Simulating the impact of digital and AI on jobs and economy


The NumJobs project was created in 2017 by Jean-Daniel Kant and Gérard Ballot. Since October 1, 2018, it has been the result of a collaboration between Sorbonne University Sciences, the University Panthéon-Assas and Pôle Emploi.

Issues and objectives

The aim of the NumJobs project is to provide to firms and organizations an innovative tool to simulate the impact of digital and AI technologies on economies and societies.

Digitalization, automation, robotization, rise of artificial intelligence (AI), as many movements and factors which constitute the digital revolution, often founded on innovations of rupture. The bigger issue (or fear depending on point of view) here concerning these digital innovations is that of the impact on employment. Are they sources of job creation, or will they involve destruction – possibly massive – of employment ? If there is job creation, is it through new types of occupation ? And which actual professions would be threatened of extinction?

Moreover, beyond the question of creation or destruction, how this digital revolution will modify the nature of work, the actual tasks, the work conditions, the place of work in the life of the individuals and the society ? Thus, the impact of digitalization about employment not only concerns the economy but also the whole society.

Innovations lead destructions of existing employment and creations of new jobs. Among them, disruptive innovations create new sectors, and could trigger waves of innovations that are potentially engines of wealth creation by diffusion of purchasing power to the other sectors of the economy. They have however destroying effects on certain types of jobs. Previous industrial revolutions have not induced a net negative effect on the number of jobs. Productivity and new products have raised incomes and demand for products. The key issue we aim to tackle is whether the digital technology has similar effects or may involve negative effects on the number of jobs and their structure.

Why current studies are insufficient ?

This question is the object of a great interest these last years and several studies sought to measure the impact of digital on employment. The first study is Frey & Osborne (2013) who estimates that 47% of employment in the USA and 35% in the United Kingdom have a high probability of automation within 20 years. Applied to the French case, 42% of employment at the risk (Roland Berger, 2014) is obtained. However these studies relate to the occupations , whereas an analysis – finer – by tasks is necessary to give an account of the heterogeneity of the occupations. A study of OECD (Arntz and al., 2016) based on tasks but at an aggregate level (?) ( considers the risk with 9% of employment (for the USA and on average in 21 OECD countries), while another of France Strategy (Ru, 2016) is based on an investigation DARES into the work conditions and estimates the risk at 15% for France.

A recent report of the COE (the French Council of Orientation for Employment) recently synthesized these studies (COE, 2017) and underlines their limits rightly. They are focused on the potential destruction of employment due to digitalization and automation, but do not allow to measure the creation opportunities related to these technologies. They are based on current technologies, and do not take into account the evolution of the innovations. The social, institutional and economic brakes - that can limit automation - are also not considered. The COE proposes its own study while being based on the investigation work conditions of the DARES, in order to work on the level of the individual terms of employment, and is also interested in employment likely to evolve (not only their disappearance). Their conclusion for France is that 10% of existing jobs are threatened, and that the half of this employment is likely to evolve in an important way.

These studies highlight the difficulty of the subject. Depending on the method, they lead to rather different results, even if the last seem to converge towards a risk of disappearance of employment between 9% and 15%, and 50% of employment that would be concerned with deep change. None of these studies gives the impact on unemployment. Concerning the upgrading capability of employment, the COE gives results on the occupations, but not on their contents (tasks). In addition, the studies are static, in the sense that it are not able to study the evolution of these innovations. Lastly, these studies take primarily the point of view of the individuals (work conditions) and not that of companies (in particular their organization, their strategy, economic constraints, etc).

Our approach

Our analysis is that a tool is lacking to model and simulate the impact of the digital innovations on employment and work. With NumJobs, we aim to build such a tool, the first to cover all the necessary dimensions of this problem. This tool will allow to :

  • Simulate the impact of a technology on work (organization, tasks,…) and employment (creations of new jobs, destruction, unemployment,…)
  • Take into account all the main elements of the job market, with its two sides: individuals and firms
  • Account for all the economic aspects, thanks to a macroeconomic looping of the job market
  • Model the processes of innovation (processes and products) , via a competition between companies, in order to account for the dynamics of digital innovations, and to measure the evolution of their impact over time

Based on our experience in the field of simulation using multi-agent systems applied to labor market and innovation, we can now propose a method to build such a tool successfully.

We propose to base our tool on 2 models of simulation multi-agents we have developed for several years:

  • WorkSim is a multi-agent model of labor markets, created by J.-D. Kant and G. Ballot. These agents (individuals and companies) make decisions based on microeconomic analyses, under bounded rationality. For example the decision for a company to create or not a new job distinguishing open ended contracts and short fixed duration contracts,according to the request; or the decision for an unemployed to apply or not with a job offer, or to leave the job market. WorkSim scales down the French Labor Market at a scale factor of 1/2000 and is calibrated on more than hundred important variables such for example as the unemployment rate or the wages by broad occupation (coming from INSEE, of the DARES, etc.). It reproduces the important stylized facts of the French job market and has been published in international publications (e.g. Goudet and al., 2016). It already allowed us to carry out the first quantitative evaluation ex-ante of the recent law work “El Khomri” and several other policies of the job market (reduction in charges, contract of generation, etc.).
  • A multi-agent model of endogenous growth with multiple firms in competition to study the dynamics of innovation in new consumer goods and quality for each one of these new goods (allowing new activities, with associated innovations of process), is being developed per G. Ballot and T. Huynh (e.g. Ballot & Huynh, 2016). Certain goods may disappear, high qualities eliminate from low qualities when the standard of living increases in an endogenous way, the worker-consumers also being represented individually. The model of production rests on a triplet: competences (R & D, to produce), tasks (organization of work) and alliances in R & D between companies, which are a source of diffusion and accumulation of competences. Moreover, for the first time, a computational model of the innovation is buckled by an endogenous competition of the companies which position on a market differentiated (dynamically) in goods and qualities, but can also go bankrupt. The nature of each market emerges at the same time as a macroeconomic state of growth of long run.



Jean-Daniel Kant

Head of NumJobs project

Associate Professor with Habilitation at Sorbonne Université - Sciences

Multi-Agent Systems Team (SMA) - Computer Science Laboratory of Paris 6 (LIP6)


Gérard Ballot

Emeritus Professor in Economics at Université Panthéon-Assas - Paris II

Paris Center for Law and Economics (CRED)



Jean-Daniel KANT

Email (click to send): Jean-Daniel[dot]Kant[at]lip6.fr

Phone +33 1 44 27 88 05

Sorbonne University - Sciences
LIP6 (Laboratoire Informatique de Paris 6)
4, place Jussieu
75005 Paris