AI and Economic Growth Philippe Aghion Banque de France 18 June 2018
Introduction AI defined as the capability of a machine to imitate intelligent human behavior AI can be seen as the latest form of automation Combustion engines, electricity, and semiconductors facilitated automation in the last century AI automates tasks once thought to be out of reach, e.g. driving or medical recommendations
Some questions When AI automates the production of goods and services, do we still generate balanced growth and the Kaldor facts? How does AI affect employment and wage inequality? When AI automates the production of ideas, can it generate a *singularity*? Will AI stop the secular decline in TFP growth? Role of the State (competition policy)?
AI in the production of goods and services
A fix to the Zeira problem
Acemoglu-Restrepo Here the dynamics of I and N results from endogenous directed technical change (Acemoglu 1998, 2002, 2003, 2007; Aghion and Howitt, 1996) Under reasonable parameter values guaranteeing that innovation is directed towards cheapest factor, there exists a unique stable BGP equilibrium.
Remark However, we don t need new line creation to reproduce the Kaldor facts I will now present an alternative and simpler model which reconciles AI with Kaldor facts
An alternative framework
Thus suppose
AI and employment
AI and employment Robots and jobs: Evidence from the US. Acemoglu and Restrepo, NBER Working Paper (2017) The impact of robots on workers: Evidence from France, joint with C. Antonin, S. Bunel and S. Margolin (work in progress)
Robot Exposure and Employment Growth
Without the automobile industry
Unskilled Workers
Skilled Workers
Questions Long-term effects on employment? Effects on self-employment?
AI and wage inequality
AI and wage inequality
Questions Overall effect of AI on social mobility? Effect of AI on internal organization of firms?
AI in the production of ideas
First suppose
Then:
Can we have explosive growth? Now suppose full automation in finite time, i.e, for t >T:
What may prevent a singularity Some essential inputs to production and/or research cannot be automated Then Baumol s cost Disease is back! Ideas harder to get (?) Creative destruction with indispensable researchers Can get the paradox that AI in the production of ideas may stall the growth process
Back to the secular stagnation debate
FACT 1: Rising then declining productivity growth in OECD countries 45
Motivation: a general productivity slowdown Average annual growth rate of TFP Smoothed indicator (HP filter, λ = 500) - Whole economy 1891-2016 In % Source: Bergeaud, Cette and Lecat (2016) - See: www.longtermproductivity.com o US: one big wave over the 20 th Century, pause during Great Depression, small wave between 1995-2005 46
FACT 2: Rising then declining productivity growth in OECD countries 47
Fact 3: Rising Market Share concentration Increased concentration of market share in most industries over period 1997-2017 : In particular in IT sector with emergence of GAFAM (Google, Apple, Facebook, Amazon, Microsoft). Fraction of industries where the four biggest firms account for more than 30% of total market share, has increased from 30% to 40% ;
Market Share Concentration Autor et al (2018) Manufacturing Sector Finance Sector
Fact 4: Declining Average Labor Share Labor share has decreased from 65% to 58% since 2007: Mainly driven by reallocation across firms as opposed to a declining labor share within firms
Declining labor share Source : Karabarbounis and Neiman, 2014
Fact 5: Non Increasing Investment (Gutierrez-Philippon)
Fact 6: Increase in Average Mark Up Essentially driven by between-firm reallocation (Baqaee-Farhi, 2018 vs De Loecker and Eeckhout, 2017) Average mark-up has increased but the average change in mark up has been negative over the recent period
Explaining these facts with one model: Ongoing work with A. Bergeaud, T. Boppart, P. Klenow, and H. Li Schumpeterian growth model with super-star waves Final good produced with continuum of intermediate inputs Growth results from quality improving innovations on the production of intermediate inputs There are high and low ability entrepreneurs
And a low-ability entrepreneur makes no profit on a peripheral line
Equilibrium innovation and growth Innovation by high-ability entrepreneurs on peripheral lines will be partly discouraged Innovation by low-ability entrepreneurs on peripheral lines will be fully discouraged Thus productivity growth will first increase when super-star firm appears..but then it will decrease
Explaining the facts This model can explain all the above facts, in particular: Increasing then decreasing productivity growth Average mark-up increase but without a positive average change in mark.even without entry barriers but even more with entry barriers
Role of the State Competition and IP policy AI helps develop platforms and networks which in turn threaten competition Data access may act as an entry barrier for creating new competing networks Networks can take advantage of their monopoly positions to impose large fees on market participants which in turn may discourage innovation by them
Role of the State Thus a new take on competition and innovation.in relation to AI!