fabernovel loader

Nov 6, 2017 | 13 min read

Economy

AI and Inequality - the Real Threat

Tom Morisse

Research Manager


FABERNOVEL
The end of employment will not happen. The great risk that AI casts over the economy is to amplify the already strong tendency of the digital revolution to increase the inequality between companies and between individuals. There is an urgent need to react.

Starting point: widening inequality

 

Thomas Piketty’s bestseller highlighted the increase in income and wealth inequality in recent decades in many countries after the first half of the twentieth century, which had led to a reduction in inequalities never seen in the history of humankind – a happy consequence of very unfortunate shocks, the sum of the two world wars and the crisis of 1929.

To put it simply: the rich are capturing a good deal of economic growth.

The evolution is striking: the fiscal income share received by the richest 1% is increasing everywhere – except in France, because the 2007-08 crisis erased the late gains made in the 2000s. Even the Nordic countries, known for their lower level of inequality, are affected – the case of Finland. The country which best crystallizes the current trend is of course the United States, where the fiscal income share captured by the richest 1% has almost tripled in 40 years.

 

Even if everyone has their own philosophical and political vision about inequalities (are they desirable, are they right, what level is acceptable?), there are at least 4 good reasons to worry about their regular progress:

  1. The extent of inequality stretches at both ends of the spectrum: the big problem is that in recent years the standards of living of the poorest have flowed back. Thus, in the United States, the median salary in real terms peaked in 1999! And the volume of households living with $15,000 a year or less, measured in percentage of the population or in absolute figures, has increased since 1999 – a historical low.
  2. Inequality affects economic growth, as the OECD has shown, for example.
  3. The risk of strong social and political tensions increases.
  4. Nobody would like to go back to the world of late-nineteenth-century annuitants dreaded by Piketty, due to a greater fiscal income share derived from capital and a concentration of capital, and in which to choose your wife or spouse was the only real stake in life.

The evolution of the technological landscape during the second half of the twentieth century and the beginning of the twenty-first is one of the causes of this growing inequality; analyzing its impact can therefore shed some light on the possible effects of artificial intelligence on inequality in the future.

 

1// At the individual level: growth is biased towards the most qualified workers

 

Many labor economists, including MIT Professor David Autor, emphasize the polarization of jobs (both in employment and wages) according to skills.

With many graphs of this sort (in this case, taken from the article Why Are There Still So Many Jobs? The History and Future of Workplace Automation) :

On the x-axis are US occupations ranked from left to right according to their average wage in 1979, and attention is paid to the evolution of the average wage (top graph) and the share of these occupations in employment (bottom graph).

What we observe is that for both graphs, times have been hard for middle-skilled occupations especially as regards the evolution of employment levels. Economists put forward the concept of skill-biased technical change to explain the trend: new digital technologies have increased productivity, hence the value in the labor market (read: the demand) of the most highly-skilled employees, resulting in a rise in wages and employment for them. The occupations in the middle of the ranking, relatively skilled but still more routine, those of the skilled worker or the clerk (responsible for general administrative tasks) and which consisted ultimately of processing and transmitting information, have receded. Those that managed to find new jobs entered the non-routine or hardly-automatable manual occupations, for instance in the food service industry.

 

A clarification is needed to avoid any confusion for the reader: when we speak of polarization, we are talking about a gradual movement, not an end state. Middle-skilled occupations remain very important in the total employment stock of developed countries.

This thesis of polarization is also applied to the EU countries; I am skeptical, however, about its ability to account for recent changes, and it is perhaps most valid in the United States (and in terms of methodology, it all depends on how low, middle and highly-skilled occupations are delineated). In the three major European economies of Germany, France and the United Kingdom, between 2011 and 2016, occupations in the 3 most skilled job categories (« Managers », « Professionals », « Technicians ») were responsible for  65% to 80% of job creation. And among the same 3 countries, over the 2006-2014 period, there is no common trend on the skill categories for which wages have risen the quickest or the slowest.

