Fully Automated Luxury Communism A Manifesto

Just as factory jobs were eliminated in the twentieth century by new assembly-line robots, Brad and I were the first knowledge-industry workers put out of work by the new generation of ‘thinking’ machines. ‘Quiz show contestant’ may be the first job made redundant by Watson, but I’m sure it won’t be the last.

That was an insightful conclusion. While machines had bested humans at things like chess and solving maths problems – feats we typically associate with genius – they did so by brute force, completing incomprehensible numbers of calculations. Deep Blue assessed 200 million chess positions per second – a colossal number made possible by riding the wave of Moore ’s Law and exponential progress. Those trends have only continued, meaning that today you can download a chess engine programme like Houdini 6 for your home computer and it would beat Deep Blue almost every time.

And yet a paradox has emerged. It has become clear that more ‘processor power’ is actually required for managing what we have historically considered to be low-level tasks for humans, such as motor-sensory coupling, spatial awareness and unanticipated responses. In other words, it is harder to build a machine that can wash the dishes than one that can solve complex mathematical problems. This contradiction is known as Moravec’s Paradox, after the technologist who defined it. From the perspective of technological unemployment it was a hugely important observation, showing how even ‘low-skilled’ jobs, from construction to fruit picking, could remain immune from automation. Even as machines beat chess grandmasters and former supercomputers found their processing power equalled by \$400 games consoles, they could barely walk up a flight of stairs.

For a while this paradox appeared intractable. Even at the turn of the twenty-first century, some fifty years after the

Full Automation

81

Third Disruption began, the possibility of a machine with even the balance and coordination of a small child seemed remote.

But then the impossible suddenly became inevitable. Enter Atlas, the robot who learned to somersault.

Atlas Somersaults

If you go to YouTube and type ‘PETMAN prototype ’ into the search bar, the first video that appears, posted in October 2009, is a demonstration of a biped robot developed by Massachusetts-based company Boston Dynamics. Awkward and attached to several cables, PETMAN looks like the lovechild of a subwoofer and Bambi on ice.

Now type in ‘What’s new, Atlas?’ On your screen will appear a video of another robot manufactured by the same company. Only this video was published in late 2017 and the robot isn’t just walking without cables, it’s doing box jumps and backflips. It doesn’t end there; elsewhere on the company’s YouTube channel you can see videos of Atlas jogging outside or doing ‘parkour’ as it jumps three successive steps of forty centimetres each without breaking stride. This would appear to indicate that Moravec’s paradox is close to being overcome, with machines able to match humans in fine dexterity and spatial awareness sooner than we think. The descendants of Atlas another nine years from now may plausibly have the kind of coordination typically associated with an ice skater, gymnast or sculptor.

The reason why is simple: the progress from PETMAN to Atlas is underpinned by the improvements outlined in the second chapter, as we see exponential gains in the price– performance ratio of digital technologies, from cameras and sensors to chips, and the experience curve in areas such as energy storage. A case in point: until 2015 Atlas had to be permanently plugged into a wall socket. Now, with its 3.7-kilowatt-hour lithium-ion battery pack, it can walk around for about an hour. These trends are only set to continue.

82

n e w t r av e l l e r s

But while robots whose movements authentically resemble those of humans aren’t quite here yet, another category of machine – drawing on the same gains in digitisation and the dividend of exponential progress – is on the verge of transforming whole industries. It is the leading edge of a transformation which will mean not only the loss of countless jobs, but entire professions. And just like the acrobatics of Atlas, nobody saw it coming – until it was right in front of them.

Autonomous Vehicles

In 2002 the American defence agency DARPA announced a ‘Grand Challenge ’ for driverless cars scheduled to take place in the Mojave Desert in spring 2004. The proposed route was two hundred and forty kilometres long and the prize, for whichever car finished first, was set at \$1 million.

While some of the most brilliant minds in America applied themselves to the task, not one of the fifteen teams present at the start line was able to complete the course. The ‘winner’, built by Carnegie Mellon University, was only able to successfully navigate 5 per cent of the route. While the challenge had been ambitious – after all, the point was to stretch the entrants’ abilities – few thought it would descend into such farce. One observer even labelled the episode ‘the debacle in the desert’. To any reasonable person the possibility of autonomous vehicles seemed decades away.

And yet, just six years later in 2010, Google announced their self-driving cars had ‘logged in over 140,000 miles’ with seven test vehicles completing over 1,000 miles each without any human intervention – including dificult terrain like San Francisco’s notoriously steep Lombard Street. Since then the likes of Apple, Tesla and Uber have entered the game, not to mention the older incumbents of the automobile industry. By 2016 Uber’s then-CEO Travis Kalanick was clear about the importance of self-driving vehicles for any transport company: ‘It starts with understanding that the world is going to go self-driving and autonomous … what would happen if we weren’t a part of that future? If we weren’t part of the autonomy thing? Then the future passes us by.’ In the span of just eleven years the technology underpinning autonomous vehicles had improved so dramatically that they went from a totem of public ridicule to influencing the business models of some of the world’s most valuable companies.

