Currently, there is a lot of discussion about the impact of technologies such as artificial intelligence on the world of work and employment.
Some of this is alarmist, and some excessively excited. There will indeed be dramatic changes, but history and economic theory both suggest that these will not radically alter the nature of the economic system.
However, while we should not fear the changes brought by widespread and extensive automation, we should be concerned about the way the process works and about its short-term and transitional aspects. Devising ways of dealing with these is a real challenge for both public policy and civil society.
In the last few years, there has been a lot of discussion about a new wave of automation that is already under way and starting to transform much of the economy. The central element in this is the combination of telecommunications with artificial intelligence (AI). This makes possible, people both hope and fear, the replacement of a great deal of human labor of many types. It is AI in particular that attracts the attention, not least because of recent dramatic breakthroughs such as the development by the Google-owned firm Deep Mind of an AI that could defeat the world’s best Go player (Go is a game played mainly in the Far East that in terms of its complexity is at least one order of magnitude higher than chess).
There have of course been many episodes of mechanization and automation before over the last 250 years. The argument made by many is that this time really is different for two reasons:
AI replaces not just human labor but the human mind and judgment as well;
and the automation will make market signals unnecessary because the key resource will be information, which is inherently abundant and can be reproduced at zero marginal cost.
There is a consensus that a lot of jobs or kinds of employment are going to disappear in the next two decades or so. There is disagreement over just how many will go as a proportion of currently existing employment. The OECD estimates that just under 10 percent of existing jobs are at high risk of automation. Other studies all conclude that the correct figure is somewhere in the high 40 percent range. The weight of opinion is therefore on the higher side of the two kinds of estimate.
Kinds of Jobs
There is a general agreement about the kinds of jobs that are likely to vanish. They all have certain qualities. One is that they are routine and repetitive, involving the repeated performance of standardized tasks. This includes both simple manual jobs and process-driven desk jobs. Another is that the job or role can be captured in a decision-making tree or flow diagram so that the range of decisions that have to be made is finite (it may still be large) — this means it can be done by an algorithm.
There is also general agreement about the kinds of work that are at low risk of automation. One is work requiring manual dexterity and manipulation (because of the difficulty of replicating the human hand); another is anything that requires human judgment or creativity, dealing with something that cannot be captured in an algorithm. The OECD study also argues that work involving human interaction is likely to survive simply because people crave human contact. Others are skeptical about this. Finally, there are some cases where stubborn human prejudice will keep the job in existence: it would be much safer if airplanes were entirely automated, but in polls most people would (irrationally) prefer a human pilot.
Given this, we can easily construct a list of the kinds of employment that are likely to vanish in the next decade or two. These range from jobs such as truck drivers and taxi drivers (replaced by autonomous vehicles), to a lot of logistics and warehouse work (replaced by automated handling systems), to a lot of routine work in the financial-services sector both high and low paid, to most legal work (but not trial lawyers) and most medical work, including diagnosis and prescription (but not surgery). There will thus be substantial losses of both blue collar and white collar jobs — in fact the losses of the second kind are likely to be larger.
Faced with this prospect of a huge upheaval in employment with many kinds of work simply vanishing, one response is to panic. The fear is that there will simply be no work available, or not enough for the people looking for and needing paid work. Others are excited and see this as a huge opportunity.
A popular reaction at the moment is to see this as the way toward a radical reconstruction of the entire economic system and a move beyond capitalism to some other kind of economic order. The idea is that the connection between work and income will be decisively severed and that we will also move into a world in which many products will be capable of being reproduced at zero marginal cost, which means they will be effectively free: in that case, the price mechanism will no longer operate.
This view has been eloquently put forth by people on the Marxist left such as Paul Mason (in Post-Capitalism) and Aaron Bastani (in Fully Automated Luxury Communism). These authors see the chance to realize the vision of the young Marx, in which the alienation of work is abolished along with the division of labor.
There are also people on the free market side who see this kind of outcome as likely, although what they envisage is a capitalist economy in which a large part of the population subsists on free stuff while not working or doing low-paid work.
