Exponential: Order and Chaos in an Age of Accelerating Technology

by Azeem Azhar

During the course of my career, I’ve watched well-informed people pooh-pooh mobile phones, the internet, social networks, online shopping and electric vehicles as niche playthings destined for eternal obscurity. Over two decades I’ve observed executives in established industries regularly, perhaps even deliberately, look at the spread of a new product or service and dismiss it. Often it was because the absolute numbers were small, in spite of signs of hockey stick growth.

Every self-driving car company has missed its targets. It turns out that the problem is much harder, from a purely technical perspective, than the teams building the technologies were willing to acknowledge. When you jump into the car for a quick trip to the grocery store, you make roughly 160 decisions for every mile you travel. While that might not sound like a lot, when the decisions are based on a near-limitless number of variables, the scale of the challenge comes into sharper focus.

When people got into a shop queue, they would once have spent the time browsing the goodies for sale at the counter – and gum was the obvious choice. Now they were spending that time playing with their phones. So gum sales plummeted.25 Nobody saw that one coming. Predicting the impact of the iPhone on grocery-store gum sales would have needed a modern-day Nostradamus.

In the early days, this exponential growth was something the public and policymakers – in America and Europe, at least – proved unable to grasp. Politicians from Donald Trump to Boris Johnson consistently downplayed the risk exponential growth represented. Early research, released during the first year of the pandemic, demonstrated exponential growth bias at play. At all stages of the pandemic, people underestimated the future course of the spread. Given three weeks of actual data for the growth of the virus, participants were asked to predict infection levels one week and two weeks later. Despite having the data to hand, people guessed the future trajectory of the disease very poorly. On average, participants underestimated the first week by 46 per cent, the second by 66 per cent.26 Making predictions in the Exponential Age is hard – and we often get it wrong.

If you have ever worked at a very large company, you’ll be familiar with how bureaucracy creeps in and slows everything down. The economist Ronald Coase described this predicament well: in his rendering, organisational costs grow as firms get bigger and come to act as a force of gravity, slowing a company’s expansion and eating away at the advantages of scale.

Google’s market share of search queries is almost 80 per cent in the US, 85 per cent in the UK, and nearly 95 per cent in Brazil. In the smartphone market, Android is installed on four out of five phones globally, with Apple’s iOS used on the vast majority of the others. No other operating system really figures. This dominance is even more pronounced among certain demographics and age groups. In the US, more than 85 per cent of teens own an iPhone.7 In online advertising, Facebook and Google between them account for more than 90 per cent of all global spend.

Such network effects are one reason why many Exponential Age firms, like Facebook, PayPal, Microsoft, Google and eBay, are so large and successful. As we’ve seen, these companies were made possible by the exponentially increasing power of computing. But the network effect is what drives them to ever-bigger gains. Most people who use a consumer social network use Facebook because that is where everyone else is. Businesses accept Visa or Mastercard because they are the platforms with the most buyers on them. This positive feedback loop makes the business better and stronger as it gets bigger.

Network effects aren’t just the preserve of profit-seeking firms. Wikipedia, the free online encyclopedia managed by a foundation, also benefits from network externalities. Anyone can create competitors to Wikipedia, and there are many niche ones (such as Wookieepedia, dedicated to the Star Wars universe). But Wikipedia is where the majority of contributors gather to share their wisdom and critique each other, so it is where readers congregate. And because this is where the readers gather, Wikipedia becomes the space where contributors want to write. It is a positive externality: every additional contributor brings in more readers, which in turn attracts more contributors.

In fact, the World Wide Web itself benefited from the network effect. When Tim Berners-Lee first developed the web in 1989, there were several contenders for methods of storing information on the internet, like Gopher and WAIS. There were also commercial products like GEnie, CompuServe and Delphi. In 1994, Microsoft even targeted the market with its own offering, Microsoft Network. But Berners-Lee’s web became not just dominant but the sole viable information network on the internet, because of network externalities – once a lot of people were there, everyone else had to congregate there too.

Wherever there are network effects, there is the chance of a winn...

Once a company has established itself as the market leader, it becomes extremely difficult to challenge it. And this effect is being driven by the consumers themselves. It is in the consumers’ own interest to join the biggest networ...

The speed of the shift to intangible assets has been remarkable. In 1975, about 83 per cent of the market value of companies on the Standard and Poor 500 stock market index comprised tangible assets. Intangible assets accounted for the remaining 17 per cent. By 2015, those proportions had reversed. Only 16 per cent of the S&P 500’s value could be accounted for by tangible assets, and 84 per cent by intangibles.