Full Automation

83

That is how exponential technologies work: ponderously at first, and then a sudden transformation – a tendency historically visible with personal computing, smartphones, the internet and soon the descendants of Atlas. For now, however, the technology that will turn self-driving cars from engineering possibility to background feature in our everyday lives remains to be perfected.

Importantly, the way this challenge is being approached by the likes of Google and Uber ofers an insight into how automation may difuse across other parts of the economy and eliminate jobs. The strategy runs something like this: begin by acquiring massive amounts of data to allow algorithms to model and reproduce outcomes and work their way through highly repetitive tasks. After that, incorporate machine learning which is able to respond to unexpected situations arising beyond the data viewed as otherwise typical. Combining these steps yields something which can perform a wide range of jobs – from complex surgery to picking fruit and even writing journalism.

84

n e w t r av e l l e r s

Such an approach is feasible because processor power is constantly improving and data sets are getting larger every second. But the nature of jobs under capitalism – comprised of tasks rather than the generalist approach seen with artisanal labour – also plays a part. Industrial change, particularly since the 1880s, has meant each job is reduced to a managed set of components, all of which are measured and managed as scientifically as possible. Without knowing it, the project of Frederick Taylor and his productivity revolution – for Drucker the first step in making information the primary factor of production – has turned out to be just as crucial to peak human as the exponential progress of digital technologies.

Autonomous vehicles ofer an instructive example. To create cars that drive themselves, the likes of Uber, Tesla and Google didn’t model and then replicate how humans drive – this remains well beyond our present technology. Rather, they tried to solve the problem by breaking it down into a set of component operations and putting a data processing system on wheels. As a result these vehicles can navigate streets and motorways by relying on precise GPS data, huge amounts of information regarding maps, and a continuous stream of realtime updates on other cars, potential obstacles, pedestrians and all the variables human drivers have to consider. All of this is achieved with a myriad of sensors, lasers and cameras processing information as 1s and 0s.

Even in isolation the arrival of autonomous vehicles likely spells the disappearance of whole professions. In 2014, driving accounted for around 4 million jobs in the US alone, and according to a report by Goldman Sachs the country could see job losses at a rate of 300,000 a year as autonomous vehicles become an integrated feature of modern society. From the perspective of business that would be entirely understandable: logistics vehicles running twenty-four hours a day, seven days a week, ofer massive savings. And while there is a temptation to say machines can’t be liable for accidents, with over 1.3 million annual road deaths worldwide, and 40,000 in the US alone, it won’t be long before the technology is suficiently advanced that such an argument could be reversed. That’s before mentioning taxis, buses, trains, planes and warehousing. All of these industries will be impacted in a similar way, if at varying paces, and near-entirely automated in little more than a generation.

Full Automation

85

Technological Unemployment Is Coming

A 2015 study by the Bank of England isolated how technological change, in particular the rise of machine learning, would mean the loss of 15 million jobs in the UK – 40 per cent of the labour market – over the next few decades. Underpinning that would be the shrinking space for uniquely human skills, with this limiting any chance for workers to up-skill in response. A year later, the bank’s governor, Mark Carney, repeated those forecasts saying many livelihoods could be ‘mercilessly destroyed’ by technological change, and that ever-higher income inequality could be one of the major consequences.

Those findings confirmed the conclusions of an earlier report published by two Oxford University academics, Carl Benedikt Frey and Michael Osborne. In 2013 they claimed that 47 per cent of all US jobs were at ‘high risk’ of being automated, with a further 19 per cent facing medium risk. Elsewhere Peter Sondergaard, research director for the consultancy Gartner, predicted that by 2025 one in three jobs will be automated as the result of an emerging ‘super class’ of technologies, with general purpose robotics and machine learning leading the way. Finally, in a 2016 report to Congress, White House economists forecast an 83 per cent chance that workers earning less than \$20 per hour will lose their jobs to robots in the medium term.

86

n e w t r av e l l e r s

The Bank of England, Oxford University, a global technology consultancy and the United States Congress are far from siren voices that are easy to dismiss. This is the heart of the economics and business establishment. While not everyone agrees on the extent to which technology will create unemployment in the short term, even more conservative voices think unavoidable change is not far away.

Take the Millennium Project. Launched in the 1990s by several UN organisations, it expects global unemployment to increase to 16 per cent by 2030 before rising to 24 per cent by the middle of the century. While that is more guarded than the Bank of England’s predictions, or the claims of Peter Sondergaard, such a shift would more than test business as usual. A world of 10 billion people facing the challenges of climate change, ageing and resource shortages would endure levels of joblessness similar to those confronting Greece today – a country where 50 per cent youth unemployment has given rise to the most polarised society in Europe. Not only would such a scenario generate political and social turbulence on a global scale but importantly – and unlike with Greece – there would be no promise of a brighter tomorrow, however far away.

