Commentary on Capitalism in the age of Robots: Adair Turner’s Speech on Artificial Intelligence.
Introduction
AI will automate a large number of jobs, and potentially put a lot of people out of work; this was reportedly Matt Hancock’s nightmare when in his recent role at DCMS. It has made for dramatic headlines about our rapidly approaching dystopian future where the age of the Matrix, with people plugged into devices, if not quite slaves to the machine in the exact same way, could indeed happen. Adair Turner’s speech outlines that many jobs will indeed be automated in areas that currently employ a lot of people – and AI will generate or replace jobs affecting a lot of “valuable” GDP. Dislocation, with those jobs being replaced by robots and AI will, to some extent lead to jobs being generated in other parts of the economy, or at least that is the established economic theory. So, while bad news for those being replaced, if people look for other work, and are retrained with additional skills, they may find other more suitable work- again according to standard government responses and experience to date.
What is remarkable about this speech is that it doesn’t support the standard analysis, and the author, a former Chairman of the Financial Services Authority and pillar of the establishment – makes a cogent argument that runs against this establishment thinking.
Adair Turner points to certain types of jobs where the impact of AI is likely to be limited. However, those types of jobs have limited consequences for GDP, and the central theme of the speech is that, in a world dominated by the Technopoly and AI, GDP as a measure of general welfare is increasingly a misplaced guide and the wrong target for government policy. The speech’s conclusion is that the justification of markets in terms of their productivity and the benefits of increased productivity in terms of growth and worthwhile jobs must be revisited. The narrow focus and assumptions of the system naturally allocating resources and rewarding merit-based activity also must be revisited.
We have dealt with what we see as the main highlights of Adair Turner’s speech below.
Explaining the Solow Paradox
Nobel Prize winner Professor Robert Solow famously commented that “computers are everywhere but in the productivity statistics.” Adair Turner explains that this is no inexplicable paradox: rapid technological growth is bound to result in a proliferation of low productivity jobs, redistributive or “zero-sum” activities, and increases in real consumption which never show up in GDP statistics. This is a major insight. It means that the latest wave of technology has not necessarily increased productivity and continuing to look at GDP as a proxy for welfare and whether society is benefitting is likely to mean that we are looking at the wrong target.
How can a wave of technological progress not increase productivity? It is normally argued that economic progress drives innovation and new opportunities, which arise through the operation of the market. The speech points out that is not always right. He refers to [Baumol 1967] who puts it as follows: “Total productivity growth is as much driven by the productivity and productivity growth potential of the sectors into which workers move, as in the sectors where jobs are automated away. Rapid technological growth can be and often is accompanied by a proliferation of low productivity jobs.” There is a current assumption that productivity improvements will give rise to endlessly repeatable productivity growth. Anyone suggesting otherwise is often branded a luddite. The truth is more complex. The following example is worth repeating:
“Indeed it could be argued that that is what essentially happened in the aftermath of the first agricultural revolution, which developed in the fertile crescent of modern day northern Iraq some 8,000 -10,000 years ago and which subsequently spread to other parts of the world. There was a revolution in agricultural productivity as people moved from being hunter gatherers and pastoralists to settled farmers with cultivated plants and domesticated animals; feasible food production per acre and per person employed increased – … but there was no shift to a path of ever increasing productivity across all sectors – …. and it seems likely that average living standards for the mass of the population actually declined [Scott 2017] – … as the surplus created by increased agricultural productivity was absorbed by a proliferation of unautomatable jobs devoted to serving the owners or controllers of the agricultural surplus – whether in the form of huge numbers of low paid domestic servants and temple/pyramid builders, or smaller numbers of higher paid priests and skilled artists. We had an initial productivity revolution in one sector, followed by overall productivity stagnation for several millennia thereafter. We tend however to assume that this pattern will not repeat in the modern world, and in one respect we are almost certainly right to do so. For one reason why there was only a one- off spurt of productivity in the first agricultural revolution, is that human beings simply couldn’t work out how to increase productivity in anything other than agriculture, whereas today we have processes of organised scientific discovery and entrepreneurship which continually identify ever more opportunities for technological advance and productivity improvement.”
