Exponential: Order and Chaos in an Age of Accelerating Technology

by Azeem Azhar

But the pandemic also hammered home some of the key points of this book. The spread of the virus demonstrated that exponential growth is hard to control. It creeps up on you and then explodes – one moment everything seems fine, the next your health service is on the verge of being overwhelmed by a new disease.

He eventually settled on the estimate that, every 18–24 months, chips would get twice as powerful for the same cost.

I define an exponential technology as one that can, for a roughly fixed cost, improve at a rate of more than 10 per cent per year for several decades.

Additive manufacturing is still a tiny business. You’ll find it in prestige products and in highly specialist sectors of the economy – lightweight parts for fighter jets, or medical implants. But the underlying technologies are on an exponential course. Researchers estimate that most additive manufacturing methods are developing at a pace of between 16.7 per cent and 37.6 per cent every year, with the average rate falling in the high thirties.16 Over the next 10 years we will see performance improve 14 times – and, of course, see prices drop concomitantly. Terry Wohlers, an analyst of the additive manufacturing sector, tells me that the 3D printing market grew 11 times in the decade to 2019 – a rate of 27 per cent per annum.17

As we saw in the last chapter, Moore’s Law is a social fact, willed into existence by industry. And that means human behaviour is key: if we stopped endeavouring to make Moore’s Law true, it would cease to be true.

His theory was that for every doubling in units produced, costs would fall by a constant percentage. The exact nature of the decline would depend on the engineering in question. In the case of the aircraft Wright studied, it was a 15 per cent improvement for every doubling of production. This 15 per cent improvement is known as the ‘learning rate’.

For this reason, the key to the continuation of Wright’s Law is increasing volume.

Exponential technologies are being driven by three mutually reinforcing factors – the power of learning by doing, the increasing interaction and combination of new technologies, and the emergence of new networks of information and trade.

This dynamic means that, in practice, much of the labour economists consider ‘unskilled’ might prove difficult to automate. A workplace manual is usually barely even a rough guide to what a job is – it doesn’t cover half of the things you need to be successful. And when such tacit knowledge exists in a workplace, it makes artificial intelligence that can do the job very hard to build. An AI system needs a clear and unambiguous goal, and modern systems need to be trained on that data. If the know-how about a job is largely hidden, an AI system will be trained on only half the picture. In short: if an environment is designed for a person, it is likely to be too complicated for machines now or any time in the near future.

Amazon’s example reveals that, on the level of individual companies, automation can create more jobs than it destroys.

If a firm is like Fred – that is, it is able to move quickly to adapt to changing circumstances – it may well thrive and grow in the age of automation, hiring more workers. However, if firms or organisations misread the nature of Exponential Age changes, their employees will get eaten by the bear. Blockbuster, the video rental service, was driven out of business by the rise of Netflix, a company originally

created to send DVDs through the mail. In 2013, the last 300 Blockbuster outlets shut down, putting thousands of Blockbuster employees out of work, and Netflix emerged as the dominant player in the video rental space.37 But those jobs were lost not because Blockbuster built some amazing automated service that reduced workforces in the 9,000 stores they operated at their peak. Rather, Netflix came up with a digitally enabled business model, first allowing customers to subscribe to DVDs from a website, and later offering them shows and films online. People preferred Netflix’s offer; Blockbuster was slow to adapt and went bust.

‘firms adopting robots … became more profitable and productive’. They also created jobs, increasing employment by 10.9 per cent. Whether the total number of production workers increased tended to depend on how fast the firm’s sales grew. But in most cases, new roles would be created in other parts of the firm – resulting in a gain in employment. The problem was not so much the forward-leaning companies, but the laggards. 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.

Crucially, this dynamic only emerges in the longue durée. The historical record is unambiguous. Technologies have created more jobs than they have destroyed, but the short-term damage can be profound. Carl Frey points out that: ‘Throughout history, the long-term benefits of new technologies to average people have been immense and indisputable. But new technologies tend to put people out of work in the short run, and what economists regard as the short run can be many years.’44

Overall, though, the lasting impact of automation will not be the loss of jobs. If we’re looking at the long-term fallout of exponential technologies, our concern should not be with the quantity of work around for humans to do. It should be with the quality of options that are available.

This problem is visible in the precarity of gig work, and the treatment of employees as fungible assets to be controlled through management algorithms. But its clearest manifestation lies in how the rewards are doled out. Economists have long tried to understand how equitable a labour market is by looking at what share of a country’s income goes to workers, and what share goes to the owners of capital – through stock gains, dividends and corporate profits. The story of the last 50 years is striking. Between 1980 and 2014, labour’s share of national income – the percentage of GDP paid out in wages, salaries and benefits – declined on average by 6.5 per cent, as measured in 34 advanced economies. In the US, the decline has been even more staggering. In 1947, workers received 65 per cent of the national income in America; by 2018, this had declined to 56.7 per cent. More than three-quarters of the decline since the end of World War Two occurred in the first two decades of the millennium. For decades, workers have been systematically getting a smaller share of the economic pie.75

