In A World Without Work, Daniel Susskind asks a very important question: “will there be enough work for everyone to do in the twenty-first century?”
“No”, he answers — the threat of “technological unemployment” is now real. Susskind isn’t the first to write a robots-will-replace-you book. He won’t be the last. But he does set out his reasoning with unusual depth and clarity — for which I’m grateful because it makes it easier to understand the flaws in an increasingly influential argument.
Susskind’s own version of the argument begins with a pile of crap and then goes on to a load of balls. No, I’m not being rude — those are two of the images he conjures up to explain his thinking.
Let’s start with the crap — specifically, the “great manure crisis” of the late 19th century. Like a lot of stories about our Victorian forebears, this one doesn’t really stand up. Horse-dependent cities such as London and New York weren’t in any danger of being buried under horse manure, for the very simple reason that it was regularly removed and used as fertiliser. But, as the author points out, it ceased to be an issue at all when transport was motorised. Horses, once integral to the urban economy, disappeared from our cities, never to return.
Will technological progress do the same to the human workforce? That’s what economists such as John Maynard Keynes and Wassily Leontief have supposed — and Susskind, also an economist, agrees with them. In the final third of his book he sets out a policy response, but I’m more interested in why he thinks it will be necessary.
To sum up what he thinks the future of work looks like, he imagines the economy as a giant ball pit. Blue balls represent the tasks that human beings are the best at doing, and red balls those where machines have the advantage. He then paints a picture of how the ball pit changes over time. For most of human history, almost all the balls are blue. A few go red with early mechanical inventions, but then the industrial revolution happens and the process massively speeds up. As we hurtle past the present day and into the future, there are more and more red balls. Though some blue ones persist in the mix, they’re too few in number and small in size to provide a basis for anything like full employment.
This is Susskind’s vision of the future. It’s not “a boot stamping on a human face forever”, like in Orwell’s Nineteen-Eighty-Four, but rather the relentless shrinking of our balls.
Of course, it’s not the future yet. Though machines have never been more productive as they are today, the same can be said of human beings, which is why there have never been as many jobs as there are now.
Susskind himself speaks of an “Age of Labour” in which “successive waves of technological progress have broadly benefitted rather than harmed workers”. He does a great job of explaining why the “substituting force” that machines have on the demand for human labour has, on the whole, been exceeded by the “complementing force” — i.e. the tendency of mechanisation to increase the productivity of, and demand for, human workers.
The Age of Labour will only end if the complementing force is persistently and pervasively overwhelmed by the substituting force. This has never happened before. For it to happen in the foreseeable future, requires some disruptive technology that’s never existed until now.
There’s an obvious candidate: AI or artificial intelligence. (If we define an AI-controlled machine as a robot, then we can include robotics under this heading too.)
The story of research into AI stretches back to the post-war era. But, as Susskind reminds us, the founders of AI got off to a false start. Their big mistake was to think they could build a human-like intelligence. Needless to say, they got nowhere. Eventually the funding dried up and research ground to a halt (the “AI winter”). The field revived when developers realised that they didn’t have to mimic the human brain to computerise tasks that previously only a human could undertake. Instead, they exploited the capacity of computers to learn through trial and error. In the right circumstances, they’re very good at this — possessing vast processing power and infinite patience.
Not that they are patient of course. Not being conscious, nor possessing anything that could be called a mind, they don’t have human qualities. Even the ‘intelligence’ in ‘artificial intelligence’ and the ‘learning’ in ‘machine learning’ are just metaphors. But that’s beside the point, says Susskind. No matter how mindless and mechanistic AI might be, what really counts is what it can do, not how it does it.
He describes the new wave of AI research as the ‘pragmatism revolution’ and there’s no doubt that it’s made breakthroughs. For instance, AI systems have been trained up to achieve total supremacy in games like chess and Go. There have been big advances in speech recognition, language translation and medical diagnosis software. AI is literally driving progress in automated vehicle technology. Facial recognition systems are now sufficiently advanced to raise serious concerns over civil liberties.
So just how far will the revolution go?
There’s long been a distinction drawn between routine and non-routine jobs — or, rather, routine and non-routine tasks. A routine task is one that can be taught as a ‘how to’ set of instructions — which is not the case with non-routine tasks. For instance, operating a till in a supermarket is routine; cutting someone’s hair (to an acceptable standard) is not.
The conventional thinking is that the first kind of work can be automated, but not the second — which is why we have automated tills and cash points, but no robot hairdressers. The key point is that because most jobs involve at least some non-routine tasks they’re safe from the robots.
Or are they? Look at the way that smart meters have replaced human meter readers. It wasn’t necessary to create a robot that can knock on your door, ask to be let in, find the meter, read it and report back to HQ. Instead, you just need a small box that sits in the meter cupboard, sending in the readings electronically. All the non-routine aspects of the meter reader’s job are thus eliminated, allowing the role to be fully automated.
Of course, that won’t work with every job. You can’t eliminate the non-routine aspect of a hairdresser’s work, i.e. cutting hair, because that’s the whole point of it. Susskind, however, argues that AI is finding ways of turning non-routine tasks into routine ones.
For instance, there might not seem to be anything routine about playing chess at a professional level. There are various principles and strategies that can be taught to any reasonably intelligent individual, but there is no list of instructions that one can follow to become a grandmaster. Except now there is — if you’re a sufficiently advanced chess computer. AI systems like AlphaZero have taught themselves how to play chess, encoding the instructions as a very complex algorithm that can be applied to any particular game.