The expansion that is currently expected to come from productivity improvements arises when activity allows people to improve output. However, as the example shows, much depends on what the new technology allows people to do and the mix of activity that it replaces and enables. The hunter gatherers that became farmers were more productive, but the jobs that were created were not all better jobs, with larger numbers of lower paid more domestic servants than had been the case before the changes.
The impact of Technology, described elsewhere as the Technopoly, could be observed as creating, and further accelerating, rapid technological progress in some sectors of the economy, but declining productivity growth overall. Adair Turner describes numerous indicators of this effect at work in modern societies:
- “Consider for instance the case of “Deliveroo” drivers in London, riding around town on that cutting-edge piece of technology known as the bicycle, to deliver whatever you want direct to your door (Exhibit 12)
- Or look at the US Bureau of Labor Statistics forecast for job creation by category between 2014 to 24, with 8 out of 10 of the fastest growing job categories paid far below the average wage, and with these jobs concentrated in those activities which at least for now we find more difficult to automate – personal care aides, home health aides, nursing assistants and cooks (Exhibit 13)
- Or consider a recent report in the Economist magazine about factory automation in India (Exhibit 14), with the factory manager clear that production jobs are bound to disappear, but confident that he can “somehow find jobs for everyone”, “as drivers or watchman if necessary”, but almost certainly over time, at a considerably lower rate of pay. [The Economist 2017] That phrase “somehow find jobs for everyone” captures indeed part of what goes on in a modern economy facing multiple automation possibilities. We find things to do and that means that the rate of measured productivity growth is far below what it could potentially have been.
- Or think about the example of how technology is now radically changing supply chains internationally, for example artificial intelligence allows “sewbots” to automate footwear manufacturing. One recent ILO report suggested that 60-90% of low paid jobs in ASEAN textiles and clothing industries could disappear through automation. [ILO 2016] This allows re-shoring at developed world costs. The recently opened Adidas “Speedfactory” in Bavaria which will employ 160 workers to make 500,000 pairs of shoes per annum: if the rest of its production becomes equally automated, the roughly 1 million workers currently employed in Adidas’s supply chain across the world could fall by over 90%.”
The speech identifies a category of what is called “zero sum” activities, or distribution activities, for example the activity of lawyers, government, and civil society, where the process of reaching agreement and deciding on things is regarded as not ‘adding value’ and essentially distributive. The expression zero sum is used to refer to them. He argues (unconvincingly) that this is not a disparaging term and disparagement is not his intent- the language involves assessment of work in a mathematical sense. So, when looking at productivity the fact that one side of an argument employs a lawyer and is disputed and debated by another lawyer the activity is considered to be “zero sum”- the political process is also “zero sum” as it involves no value add in a GDP sense, and reaching agreement among different people is thus not valuable in increasing productivity. In a broader economic sense legal and political systems and ways of addressing crime and security, which are well known to be critical for the predictability needed for investment. In a value-based sense the language is also disparaging: “zero sum” can be taken to mean “useless”. If we didn’t have mechanisms for conflict resolution of all types, we would be in a dysfunctional and strange society; one that would attract no investment altogether.
Clearly the benefits to society of these “zero sum” activities are more important than accounted for in the speech. Indeed, we should call them pro social or otherwise activities and work to be sought out and endorsed as public goods. Adair Turner appears to acknowledge this – and his point is that these activities are likely to be unaffected by AI. We may even need more politicians and lawyers and where people with more time on their hands can get more involved in doing things for others and helping each other rather than spending time on endlessly repeatable jobs producing stuff that can increasingly easily be automated. Perhaps the refocusing that is called for could include an aim of more valuable contributions from people involved in zero sum or distributive roles that cannot be performed by robots.
