You may not be keen on the idea of your chosen line of work being automated away. Like the horse in the old phrase “horse and buggy”, or the driver in the (soon-to-be-old) phrase “truck driver”, many people don’t relish the thought of a machine cutting into their livelihood and perhaps even their sense of meaning in life. (With the rise of the automobile, who knows how many horses fell into despair?)
But in this century, many, many jobs – interesting, fulfilling jobs – will receive automation’s kiss of death. Some predict that the death of such jobs will eventually lead us to a post-work heaven of leisure, while others predict a hellish state of poverty and boredom. So it’s worth thinking through a few questions: 1) in the coming decades, what sorts of jobs might be destroyed, and what sorts might endure and be created? 2) to remain employed, how might we prepare for this? and 3) how will these shifts affect how we find meaning in our work? To think more clearly about work, automation, and meaning, we’ll first imagine how they play out in three generations of shoemakers. Then we’ll examine the form of AI called machine learning and its impact on different kinds of work. Finally, we’ll look to Hegel and Hannah Arendt for a few insights into our strange, automated future.
In our fictional family of shoemakers, the first generation is a man named Cobbler Schumacher living in pre-industrial times. With no help from machines, he performs all the steps of shoemaking himself: he imagines the design of the shoes, he executes the design by making and unifying all the parts (sole, upper, laces), and finally he sells the shoes out of his shop. By being the decision-maker at each step, he has created an end product that he is proud of. He gains not only money but a sense of meaning from his creations.
Fast forward to the industrialised era. Although Cobbler Schumacher’s son had taken over the family business, he was forced to close it due to a flood of factory-produced shoes. Now the cobbler’s grandson works at an assembly line at Shoes Incorporated, with shoe soles whizzing by him all day. His job is to put a strip of glue around the edge of the soles, so that a mechanism further down the line can press on the shoe’s upper part. The cobbler’s grandson is not a shoe designer, nor a shoemaker, nor a craftsman – he is a gluer. Gluer Schumacher’s job is simply to serve the machine: from his left, the machine tirelessly conveys to him sole after sole, and to his right, it relentlessly demands freshly glued soles. He thus finds his own soul coming undone, gaining no sense of purpose from his work, only a pay check. Is automation partly to blame for this? To be sure. On the other hand, now even the poorest people can afford to buy shoes.
Fast forward to 2018. The 17-year-old great-great-great-grandson of Gluer Schumacher was planning on a humdrum but comfortable factory career at Shoes Inc. But the company has just built a fully automated factory, complete with 3D printing and specialised robotics. With shoes now cheaper than ever, and with a wealthier population demanding more variety in styles, Shoes Inc. needs more shoe designers. So the great-great-great-grandson attends a 2-year design school and starts a career at Shoes Inc. designing trendy new styles of shoes. Designer Schumacher takes pride in his creative work. Like his cobbler ancestors, he is a kind of craftsman, and finds his job interesting and meaningful – much more so than his factory job would have been. Automation destroyed his first career choice, but paved the way for a second, more fulfilling one.
But Designer Schumacher will also lose his job to automation. You see, he designs shoes not on paper but with 3D CAD (computer-aided design) software. This program automates some of the smaller, tedious design tasks to save him time, but it also was programmed by the software manufacturer to record every mouse and keyboard stroke he makes. The software company compiles all of this data with the data from thousands of other designers, and uses it to train new, AI-powered software to make design choices on its own. After seeing an impressive demo, Shoes Inc. “hires” the AI-design software and fires Designer Schumacher. The software that was formerly his digital assistant has now morphed into his replacement. So after 300 years of progressive automation, the Schumachers are shoemakers no more.