Overall, what seems to me undeniable is not polarization; it is that jobs require more and more skills, and that the situation is becoming therefore very difficult for the unskilled, in terms of access to employment and wages.

 

Lessons for AI

Some will see in the analysis of polarization a pattern feared in the advent of artificial intelligence: on the one hand highly-skilled engineers being paid millions of dollars each year, and on the other hand pauperized algorithm trainers.

If you have read our last article on the permanence of employment (obviously unexpected for some), you may know that a great number of job creation mechanisms are being missed by this focus on the jobs most directly related to the AI ​​boom.

Looking at the big picture, my belief is that, both in the transformation of existing jobs and in the creation of new ones, AI will permeate these occupations just as computing did previously. And that the skill divide will once more rest upon one’s ability to master and integrate in workflows the new tools that will undoubtedly emerge.

 

2// At the company level: the « winner takes all » trend

 

It is at this level that we can fear the most. Two recent papers titled Are US Industries Becoming More Concentrated? and The Fall of the Labor Share and the Rise of Superstar Firms underlined alarming findings:

  1. Over the last two decades, business concentration has increased in 75% of the US economy sectors – a trend that has also played out in Europe.
  2. Differences in productivity and wages are thus driven by differences between firms within the same sector and not between sectors.
  3. Alongside the decline of the action of governments or courts to fend off concentrations, and in conjunction with globalization, information technologies provide a part of the explanation. Network effects and the rise of digital services, with their marginal reduced cost, tend to concentrate market shares in one or a handful of firms: the winner takes all, or at least the winner takes most. The sectors where technological progress (as measured by total factor productivity) was the fastest are the most concentrated.
  4. Innovation can allow new entrants to take on large market shares, but technological barriers thus created subsequently limit competition. And some studies point to the fact that innovation that has become more costly and that it now relies on economies of scale favoring the largest companies on the market. In the long run, innovation has become more concentrated, as proves this graph depicting the evolution of the share of patents owned by the four largest firms in all sectors of the US (case study from Grullon and al.):

To illustrate this trend of a divergence of productivity between firms, influenced by the digital revolution, we can use the comparison we draw in our GAFAnomics studies between revenue per GAFA employee and that of older companies. Significant salary differences result.

However, the explanation of the phenomenon is not based solely on the advantage built up by the “superstars”, as if the disparities could only be explained by a handful of incredibly productive giants – the way of thinking proposed by this Harvard Business Review article. An OECD study, The Great Divergence(s)also shows that the least productive firms are falling behind – the 90-50 line tracks the gap between the most productive firms and the middle of the statistical distribution, and the 50-10 line tracks the gap between the middle and the least productive firms:

 

Lessons for AI

The previous consequences of new technologies seem to be amplified by the development of AI. On the one hand because the importance of data and computing power requirements in the learning process of the AI models will lead to a concentration of suppliers of generalist technological solutions. On the other hand for the complexity / cost of deployment as well as the scale effects in the return on investment could be even higher than in the previous phases of the digital revolution, thus favoring larger companies.

 

3// At the crossroads between individuals and companies: the fragmented labor market

 

In recent years, Uber’s explosive growth and the simultaneous explosion of the number of independent drivers has led to fears of a fragmentation of the labor market, accompanied by a reduction – if not a disappearance – of employment-related benefits. This tendency of the « Uber for X » has – for the moment – dwindled and the new digital freelancers remain confined to transport and delivery.

Indeed, in the G7 countries’ national statistics, the share of self-employed workers has reached historically low levels, except in France (a slight increase since the early 2000s) and in the UK (where the share of self-employed earners rose sharply during the 1980s, and has been stable since). But no upward trend is to be noted – unlike other indicators of precariousness, such as part-time work.