In short, AI could usher in the opportunity for people to be more pro-social. This raises a central question that many have talked about, including Adair Turner: we need to set up the goals of the system in a different way. GDP goals are meaningless; we need other goals for society. The education system is currently designed to channel people into productive work that could now more easily be automated than in the past; perhaps the current, instrumental, view of education needs to change? Otherwise we will be educating for roles that have disappeared and thus rendering individuals unfit for a future where people don’t need to work in factories and where roles that are needed are creative, social and fulfilling. However, as we point out separately in the Technopoly paper, in a world where AI is the servant, not the master of mankind, the risk of exploitation also needs to be addressed, which involves action to address the structure of industry.
On GDP measurement, Turner asks the following question:
“– are today’s children, equipped with computer games and with access to social networks of a technological power unimaginable 30 years ago, happier, more fulfilled, and less stressful than children 30 years ago, it is not certain that the answer is yes; some commentators indeed would argue that there have been significant adverse welfare effects.”
He contends that it is important to recognise two quite separate questions:
- do statistical measures of real GDP per output tend to indicate less rapid technological progress than is actually occurring,
…and then separately…:
- is the impact of that technological progress a clear positive for human welfare?
The speech suggests that the answer to (i) is “clearly yes,” while the answer to (ii) is “sometimes yes” (healthcare innovations) and “sometimes perhaps no” (ever more complex computer games). Also, one of the reasons why “computers are everywhere but in the productivity statistics” is that our estimates of real productivity growth fail to account effectively for some of the most dramatic increases in productivity.
Meaningless measures in the hi-tech hi-touch economy
The discussion in the speech covers the fact that the three effects which explain Solow’s paradox have always been present to some degree. Industrialisation in 19th-century Britain was accompanied by a rapid growth in domestic service employment. A significant proportion of all human work activity has always been “zero-sum” in nature – often involving activity to refine and address conflicts and address problems created by progress. GDP measures have always been imperfect. The observation in the speech is that these effects increase as we get richer, and as technological progress accelerates. GDP measures have always failed to capture improvements in human welfare. Now the economics of information and communication technology (high fixed costs, low variable costs, economies of scale, and, most importantly, economies of scope) greatly increase the problem of measurement and focus on GDP. In addition, a point that we raise in Technopoly, it can be observed that digitisation, the process through which physical goods are substituted for by low cost digital services, compounds these issues.
According to the speech, consequences include:
- Our standard assumptions about how to measure economic and human welfare progress are breaking down; and
- The measured economy becomes dominated not by the production of welfare enhancing goods and services, but by non-produced assets, rents and distributive games.
In the developed world, and still in much of the developing world today, certain assumptions may well have been “good enough” to allow GDP and proxy measures to provide a reasonable guide for economic policy. When people first move from farms to factories and then from gradually automating factories into newly emerging service sectors, so that for the first time in history people can own cars, washing machines and televisions and enjoy piped sewage systems, restaurant meals, and hotel stays by the beach, it may be broadly true that underlying productivity improvements of say 3% per annum are reflected in measured per capita income growth of around 3%, with human welfare also improving quite rapidly, so that each generation feels better off than the one before. The speech questions whether these assumptions are now useful. As a result, our standard GDP measures are becoming both:
- less meaningful indicators of the underlying pace of technological change; and
- less useful indicators of increases in human welfare.
The speech also provides an interesting discussion of a possible vision of a world in 2100 which references increased rental income and a rentier society (if that is truly what people want, and action to prevent that vision becoming a nightmare is not taken). Adair Turner discusses the outcome he predicts, so in a world of increased automation with fewer jobs and where leisure time is increased, he considers that people will not be free to do what they want- because incomes will collapse.
Incomes not jobs?
Widespread reporting has captured the fear that automation means mass unemployment. Some reassure us that waves of automation have led to greater employment in the past. The speech makes the point that “Some of the arguments put forward by the optimists are unconvincing, but in the long run they are right that there is no limit to how many jobs can in principle be created. There is however, little doubt that effects of change are felt in different places and different industries at different times. For example, where Amazon has itself “created jobs” and added more employees and more people are employed in its supply chains, those additions ignore the number of jobs lost in in all of retailing that it is substituting for (physical and online combined), and total employment overall may now be falling.