In 20 years, Salesforce has grown from a handful of founders to a $200-billion business with nearly 50,000 employees. But as it has expanded, the company has figured out how to make each additional employee more productive – and how to get more money from its customers. In 2005, each of Salesforce’s then 767 employees supported $230,000 in revenue. In 2020, each of its 49,000 employees supported nearly $350,000 in revenue. Customers seemed to love Salesforce too. In 2010, the average customer paid Salesforce $18,000. By 2020, that had risen sixfold. Not only were customers paying six times as much, they were happier. In 2010, 81 per cent of customers renewed their business with the firm. By 2020, more than 9 in 10 did.

The easily scalable nature of intangible assets means that Salesforce can grow and grow, without taking on that many more staff – or becoming unwieldy.

Peter Thiel, one of Hoffman’s co-founders at PayPal, also has an ideology that focuses on growth. In Thiel’s view, ‘competition is for losers’.34 He recommends instead that founders identify markets where they are substantially better than any existing competition – so they can gobble up market share in a way that would have been unthinkable just a few decades ago.

What Thiel and Hoffman identify, in their respective ways, is that the superstar firms of the Exponential Age don’t just aspire to become large and dominant. They have no choice: second place is a very, very distant second.

Between 1997 and 2006, AI academics were 100 times more likely than life scientists to move from academia to industry.47

The rise of newly automated workplaces raises the prospect of mass redundancy. And it is framed as a more existential threat than Keynes’s fears of technological unemployment. Soon, we are told, we’ll reach a point where automated systems will render most of us unemployed and unemployable.

But the notion of a jobless future – the ‘robopocalypse’ of tabloid headlines – is overstated. It is alluring and it makes the news, yet it is muddle-headed. Historically, our economies have become more automated. And historically, employment levels have tended to increase.

How is this possible? Because automation has the potential to create more work than it destroys. Yes, there may be unemployment in the short term as some sectors wane. But automation creates jobs which often require new and distinctly human skills: from the programmers who develop systems to those who operate and maintain them. Over time, automation will end up inventing whole new sectors of the economy – ones that we can now only imagine.

When it had its largest workforce, in 1980, General Motors employed 900,000 people. It sold more than 4 million cars a year, with revenues approaching $66.3 billion.18 Each employee accounted for about $74,000 in sales. Now consider the biggest firms of the Exponential Age. Alphabet, which owns Google, employed about 120,000 people in 2019, against revenues of $162 billion, clocking in at $1.4 million in sales per employee.19 The overall trend is towards more valuable companies with fewer employees.

We’re faced with a conundrum. On the one hand, automation seems to be threatening a large fraction of the workforce. On the other, the better the technologies get, the more jobs there seem to be. What is going on? There are a few possible explanations. The first is that automation is perhaps much harder than it seems – for all the talk of an imminent robot world, we’re actually at a fairly early stage in the automation process. Although technologies are improving, many are taking longer than expected to become superhuman. Exponential processes take time to marinate.

the first AI-powered self-driving vehicles drove in the very measured environments of Phoenix, Arizona – with its wide, straight roads and perfect weather. The frantic, rainy autobahns of Germany will remain a tough ask for a while. When self-driving trucks are first launched, they will stick to straight highways, rather than manoeuvring through narrow roads in the City of London. Full automation, it seems, remains some way off.

Amazon is possibly one of most robotised large companies in the world – with an astonishing one physical robot for every four workers.

In the six months after the World Health Organization declared the coronavirus outbreak a pandemic, Amazon announced four waves of hiring, amounting to a staggering 308,000 new jobs globally in one year.

All this means that we’re left with a slightly different picture of our supposedly jobless future. The more that superstar firms like Amazon and Netflix automate, the bigger they grow; the bigger they grow, the more people they employ. There’s an exponential process here, but it doesn’t lead us to employee-free corporations.

A recent survey of 587 manufacturing companies in France supports the notion that the real threat from automation is a traditional one: the competitive threat from rival companies.

But in most cases, new roles would be created in other parts of the firm – resulting in a gain in employment.

A 10 per cent increase in robot adoption by a firm was associated with a 2.5 per cent decline in employment at its competitors.38 It was not automation itself driving job losses, but the difficulties faced by the companies that didn’t automate.

It goes against the folksy economic belief that there is only so much work to go around, and that upsetting the equilibrium of the labour force – by increasing female labour participation, or allowing immigration, or using robots – will reduce the available work for workers. But this belief is nonsense. It’s a form of zero-sum thinking that has largely been dispensed with by economic theory and historical evidence. Economists call it the ‘lump of labour fallacy’.

Automation created more opportunities, not just for Sid but for the rest of the team. With the easy tasks that comprised most of his workday now being done by code, that allowed Sid to turn to more complex, and important, tasks – many of which had previously been going unattended. And as he turned his attention to these tasks, they in turn generated more opportunities and more work.

The historical record is unambiguous. Technologies have created more jobs than they have destroyed,

Within a few years, crowdsourcing platforms had multiplied. Services like Elance and oDesk sprung up for complicated tasks like programming or copywriting; Fiverr and PeoplePerHour were created for tasks that were less complex than programming but more complex than an Amazon HIT.