Hold on, you say, that’s science fiction – algorithms can’t learn how to do things just by observing humans. Yet auto companies are already doing this today to develop self-driving cars, using a form of artificial intelligence called machine learning (ML). Software installed in thousands of cars records every detail of human driving behaviour (steering, breaking, etc.) over hundreds of thousands of miles, all the while matching this data up with the car’s camera and sensor inputs. In this way the software eventually learns, for example, that if there are two painted lines then the car should be steered to stay in between them, that if it’s snowing then the car should be driven more slowly, and that if the driver in the car ahead is texting and driving then they should be honked at loudly. The data recorded is not 100% consistent (some human drivers don’t slow down when it snows), but the ML algorithms learn to mimic the most common behaviours. Surely, though, machine learning can only learn to mimic mechanical activities, and thus blue-collar jobs; surely the mental work of white-collar jobs cannot be automated? But it can, and it already has – just ask the hundreds of equity traders that Goldman Sachs has replaced with computers.
Because ML algorithms are so central to the future of work and automation, we should pause for a moment to consider the basics. What is an algorithm, and how could it “learn”? An algorithm is a set of instructions that a computer can perform to accomplish some task. Traditionally, algorithms have been written by people. For example, in the early days of Netflix, someone wrote a very basic algorithm that recommended movies to you, e.g. “if the user has watched action movies more than 75% of the time, recommend another action movie.” A traditional algorithm is a static and unchanging entity, and can only improve by being manually re-written by a human. By contrast, an ML algorithm is, in a sense, alive: after someone writes and executes it, it can change and develop without human intervention. That is, it can re-write and improve its own code. This is what it means to say that an algorithm “learns”.
When Netflix (and Amazon too) began using ML algorithms, their recommendations became far more specific and far more effective. Why? Because these algorithms learn and adapt to your specific viewing preferences. Instead of Netflix hiring millions of programmers to write personalised recommendation code for each customer, ML algorithms do it for them. As you watch more Netflix movies and feed the ML algorithm more data, it uses inductive reasoning on the data to make generalisations about you, which it then writes into their recommendation processes – e.g. “if it’s a weeknight, recommend an action movie, but if it’s a Sunday recommend an historical drama … unless it’s raining – then recommend a light-hearted comedy.” By tapping into your local weather, along with your past watching habits, the ML algorithm has learned what you tend to watch and when you tend to watch it. All of this can be done without the human programmer of the initial ML algorithm ever lifting a finger. As the computer scientist Pedro Domingos says, writing ML algorithms is like farming: the human programmer drops a seed into the ground and then walks away. If the ML “seed” has fertile soil – i.e. lots of data – it will grow on its own, autonomously. In our Netflix example, the fertile soil is the choices we make, expressed through our taps and clicks. Our fingers write Netflix’s software for them. How nice of us.
The invention of algorithms that can learn about the world, that automate the accumulation of knowledge, is epoch-making. Like the shift from conventional bombs to nuclear bombs, the shift from traditional algorithms to ML algorithms is an unimaginable increase in power. We do not yet understand what this will mean for human life in general, or for our working lives. For, algorithms that strategically place web ads and skilfully direct self-driving cars are just the beginning. Thinking back to Designer Schumacher, we can note that shoe design makes use of generalisable rules, and involves a rather limited number of factors. This means that ML algorithms will eventually be created for this domain. Furthermore, the Netflix recommendation style of ML algorithms will spread. Imagine a Netflix-like service at your hospital that gives you not movie recommendations but cancer therapy recommendations, based on your specific genome and your specific life history. Medical start-ups are building such services as we speak. But does this mean that jobs as complicated as “cancer therapy specialist” will soon be automated away?
Jobs of the Near Future
Think about your own job, or the job you aspire to. How easy or hard would it be to automate? To consider the kinds of jobs that may or may not be automated over the next few decades, and the new kinds that might be created, let’s consider a strange fact. For almost 50 years now, pocket calculators have been far superior to humans in many mathematical abilities; yet, up until just a few years ago, the greatest minds in robotics couldn’t get a robot to walk on two legs as well as a 12-month-old baby. Why would it be so much easier to automate complex mathematics than awkward baby toddling? This counterintuitive fact is what roboticists call Moravec’s paradox, the observation that very little computational power is needed to mimic and even surpass certain human mental skills such as arithmetic and playing checkers, while a great deal of computational power is needed to mimic certain bodily skills such as walking on two feet. Applied to the world of work, Moravec’s paradox explains why it has been fairly easy to automate something like basic accounting skills (e.g. tax preparation software), but up till now impossible to automate tasks that require moving around on uneven surfaces, e.g. your local mail carrier on their daily route. If a task requires the use of sensorimotor skills in a stable and predictable environment, like a shoe factory, it likely has already been automated, but if the environment is unstable and unpredictable, automating the physical task has thus far been impossible.