Nor is it convincing to argue that the growth of a profitable firm or sector “creates jobs in the rest of the economy” simply because income earned in one job, company or sector is spent on other products and services. All economies involve an endless circular process of earning and spending, but that obvious fact tells us nothing about the equilibrium level of employment relative to the working age population. Adjustments may be painful for many at different times in their lives and lead to political and social problems.
What is not discussed in the speech is the issue that whether this level of dislocation is inevitable or is desirable is debatable. Much depends on what is done to take evasive action. The speech describes a world that might happen; not on that will inevitably happen.
As Adair Turner puts it:
“… in a world of ever increasing automation potential, that full employment equilibrium may be accompanied with ever rising inequality, and there are signs of that trend already at work. For over thirty years, incomes have diverged from the median at both the top and bottom of the US income distribution, and rapid technological progress is almost certainly one of the drivers of that divergence. At the top end of the income distribution, returns to high skill in the development and application of information and communications are and will remain huge. But they are likely to be concentrated in the hands of the small number companies and individuals needed to create all the software and apps needed by our increasingly automated economy. The giant dominant software and application companies of the Internet age create huge equity value and sky-high incomes, but with a startlingly small number of employees. In the middle and bottom of the income distribution, meanwhile, the fact that predictable physical jobs are the easiest to automate, will mean continued significant job losses, and it is likely that many of the workers displaced will only find new jobs in the low-wage sectors of economy.”
Inequality appears to be increasing. This may be the stuff that ignites populism and supports political change. Whether the inequality of outcomes is inevitable is discussed. Conventional wisdom is challenged in the speech and the ethical and moral implications of allowing merit-free wealth to be accumulated by only a few is canvassed. Standard responses, such as more work and effort for those that are affected needing more “skills and retraining” are rejected. It is pointed out that retraining is unlikely to work at all this time around because:
- firstly, we only need a small number of clever IT literate people to write all the code needed for all the robots, all the apps, and all the automation software that can then write itself; and
- secondly, however many people are employed in IT, the income related to that activity will accrue to only the most successful: there will be both successful and unsuccessful IT players in the markets concerned. The speech posits a world where an increasing percentage of the total workforce is employed in difficult to automate face-to-face services, and the equilibrium wage rate for those jobs may be influenced hardly at all by whether they have the skill required to be an adequate, but still second tier, software developer. This may be true for many reasons not discussed in the speech, in particular when considering people’s feeling of self-worth, and what they regard as a valuable use of their time, rather than how the market values their time.
Other policy responses considered:
Income support such as Universal basic income (UBI). In a world with a limitless “reserve army of robots” it is suggested that free-market competition won’t deliver the real wages and minimum adequate incomes which make people, in some sense, equal citizens of society. So, we could provide everyone with a basic income to compensate for lack of opportunity. The case for UBI may be superficially attractive there are reasons it is not a panacea. Work, and adequate remuneration from work, delivers a purpose and sense of status and self-worth which a pure monetary subsidy cannot replicate. Success based on merit, and the opportunity to achieve is a cornerstone of a just society. As Adair Turner points out, as the relative importance of property prices and rents grows – with rising average property prices relative to income but also increasing divergence between regions and cities – in the Adair Turner vision of the tech driven economy, the income required for an adequate living standard varies greatly between different locations; policies aimed at ensuring affordable housing may therefore be a more cost-effective and targeted policy response than a UBI alone.
Offsetting the concentration of income, wealth and rents. If new wealth derives from uncreated increases in land prices, there is strong case for increasing the real property taxes and inheritance tax. The speech also suggests a good case exists for restricting the breadth of application of IPR (the categories of “invention” which can be protected) and for reducing the length of IPR and copyright protection. However, the opposite would in fact be the case in an economy based on creative industry and large numbers of creative industrialists that would be prevented from merging via sensible antitrust laws designed to secure a more creative and happier population – and the argument for increased protection for the creative economy in both increased IPR protection and protection from predatory takeover by existing players is at least as compelling.