New Kinds of Work
Neither panic and despair nor excitement is justified. The question to ask is not whether new technology is going to replace many jobs but whether those jobs will be replaced by new ones. There have been several episodes before where observers have expected the end of employment because of automation, and in every case the actual result has been that while many jobs do disappear, they are replaced by even more new ones.
Of course, that does not mean the same pattern is bound to happen again — to think so is to commit the fallacy of induction. It really could be different this time. However, there are theoretical reasons to doubt the more excitable predictions.
Firstly, many theoreticians of AI have a mechanistic view of human consciousness and decision making. For them, the human brain is simply a computer, only more complex and made of neurons rather than silicon. Hence the processes that create human thought are no different from the kind that take place in a computer or an AI, and any and all of them can be reproduced in a sufficiently advanced AI. This would mean that any and all human activities could be performed by an AI.
This, however, confuses intelligence and consciousness. We truly have no idea what the latter is or how it is produced, but we do know that the two are distinct (there are animals that we can show to have one but not the other). An AI or computer procedure, no matter how advanced, can only do what its programming and algorithm allow for — it cannot originate anything. This means that genuine creativity or the exercise of judgment when confronted by something novel cannot be built in. They remain human capacities.
Secondly, there is the question of knowledge. Even if most activities can be reduced to an algorithmic decision tree, the knowledge that those decisions will be made on is mostly tacit, localized, changing, and therefore incapable of being expressed in writing or numbers. This gives humans an advantage because of their greater flexibility and adaptability (AI can also learn, but this process is not as flexible as in humans). Putting the two things together means that there are many areas where humans will retain an advantage. Even if AI can also do these things, it will do so at a higher cost.
Thirdly, this misunderstands what automation does and hence its effects. What automation of any kind does is to make work more productive and hence to free up time and resources by increasing the intensity of the use of resources. Instead of using X amount of time and Y physical resources to get a given output, you use a fraction of X and Y to get the same result. This frees up the resources and human time for doing other things.
The result is fewer people doing some things and the people no longer doing them doing other things (often in different places). One challenge is that we do not know what those other things will be (although we can make informed guesses). We should not speak of a displacement of human labor but rather of its being freed up to do new things, in the way that all of the labor once needed to grow food has been released to do a myriad of other tasks. (The argument about zero-marginal-cost production misunderstands the nature of both scarcity and the price system, but that is another argument).
So should we just chill and see what happens? Not so. There are two genuine challenges that these changes pose. The first is that the rewards from the new activity and its output will accrue to a small minority. The historical pattern is that this is what happens in the early phase of any technological transformation, due to first-mover advantage and simple good fortune.
However, as time passes and the returns to the new technologies decline as it becomes mature and widely adopted, the income and wealth gap that widened in the earlier phase starts to shrink. This is what we can observe in both the 19th and 20th centuries. However, this takes time, and in the meantime you can have serious political and social unrest, for obvious reasons. Moreover, today the high rewards to first movers are artificially heightened and prolonged by the legal system, above all the current regime of intellectual property rights.
The second challenge is that the transitional costs in human terms of rapid innovation can be very high. Outside the world of economic models, the reallocation of both capital and labor as a result of technological innovation is neither immediate nor frictionless. This is because of the heterogeneity of both labor and capital — they are not uniform.
In plain English, this means that many capital resources such as buildings and machinery will simply become useless because they are in the wrong place or cannot be readily adapted or changed to a new use. In terms of labor, a person who has a range of skills that are now redundant may find it very difficult to acquire new ones or to move physically to a different place. In human terms, this can be very painful and traumatic and very destructive of both human connections and personal happiness. This is a real challenge - you can write off capital, but writing off human beings (often in large numbers or in concentrated locations) is both wrong in itself and very dangerous.
What Should Be Done?
If that is the real challenge of AI and automation (as opposed to fantasies of automated luxury communism or panic and despondency about the replacement of humans), what then should be done? Clearly, there is a place for imaginative public policy, which should mainly take the form of institutional reform and innovation rather than paid programs (e.g., radical reform of intellectual property).
The main step though is to look to social entrepreneurialism. We need people to develop solutions to the challenge of radical change in work and employment at a local level and in a decentralized but networked way. It is mutualism and social action that we need to rediscover and employ. Fortunately, some of the results of the current wave of automation are likely to make this easier, but that is for another column.