 

There is a framework of economic analysis that is rarely discussed within tech circles but is much more significant in recent economic history. I am talking about the « fissured workplace hypothesis ». The main lines are quite simple: over the last few decades, large companies have gradually transferred to subcontractors tasks that are far from what they consider to be their core business: housekeeping, catering, security, logistics… A study of the German market found out that in 1975, 82% of retail establishments directly employed at least 1 maintenance worker; the proportion had been reduced to 20% in 2009.

A few years after outsourcing these tasks seen as non-essential, wages are now 10 to 15% lower than those who continue to occupy these positions internally. Subcontractors’ employees no longer enjoy the benefits that large companies grant after negotiations with trade unions – and anyway the weight of trade unions has also declined – and are less organized. The larger firms, more concentrated as we have seen in the preceding section, pay for services rather than wages, and they leverage their clout to push prices down.

 

Lessons for AI

This trend is not directly related to new technologies – computing is not mentioned in the articles on the « fissured workplace hypothesis », although it can be assumed that it has played an important role in coordination between client companies and subcontractors.

Nevertheless, several impacts of AI can be foreseen:

  • If AI ​​reinforces the concentration of companies, then this phenomenon will be amplified.
  • The ability of AI ​​to expand and organize the subcontracting market (matching supply and demand) could also amplify the phenomenon.
  • A new type of fissure could emerge: the increasing outsourcing of tasks that are sometimes important, but whose scale effects related to data could lead to concentration in a few specialized providers (e.g. automation solutions for advertising or customer service). Unlike recent subcontracting movements, this would affect skilled jobs too.

 

4// Calcification (not overheating) is the enemy!

 

Talking about the digital revolution generally means highlighting the incredibly dynamic « metabolism » of the industry, from startup lifecycles to technological transformations or use case mutations.

Beware, however: the economic ecosystem is not renewed as fast as one might expect. In the case of the United States, for example, newly created firms accounted for 14% of existing firms in the late 1980s, compared to less than 10% in 2014. And the renewal of jobs is not exponential when looking at job openings as a percentage of existing jobs:

Source: US Bureau of Labor Statistics, JOLTS

Job openings in the « Information » sector (which, apart from tech jobs, include those in the media and telecommunications industries) are created at a slower pace than during the dotcom boom or even the years before the Great Recession of 2007-08.

(NB: unfortunately these statistics are not available for longer periods or for sub-sectors, for a more detailed analysis)

End-to-end, the trends we have attempted to analyze so far raise concerns about a fossilization of competitive positions at the scale of individuals and businesses, rather than the potential advent a treadmill-economy in which adapting to unbridled changes would be the preserve of a small number.

In this sense, as some thinkers suggest, we have more to fear from an explosion of AI that would calcify the achievements of some and the shortcomings of others. I would rather dream of giant shake-up.

 

5// My recommendations

 

Once the risks due to the rising inequality level have been established, what can be done? This article would be incomplete without some guidelines for public policy, but let it be clear that the ones that follow are only my own.

There are two main ways to reduce inequality. The first is to tackle it ex post as economists say, that is seeking to alleviate the impact of inequality through redistribution mechanisms enabled by taxation. The second is to tackle it ex ante, meaning to compress the scale of income and wealth generated by the economy, so that redistribution becomes less necessary. My preference goes clearly to the second option.

 

Redistribution: a good idea – in theory

 

Let us immediately dispel any questions about the title you have just read: I am in favor of the redistribution of wealth … but this should be, in my opinion, a last resort, not a first reflex.

This is why the robot and universal income tax proposals, within the precise framework of the fight against inequalities (the universal income is such a vast concept that it deserves in-depth articles well beyond AI) seem to be very dull. In addition to the fact that their assumption of a massive disappearance of employment through automation is unfounded, they are proposals of abdication: no, would one try and read between the lines, nothing can be done against primary inequality… I am not pessimistic to the point of asserting that the advocates of these « AI-specific » proposals are trying to distract our attention, but some commentators clearly think so.