High quality urban development. The speech suggests that macro and micro economists pay little attention to the physical realities of spatial development, city design, architecture and transport systems, let alone smart city developments. In a world where land in desired locations is likely to account for a rising share of all wealth, the geography of economic development is also likely to play a vital role. Smart Cities and their development deserve close attention.
Adequate wages and status for caring services. Jobs in face-to-face caring services – the need for personal care aides, nursing assistants, home health aides is growing rapidly in the developed and aging economies. These are among the low paid, and (at least for now) unautomatable jobs to which labour shifts as automation reduces jobs in other sectors. They are also jobs which deliver high inherent social benefits. Reduced taxation or increased minimum wages could be part of an answer to encourage people into such jobs and make such businesses more attractive.
Celebrating craft skills. More people are likely to find satisfying jobs as skilled gardeners, artists, cooks, craft brewers, small urban organic farmers and beekeepers than as software developers. This should be celebrated and promoted.
Increased leisure. It would be ideal if technological advance led to increased leisure. Whether it will is debatable.
Education for life and citizenship. Education policy has tended to reflect an instrumental and narrowly economic philosophy: better education and skills for business use is desirable because they will raise personal income and productivity growth. Where productivity growth is driven by technology and benefit a small proportion or the population, why is still higher productivity growth the key objective? Adair Turner suggest that re-education and intervention is needed to address the technology that has provided “the impetus it has given to fake news and the manipulative reinforcement of initial prejudices.”
Freedom– we wouldn’t know what to do with it- or would we?
John Maynard Keynes put it as follows in his 1930 essay, Economic Possibilities for our Grandchildren: once the problem of production is solved, “for the first time since his creation man will be faced with his real, his permanent problem, how to use his freedom from pressing economic cares, how to occupy the leisure, which science and compound interest will have won for him, to live wisely and agreeably and well.” Adair Turner states: “I believe we are clearly on route to “solving the production problem” so completely that driving further productivity growth should no longer be a primary objective of policy.”
This may work for the developed world, but a thought is spared for those developing countries where radical automation combined with rapid population growth could make economic catch up impossible. The speech points out that the classic development ladder on which countries have climbed out of poverty is being destroyed. For example, between 1800 and 1950 a huge gap opened up between the standard of living of rich countries, in particular in North America and Western Europe, and most of the rest of humanity. Over the last 70 years, a number of countries have caught up remarkably, growing far faster than rich countries, and at least partially closing the gap. Almost all of these countries have achieved this using the same development model. Export-oriented labour-intensive manufacture has made it possible to absorb surplus agricultural labour. Rising income from manufacturing has made possible high savings and investments. High investments, in both plant and machinery, and infrastructure, has supported rising productivity and real wages. That is essentially the path which Japan followed in the 1950s and 60s, Korea and Taiwan from the early 1960s, and most recently China. In each case, labour-intensive and initially low-wage manufacture, producing goods for export, played a crucial role in unleashing the subsequently self-reinforcing process of development.
This is now radically changing. The Adidas “sewbots” are only one example of the type of re-shoring that information technology and the application of AI will facilitate. Many outsourcing activities that exploited comparative advantages and relied on technological capability, for example the computer science capability of Infosys in India, will likely change with the advent and deployment for AI. Whether that leads to more capability in far flung paces connected by technology platforms or less is difficult to predict.
Adair Turner accepts that the fact that something can be done doesn’t mean that it will be; much depends on the relative cost of robots and AI versus employees, and on relative wage rates in the developed and developing worlds. Provided wages remain low enough in developing economies, automation of basic jobs may be delayed far beyond the point when it becomes economically feasible. Cold comfort perhaps for those currently in imperiled jobs that the march of the robots depends on those accounting for their costs and the risks of implementation. India also already displays the developed world phenomenon in which rapid technological progress in some sectors and jobs is offset by the proliferation of low paid service jobs in activities which are for now difficult to automate. The effects on developing economies are canvassed and potential policy responses suggested. The issue of technology outsourcing, and the globalisation and interdependence that it creates between regions is not reviewed and is potentially a significant missing piece of the analysis: technology platforms have created interdependence, and flattened differences between regions, as discussed in Tom Freidman’s bestselling book The World is Flat.