The most frequent rejoinder to all of this is that while the jobs of today may well disappear, others will emerge in their place. After all, that is what has always happened in the past. And yet that isn’t quite true. Eighty per cent of today’s professions existed a century ago, with the number of people employed in the 20 per cent of new occupations comprising only one in ten jobs. While the world economy may be much bigger now than it was in 1900, employing more people and enjoying far higher output per person, the lines of work nearly everyone performs – drivers, nurses, teachers and cashiers – aren’t particularly new.

Full Automation

87

Actually Existing Automation

In March 2017 Amazon launched its Amazon GO store in downtown Seattle. Using computer vision, deep learning algorithms, and sensor fusion to identify selected items the company looked to build a near fully automated store without cashiers. Here Amazon customers would be able to buy items simply by swiping in with a phone, choosing the things they wanted and swiping out to leave, their purchases automatically debited to their Amazon account.

Several months later Amazon acquired Whole Foods Market for \$13.7 billion. While that might have appeared a strange acquisition for a company whose core business is online retail, the purchase provided them with the supply chain capabilities to support Amazon GO and take aim at the \$800 billion global grocery market.

Company management plans to use six people per shift in each Amazon Go store compared to the seventy-two employees found in the average US supermarket. When you consider the labour costs as well as Amazon’s singular advantage in high automation warehousing – here too they are a world leader with their ‘KIVA’ robot – it quickly becomes clear the company could come to dominate areas of ofline retail just as they presently do online. That is, except in China, where in late 2017 the local retailer JD.com announced the opening of hundreds of ‘unmanned stores’ ahead of anyone else.

88

n e w t r av e l l e r s

Regardless whether it is Amazon or a rival who gains firstmover advantage, the trends are clear. The future of retail, as with logistics and warehousing, is automated. Yes, some jobs will remain, but when you consider that salesperson and cashier are the two leading jobs in the United States – and indeed most other countries – the prospect is a frightening one. Some might say customers want an emotional connection when they shop, and in certain contexts that may well be true, but most of the time the primary consideration will be the best product at the most afordable price. That will mean cutting labour costs wherever possible.

Rather than distant challenges, the retail industry now anticipates major layofs in the area. Before Amazon Go was even announced, the British Retail Consortium predicted almost a third of the country’s 3 million retail jobs would disappear by 2025, resulting in 900,000 lost jobs as companies turn to technology to replace workers.

As with self-driving cars and Atlas, all of this is possible because of extreme supply in information – from things like image and range sensors, to stereo cameras, deep learning algorithms, and the ubiquity of smartphones and online accounts. The same holds true elsewhere in the supply chain, from the warehousing robots using sensors and barcodes controlled by a central server, to the autonomous vehicles set to oversee distribution and delivery – whether by vehicle or drone.

But even among those who accept that common jobs like warehousing, retail, logistics and taxi-driving could be eliminated by advancing technology, there remains an insistence that jobs in ‘high value ’ services will somehow remain immune. Here too, however, the evidence increasingly indicates the truth is rather diferent.

Full Automation

89

Speaking at a technology event in 2017, Mark Cuban, the billionaire owner of the Dallas Mavericks, predicted that the world’s first trillionaire will be whoever masters commercial applications for AI, the reason being that artificial intelligence could prove particularly lucrative when applied to traditionally ‘white collar’ industries like insurance, software development or accountancy. ‘I would rather be a philosophy major,’ Cuban said of those training to enter such professions today.

Attention-seeking? Perhaps. Hyperbolic? Absolutely. Wrong? Probably not, because from cardiac surgery to calculating taxes, historically well-paid professions are just as repetitive and subject to the division of labour, and therefore automation, as anything else.

Take the da Vinci surgery robot. In 2017, University College London announced that this relatively low-cost machine had saved the lives of around 500 men with prostate cancer. While the robot itself is not automated – it instead grants a human surgeon far higher levels of dexterity and precision – the paths to automating a range of its regular operations resemble the blueprint for a self-driving car: you give a powerful data processor huge amounts of information, machine learning and a scalpel. The first part allows algorithms to model and reproduce outcomes and work their way through highly repetitive tasks, while the second allows for immediate and smart responses to unexpected situations.

In medicine, you can see how that would be applied to pretty much anything – from eye examinations to treating prostate cancer or taking blood. In areas more dependent on pattern recognition, such as radiology, machines have even more of an advantage. Radiologists use medical images like X-rays, CT and PET scans, MRIs and ultrasounds to diagnose and treat patients. While the field has greatly improved patient care over the last few decades, it has contributed to escalating costs and is relatively labour intensive. That is, until now.

90

n e w t r av e l l e r s