postulating the reality of productivity gains due to GenAIS without having taken the time to study them seriously in all their dimensions, companies run the risk of setting themselves utopian objectives and implementing unsuitable, if not counterproductive, strategies.

To avoid these potentially costly mistakes, it is essential to :

·      Start with a solid definition of productivity.

·      Implement precise evaluation processes.

·      Anticipate costs, both financial and in terms of internal organization, and normative issues (ethical and legal).

·      Determine precisely the use cases and tasks for which a GenAIS is relevant.

·      Train employees in the use of GenAIS deployed within the company.

·      Carefully select relevant GenAIS.

·      Measure the potential negative impact of both the use of GenAIS and a productivity policy, particularly on employees' mental health.

Further readings

Meredith Somers. How generative AI can boost highly skilled workers’ productivity. MIT Management Sloan School, October 19,2023. https://mitsloan.mit.edu/ideas-made-to-matter/how-generative-ai-can-boost-highly-skilled-workers-productivity

Fabrizio Dell’Acqua et al. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Technology & Operations Management Unit Working Paper No. 24-013. September 27, 2023. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321

Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond. Generative AI at Work. National Bureau of Economic Research Working Paper Series, Working Paper 31161. April 2023, revised November 2023. https://www.nber.org/papers/w31161

Christian Keller et al. AI revolution: productivity boom and beyond. Barclays Research with IBM Institute for Business Value, Impact Series 12, January 11, 2024. https://www.ib.barclays/our-insights/AI-productivity-boom.html

‍