Implications for economic theory
In terms of implications for economic theory, the speech is radical, and important coming as it does from one so influential. Adair Turner states:
“The theoretical case for believing that free markets will not merely perform better than planned economies but produce an optimal result of human welfare rests on the neoclassical theory of competitive equilibrium. Within that theory, three assumptions are presumed valid:
- First that free-market competition will produce an optimal allocation of the two factors of production, capital and labour, combining within a production function to produce the maximum income deliverable from any combination of capital and labour. Within this theoretical framework all labour is devoted directly or indirectly to the production of welfare enhancing goods and services, with no distributive or zero-sum activities. Free competition ensures that all workers receive income equal to their marginal private product which in turn is equal to marginal social product.
- Second that the prices of goods and services produced and consumed, reflect the marginal benefit to both private and social welfare.
- Third, as a result, that the aggregation of all goods and services into national income (GDP) accurately reflects aggregate utility, allowing us to use measures of GDP per capita to draw valid inferences about the progress of total human welfare over time or its relative level as between different countries.”
He considers that each of these assumptions is decreasingly valid since:
- an increasing percentage of all human activity will inevitably be “distributive” or zero sum in form; and
- a key problem is not that top income earners are paid incomes above their marginal private product, but that marginal private product can be immense even when marginal social product is small. Put simply, people may be richly rewarded by their activity but Adair Turner questions the social value of the activity and the merit of allowing such riches to accrue to those in distributive work. (Many would look twice at this proposition and need to carefully consider it further. For example, the idea that because distributive activity is not productive in an economic sense does not make it of low social value and a major criticism of the speech could be that it confuses what people actually value and want to spend their time on, with what economists describe as socially valuable. The risk here is that Adair Turner is making a personal value judgment that many would disagree with: in concrete terms if more people were involved in social activity and society, and in dispute resolution and conflict avoidance, would they be less socially valuable or more? Adair Turner sees these things as lacking in social value as they involve distribution and are zero sum. It can readily be understood that they are likely to be highly regarded by many beyond the economics profession).
- Beyond some level of income, the idea that all forms of additional consumption deliver equally important increments to human welfare breaks down. The speech suggests we need to answer whether standard measures of real GDP fail to reflect technical production and increased goods and services with less labour input – the answer is almost certainly yes. Does increased flow of goods and services contribute to social benefit? Here the answer may be yes for some hidden benefits (e.g. improving healthcare) but not necessarily for some other services (such as social networks and constant device use). Adair Turner suggests that we need methods for distinguishing the potentially different relationship between additional consumption and additional welfare for different categories of goods and services.
As a result, the value of GDP per capita as a proxy measure for human welfare is already degrading – and will inevitably decline over time. The argument is put forward that maximising real GDP growth can no longer be the prime objective of economic policy when the measure we seek to maximise is gradually losing its meaning.
Conclusions
The AI and Technopoly systems and structures allow for, and potentially create, increased levels of inequality. More generally, we need to recognise that “the problem of production” could be solved so completely as to become unimportant in the developed world.
As a result, the most crucial challenge for the developed world could now be how to manage in a fair way disputes about the distribution of those goods, services, and assets (both created and natural), which automation does not make available.
Adair Turner then suggests that we need to return to the political case for economic freedom on which the “political economists” of the 18th and 19th century – Adam Smith, David Ricardo, and J.S Mill – concentrated. As Amartya Sen more recently noted, the case for the market economy has for many years been stated predominantly in terms of the maximisation of income as a “culmination outcome” rather than in terms of freedom as an end in itself. Those who, like Adair Turner, want to make the case for a market economy will need to return to that more fundamental justification, balancing the case for economic freedom versus other considerations, rather than hiding behind the argument that economic freedom is essential to drive faster productivity and income growth where all will benefit, directly or indirectly.