TaskRabbit, now owned by furniture giant Ikea, will today dispatch someone to help you assemble your new bookcase. Talkspace will help you find a therapist.

The platform UpWork is a great example of how the gig economy can help workers access the global economy, often in part-time roles.

In practical terms this has manifested through stagnating average wages and increases in inequality. Again, the most striking example comes from the US. Between the 1940s and the mid-1970s, economic productivity and workers’ pay rose in tandem: from 1948–1973, there was a 97 per cent increase in workers’ hourly pay, against a 91 per cent increase in economic productivity. Fair enough. But then, something surprising happened. The increase in wages tapered off – even as economic productivity continued to skyrocket. By 2018, US economic productivity was 255 per cent higher than it had been in 1948; but workers’ pay was only 125 per cent higher – barely a third more than it had been in 1973. In other words, the output of the US economy continued to rise – but the share workers received stagnated.

The cause of the decline of labour’s share of the economic pie is multifaceted. But it is closely related to the shift to the exponential economy. Four key causes stand out. Globalisation, which drove down wages in the West as companies offshored jobs to cheaper locations across the world. The decline of unions, which meant workers lacked the bargaining power to stop economic rewards going to the owners of capital (we’ll return to this in a moment). The rise of the intangible economy, which reduced the relative value-add of the average worker – more value was being created by know-how, software and data, stewarded by smaller numbers of highly specialist workers, than by the human sweat of the larger parts of the workforce. And superstarification – as markets consolidated around ever-fewer superstar firms, there was less competition for labour and so workers had less leverage.

These are all hallmarks of the rise of exponential technologies – the emergence of a global, high-tech, intangible economy, dominated by a handful of big firms. According

The average software engineer at Uber was paid $147,603 a year in 2020. Senior engineers with five or more years of experience might make three times that.

On the flip side, there are Uber’s drivers. As we’ve seen, gig workers are often paid relatively little. Your typical driver will make $19.73 per hour before expenses, or $30,390 a year if they drive 40 hours a week.

There is a similar dynamic at work in Facebook. Half of all employees at Facebook, from engineers to marketers, accountants to salespeople, make $240,000 a year or more.79 Facebook’s content moderators, who are not employed by the firm but rather contracted via temping agencies, are paid on average $28,000 per annum. Ordinary people wh...

All this points to the changing topology of employment in the Exponential Age. This is an economy where intangible assets – the kind produced by well-educated knowledge-workers – are all-important. Those with high levels of education are compensated handsomely. At the same time, there remains a group of less well-rewarded, less highly skilled workers, who may not even be acknowledged as employees. In aggregate, the result is a reduced share of income that goes to employees. Middle-wage earners, who used to be the engine of Western economies, are evaporating.

In the future, however, manufacturing can happen near to the consumer. Thanks to the wonders of 3D printing, components can be produced locally; while the design might come from anywhere, the finished product can be crafted in a local workshop and handed to a customer who lives close by.

As of 2020, vertical farms have a tiny share of the food market. But the market for high-intensity vertical farms is growing at more than 20 per cent per annum, on the march up our exponential curve.9

But renewables have now put every nation on a path to energy independence. Once wind turbines are installed or a solar farm is deployed, they require few raw materials – and, as we saw in Chapter 2, such power supplies are fast becoming ubiquitous. This shift to renewable energy drastically reduces the amount of ‘stuff’ that needs to be carted around.

The average electric car stores about 50 kilowatt-hours of electricity: enough to run the typical British or American home for five days. It will become commonplace for our electric cars to lend their stored electricity to our homes when it is dark. Britain alone is forecast to have as many as 11 million such cars by 2030.

If each owner were willing to share a bit of the surplus energy stored in their cars with their neighbours, it might cover the whole country’s needs.

For Moixa, his company, to buy the batteries itself, it would need large amounts of capital, perhaps running into the tens of millions of dollars. Instead, he’s persuading owners of electric cars to connect to his network. Together, these idle car batteries form a giant virtual power plant. Daniel’s platform manages them and uses sophisticated algorithms to balance usage across the whole network. At last count, he had managed to combine 20,000 batteries together in several Japanese cities.12 That’s enough to power 25,000 Japanese homes for a day. It is like alchemy – replacing a massive power station, smokestacks rising into the sky, with a web of cars, parked on driveways, keeping homes running as we sleep.

Josh Hoffman is the CEO of Zymergen, the biotech start-up. His company combines advanced machine learning with clever genetics to persuade microbes to efficiently produce industrial materials. Their first product, Hyaline, will be used in the screens of smartphones. Hoffman’s bugs use natural processes to elegantly grow the screen film, requiring much less energy than making plastics from oil. It is a method that effectively removes hydrocarbons from the production of prosthetic materials.

Cutting-edge manufacturing processes require fewer workers than older methods – which means, for the first time in decades, it makes economic sense to manufacture in places with high labour costs.