In addition, redistributive policies can be counterproductive – i.e. hamper economic growth – if they are poorly calibrated. And that their negotiation in the political arena is always difficult. David Autor summarized it well: « One could imagine that with so much available wealth, redistribution would be quite obvious to solve. But history suggests that this prediction never happens. There is always a perception of a lack and a permanent conflict over redistribution, and I do not expect this problem to ease as automation progresses.« 

 

Better distribute the benefits of progress

 

This slogan can be applied both at an individual level and at a company level.

At the company level, it means increasing competitiveness for all firms, large and small, pioneers in the adoption of new technologies and laggards, so as to level the playing field and fight against the « winner takes most » tendency. It is the responsibility of the whole ecosystem of business services, of which FABERNOVEL belongs of course, but it is above all the responsibility of suppliers of AI solutions. To ensure that the barriers to the adoption and use of these solutions are as low as possible.

To build a competitive market where the rules of the game are the same for everyone, it is also up to governments to take responsibility – and without citizen pressure, this will be impossible. The study by Grullon and al. already cited points to a drastic reduction in antitrust actions initiated by the Department of Justice and the Federal Trade Commission in the United States: 16 per year over the last 3 decades of the 20th century, 3 per year at the beginning of the 21st century.

Public authorities are facing new problems arising from digital technologies, including monopolies that many deem inevitable because of network effects. As my colleague Kevin Echraghi has shown, there is an urgent need to strike back against the platforms, using their own methods. Nevertheless, one last time, behind the GAFAs and their gigantic size (they are currently, through the lens of market capitalization, the largest groups in the world), all industries need to be monitored closely to avoid too much concentration in a few hands.

 

At the level of individuals, there are two ways of approaching the problem: either to raise the skill level of the workers or to democratize the tools related to artificial intelligence so that specific skills are never a prerequisite for their use. As far as training is concerned, we come back to eternal questions.

  1. Ensuring that the education system ensures equal opportunity for all its participants; and demonstrates ambition for all students. Clarification: the arguments of type « but if we all held master degrees the differences would be based on something else » are nonsensical because the countries where higher education is the most common are also those where employment rates of young graduates are the highest.
  2. Enabling qualitative, ongoing training of all employees, not of those who are already better-off.

 

We must be ambitious about training, but we must also remain realistic. A distinction must be made between initial and ongoing education on the results that are achievable. To use a trivial metaphor: think « elevator » for the first, « staircase » for the second. As much as one can hope to raise the educational level of the younger generations rapidly and massively, to sell ongoing education as a miracle solution for equalizing the chances for the employees already on the job market is a dead end. MOOCs are an excellent example: behind the few wonderful stories of self-taught individuals who have become superstar coders (so much the better if they exist and if they multiply), above all, they are people who are already well educated and keep on gaining additional skills. A miner may become a coder, but it is less likely that she will soon become an expert in machine learning. Continuous training as a staircase is thus a chance for everyone to climb an additional ladder of the skills scale, but not to radically change the order of positions.

Note to reader: this last paragraph is not pessimistic, but realistic. Targeting the ongoing training effort on everyone but especially on those who need it the most would be a welcome policy to fight income inequality.

 

Conclusion

 

Information and communication technologies have played an important role in transforming the economy, especially the labor market, over the past few decades. But this impact has not been without deleterious effects, contributing to the increase in inequality between individuals, either directly according to skill levels, or indirectly through the effects of concentration of firms and reconfiguration of their limits.

In this context, the rise of artificial intelligence is likely to amplify the trend. Now is thus the perfect time opportunity to address the flaws of our economy and society in depth before it is definitely too late.

 

Do you want to assess AI's impact on employment in your company through a workshop?

Contact our AI Lab
This article belongs to a story
logo business unit

FABERNOVEL

Distribute the future. Connect leaders. Change the game. We ignite ventures and transform organizations for the new economy.

next read