Taking all this into account, some thinkers break the world of work down into four categories: 1.) blue-collar jobs with repetitive, predictable tasks (e.g. factory work), 2.) blue-collar jobs with context-specific, unpredictable tasks (e.g. mail carrier; construction worker), 3.) white collar jobs with repetitive, predictable tasks (e.g. basic accounting; much customer service), and 4.) white collar jobs with context-specific, unpredictable tasks (e.g. psychotherapist; researcher). Robots in factories clearly automate predictable blue-collar tasks, and Moravec’s paradox helps us see why unpredictable blue-collar tasks are difficult to automate, while white-collar tasks of the predictable sort are already gone or not long for this world. So we might conclude that the jobs most difficult to automate, and most likely to endure for the coming decades, will be those in categories 2 and 4: blue- and white-collar work involving non-repetitive, unpredictable tasks.
Yet, creative capitalists are always seeking to make tasks more predictable and more controllable, and one way to do this is to move unpredictable blue collar work into a predictable factory setting. One example is the construction industry’s increasing use of prefabricated materials (e.g. factory-assembled walls), a trend that even established hotels like Marriot are joining. Another risk to formerly safe blue-collar work is advances in computer vision software; fruit-picking machines, for instance, now combine computer vision and a vacuum system to pick apples at a tremendous rate, with little human intervention.
These trends shed light on what the most secure jobs of the next few decades will be. Most of these jobs will be white collar jobs that involve mental work that is highly context-dependent, and whose context or environment is unpredictable, uncontrolled, and wide-open. That last point is important. Netflix’s task of giving you enticing movie recommendations is highly context-dependent, since the context is the individual viewer and their specific viewing preferences. But Netflix was able to automate this task because the environment is highly controlled: it is simply all the movies and TV shows that Netflix has in its digital library, meticulously tagged and catalogued. Recommending movies – or recommending products as a salesperson – thus differs dramatically from a job like “smartphone app designer”, where the decision-making environment is difficult to define and hugely expandable based on the strength of the designer’s imagination. After all, who would have guessed 10 years ago that calling a car with an app would be so useful, or that your life would be made complete by an app to track your cat’s bathroom habits?
Clearly, one part of such “context-dependent, uncontrolled-environment white-collar work” is creativity. Creativity is a loaded term and difficult to define, but what I mean by it here is quite broad: using your imagination to think up what is not yet actual, but merely possible, thereby bringing it closer to actuality. Under this definition, creative mental work would include not just app designers and architects, but also entrepreneurs, biology researchers, and boundary-pushing IT engineers. Therapists and counsellors would also fit the bill, because their clients present them with new and unforeseen problems that demand deeply nuanced responses. Creativity on its own will not be sufficient to secure one’s job – my wager is that the shoe design of Designer Schumacher will eventually be automated – but it will be extraordinarily usefulin the increasingly difficult quest to remain employed.
Education for the Near Future
If the job market of the future will favour high-level, creative mental skills, we should briefly consider how best to prepare ourselves for such jobs. How does one learn to question current assumptions, challenge the status-quo, dream up the not-yet-actual, and critically evaluate one’s progress while doing so? One of the best preparations for this is a liberal arts education. In stark contrast to this view, many parents and nearly all governments see the ideal college education as essentially a vocational school, with a 1-to-1 correspondence between one’s degree and one’s career. As we enter the machine learning explosion and a time of great vocational uncertainty, this 1-to-1 strategy becomes, economically speaking, increasingly imprudent. Putting all your educational eggs in one vocational basket is a risky venture when baskets are being automated out of existence. Better to study subjects like literature, biology, philosophy, and mathematics that give you a broad, flexible, creative foundation on which to adapt to the market transformations of the next several decades (specialising when needed with a professional degree or online courses). Both the students and the educational institutions that embrace this approach will be in the best position to thrive in the economy of the future.