It’s not that AI researchers have cracked the secret of human mastery of chess. The computers that can beat any grandmaster do so mindlessly — they don’t need to be conscious or creative or cunning. All that’s necessary is the brute processing power of the silicon chip — plus some smart machine learning software kindly supplied by some helpful humans.
Far from being reassured by the fact that there’s been zero progress towards the creation of a true machine mind, Susskind’s key point is that the ‘narrow’ AI systems that do exist are nevertheless routinising tasks previously thought to be non-routine. Thus a crucial limitation on the automation of human work has been lifted. The process of task encroachment is therefore fundamentally unconstrained.
The implication is that the substituting force of technology will continue, while the complementing force falters. The ongoing automation of tasks will free up resources that can be devoted to other tasks — but increasingly that demand will be satisfied by further AI-controlled machines instead of humans. The routinisation of more-and-more non-routine tasks would thus mean less-and-less need for us.
Do you feel those balls shrinking yet?
But what if there’s nothing particularly special about AI? What if there’s nothing new about routinising non-routine tasks?
Though Susskind clearly understands that an AI isn’t a true intelligence, he does suggest that they are nonetheless things of wonder. He draws a parallel between the mindlessness of AI and the godlessness of Charles Darwin’s account of the natural world. Though he had no place for a divine creator, Darwin still believed in the “grandeur” of life. The products of both AI and natural selection may arise from a mindless, bottom-up process, but nevertheless they can inspire “an almost religious sense of awe.”
As well as bigging-up our “unhuman machines”, Susskind is also keen on “demystifying human beings” and “not granting them magical powers”. Indeed, he rails against what he calls the “superiority assumption” — the idea that we’re special and will always be indispensable.
This is where the deepest flaws in his argument are exposed.
I once had a colleague who was a gifted draughtsman. Using nothing more than a pencil and a piece of paper he could draw a near perfect square or circle. There were slight deviations, of course, but they didn’t distract from the essential squareness or circularity of what he’d drawn. Beyond the basic techniques, there are no instructions for achieving that quality of draughtsmanship — it only comes with practice and then only to those with natural talent. It is certainly non-routine.
And yet the task of drawing a square or a circle can be routinised using nothing more than some simple tools. So should we discern “grandeur” in something as basic as a plastic ruler — just because it provides a commonplace substitute for a rare talent? Should we think any less of what a draughtsman can do because some of it can be replicated by the contents of a child’s geometry set?
Many of our tools and machines routinise non-routine tasks. A camera captures a landscape in greater detail than any painter; a record player plays music with greater consistency than any musician. But which is superior, the machine or the artist?
Machines, including AI systems, can mindlessly do things that people do mindfully. But even if the outputs are comparable, the difference between the inputs remain all-important. The mind that allows the draughtsman to draw a pleasing square is also what enables him to design and build a cathedral.
Creativity requires consciousness. AI has been able to automatise various tasks, both routine and non-routine, because there are ways of doing them without creativity, common sense or any other faculty that depends on a conscious mind. But just because this method works with games that have clearly defined rules and with various kinds of pattern recognition, it doesn’t mean that AI methods would work in most other domains of human endeavour. To assume there is an algorithm waiting to be discovered for just about anything human workers can do, is just that — an assumption, one unsupported by compelling evidence.
Another assumption is that there really is nothing “magical” about human consciousness, that it is purely material in origin. But even then, one would then have to ask why natural selection would go to all the trouble of evolving such an unfathomably mysterious feature, instead of relying solely upon the biological equivalent of the algorithmic processes that AI uses. Perhaps the explanation is that conscious thinking is so much more capable.
Ultimately, an AI system is just another tool. It may be a very useful tool, but, being mindless, it’s got a lot more in common with a plastic ruler than a human being. AI technology isn’t even unique in making a routine out of the non-routine. That’s been going on for millennia with much simpler machines.
To prove his point, Susskind would have to show that AI has accelerated the process of task encroachment to a level never seen before in human history. But he doesn’t.
Indeed, one has to ask whether the process is slowing down. Just think about our home lives. My grandparents, who were born at the beginning of the 20th century, lived through changes of immensely greater significance than anything I’ve experienced. In particular, my grandmothers were liberated from a life of utter drudgery by domestic technologies like electric lighting, central heating, the washing machine, the vacuum cleaner, the gas cooker and the refrigerator. Nothing in more recent decades compares to the importance of these machines. The wretched smartphone doesn’t even come close.
Where are the great labour saving devices of the 21st century? If task encroachment is all set to make us redundant, then why are we still doing the ironing?
Susskind is not a engineer and his book is not a technical manual. Rather he’s an economist and therefore what we might expect from his profession is statistical evidence that AI is transforming the economy. If a revolution really is underway, it should show up in lower employment levels and a much higher rate of productivity growth. He does cite various economic trends, for instance job losses in the manufacturing sector, but these are better explained by factors like outsourcing, offshoring and immigration.
That’s not really a criticism. The reason why killer facts are absent from his book is because they don’t exist. Not yet, anyway. What A World Without Work does provide, however, is a conceptual framework that can be tested against the evidence as it comes in.
Until it does come in, we can only argue the case on a theoretical level. On that score, I have deep disagreements with Susskind’s argument.
Nevertheless, where he does go wrong, it’s always in an interesting way. Which is more than you can say for any machine.