What conclusions can we draw about how meaningful the future of work will be? If highly creative white-collar work is the future of work, this is a remarkably encouraging situation. Thinking about what makes work meaningful in the first place will reveal why. In our discussion of the Schumacher family, we saw that both the Cobbler and the Designer were craftsmen. In The Human Condition, Hannah Arendt takes a page from Plato and notes that as a craftsman works, he has the idea of the whole, finished product in mind, while the factory worker does not. This means that for a factory worker – or for any other kind of worker disconnected from the finished product – the goal that they have in mind as they work is money. That is, the meaning of their work, their work’s connection to something beyond itself, is a pay check. By contrast, the craftsman plays a part in determining the end product, and also has that end product in mind as he works, so this purpose permeates his work with meaning. For the craftsman, the meaning of his work is found not only in a pay check but also in the product of the work itself. This provides us with a more precise definition of at least one kind of “meaningful work”. Such work is meaningful 1) when it contributes to some larger goal that one cares about, and that one has played a part in setting, and 2) when it contributes to that goal in a way specific to one’s activities. The contrast to this “specific contribution” is seen when a person works only for a pay check: their work does indeed contribute to a larger goal that they care about (their pay check), but it does so in a completely general way, i.e., in the same way that any other activity with the same pay rate would. Conversely, the creative worker who helps set the end goal and works with this goal in mind can find her activities meaningful in her particular contribution to that goal. All of this helps explain why Cobbler and Designer Schumacher found meaning in their work, while Gluer Schumacher did not.
The above account of meaningful work is, to be sure, hasty and incomplete. For example, the “factory worker/craftsman” distinction is more a spectrum than a dichotomy (I once worked a factory job that required some level of craft), and factory workers often find some meaning in contributing to an excellent product, or simply having earned their pay check. Additionally, one might work for an aid organisation or a school and find meaning in contributing to their mission through whatever activities one is assigned, not just through the specific, creative activities of one’s imagination.
But our focus here is on the meaningfulness of the work most resistant to automation in the next few decades. With this in mind, we can see how the history of automation is, on the whole, working in the interests of human meaningfulness. Industrial automation did indeed destroy the often-meaningful work of medieval craft, but further developments in industrial (and office) automation have been replacing many kinds of demoralising jobs, freeing humans up to engage in more fulfilling work. In 1821, Hegel put his finger on this strange and ironic movement of history. After speaking about factory work’s dulling of the human mind, and about machines taking over this labour, he states, “Through the completion of this mechanical progress, then, human beings become free once again … The human being is at first sacrificed, but then emerges again as free …” Machines first needed to enslave us, in order to later free us up for more fulfilling work.
The jobs of the next few decades promise to be more meaningful not only on the whole, but also for those who already find meaning in their work. This will happen through automation augmenting some jobs, as opposed to replacing them. For example, when we can automate the repetitive tasks of the cancer therapy specialists mentioned above, they will be free to devote more time to the less researched cancers and to outlier cases that might otherwise have slipped through the cracks. In education, when we can automate the repetitive tasks of middle school teachers (such as the pure information transfer of lecturing), they will be free to devote more time to working with smaller class sizes in a more discussion-based format. Society will benefit tremendously from these kinds of hyper-individualised human services that, due to lack of time and resources, currently do not exist. And many people working in those jobs will find them more fulfilling as well.
Peering into the year 2050 and beyond, the future of automation and meaningful work becomes far hazier. Both Aldous Huxley and J.M. Keynes predicted that as technology advanced in the twenty-first century, even the most well-prepared humans would find themselves with less and less work to do. Many thoughtful futurists today agree. But for the near future, automation holds great promise for meaningful work, i.e. for vocational goals freely chosen as something good even apart from compensation. While all kinds of jobs can be automated and done for us by machines, the work that can bring us a sense of meaning and purpose is the work done by us. Work can be automated, but meaning, by definition, cannot.