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.76

‘The bigger the city the more each person earns, creates and innovates and interacts – and the more each person experiences crime, disease, entertainment, and opportunity – and all of this is at a cost that requires less infrastructure and energy for each of them,’ West says. ‘This is the genius of the city.’32

At their most extreme, these pathways can lead people to violence. A report from 2018 set out to investigate whether Facebook was facilitating ISIS propaganda, and showed how important social networks have been in recruiting new members to the terrorist group. It is a study in homophily: people are gradually pulled from the mainstream of political life into more and more extreme communities. Befriending one ISIS supporter leads Facebook to recommend a dozen other pro-ISIS accounts. ‘Facebook, in their desire to connect as many people as possible, have inadvertently created a system which helps connect extremists and terrorists,’ concluded the researchers, Gregory Waters and Robert Postings.31

This type of open interchange is certainly technologically possible, if not completely straightforward. As Sinan Aral, an MIT professor who has studied social networks since the early 2000s, told me, networks ‘all have the same messaging formats now anyway: textual messages, video messages and stories’.38 To make this shift a reality, though, governments would need to act decisively. Nations could mandate interoperability, especially for services that reach a certain size – perhaps 10–15 per cent market share. By waiting until a company had grown to this scale, it would allow founders to innovate and explore markets – while signalling that future success will bring additional civic responsibilities. And the principle need not be limited to social networks. Imagine if a gig worker could bid for jobs across dozens of relevant platforms at once and carry the track-record they have earned from one service over to another. Or in the case of healthcare, where a patient could access their medical records independent of the provider behind them. In each case, power would slowly shift to the person and away from the platform.

One approach is to embrace the notion of data as property. As the pop star will.i.am wrote in The Economist in 2019, ‘The ability for people to own and control their data should be considered a central human value. The data itself should be treated like property and people should be fairly compensated for it.’39 The argument is appealing. By putting a price on our data, we could create a more efficient market for the digital labour we all undertake. Well-known figures ranging from Tim Berners-Lee, who invented the World Wide Web, to Alexandria Ocasio-Cortez, the Democratic congresswoman, have pushed the idea that pricing and compensating us for our personal data would help reform the data economy. But this would be, in my view, a mistake. As we’ve seen in this chapter, the creeping monetisation of private life is the problem. Do we want to further marketise our experiences on exponential platforms – to turn our messages to friends and family into a commodity, and allow Facebook to give it a price tag? You might also be disappointed to learn how much you would actually earn from this approach. Facebook made about $7 in profit for every user in 2019. If each user was to take half of the firm’s profit, that would net them the price of a coffee. A year. The truth is, an individual’s data is worth little – sorry, Mr i.am, even yours.

Imagine a world where data resided in data commons. Certain types of information about our lives would be collectively held, in a trusted environment, and used for our benefit. Some data commons are already in existence. The UK Biobank makes excellent use of the millions of data points about our genes that are created each year. It collates the medical and genomic data of 500,000 people. Each has consented to researchers using their data to better understand and prevent serious disease. And any legitimate researcher can access this resource. In fact by 2020, more than 20,000 researchers were doing so. The goal of the UK Biobank is to enable novel scientific discovery with the consent of those who have contributed their data, rather than merely to maximise a commercial return. It is a sharp contrast to how data is often currently managed: hoarded by one party and auctioned off to a small number of well-heeled bidders.

Yet this is only half the picture. The technology offers up new possibilities. But technology alone will not solve our problems. We should be concerned with three issues. The first is that an abundance of stuff – energy, materials, healthcare, whatever – doesn’t mean an absence of waste. Sure, some countries have decarbonised their electricity supply and broken the link between carbon output and GDP. But that doesn’t eliminate resource use altogether. Solar panels and chips need sand; batteries need metals that must be yanked out of the ground; urbanisation will require buildings, roads, sewers and more. In fact, history has shown that the cheaper something becomes, the more people can afford it, and so the more demand there is. Supply catalyses demand: my new shelving, bought to help me tidy the piles of books on my desk, resulted in books both on straining shelves and in unruly piles on my desk. Exponential technologies by no means preclude excessive resource extraction.

It is a shift with two main stages. The first is to acknowledge that technology’s shape, direction and impact are not preordained. Of course, technology builds on what has come before – new innovations layer and combine with those of earlier generations. But its path is not set. We are the ones who decide what we want from the tools we build. This is shown most clearly in the divergent ways technologies are used by different societies. Consider DDT, which was very helpful at killing mosquitos and slowing the spread of malaria, but also polluted food chains around the world. DDT was banned in the United States and United Kingdom. Yet for decades it was allowed in India, where preventing malaria was a priority – and so ecological damage was deemed an acceptable downside. Gig work tells a similar tale: in the UK, a rich country with a well-established formal labour market, gig-working platforms may be seen to unacceptably undermine workers’ rights and protections. In India or Nigeria, the same technologies may bring benefits, as these countries have a more informal work market based on day labour. In other words, it is our choices and our circumstances that determine how technology is actually used. As the historian Melvin Kranzberg put it: ‘The point is that the same technology can answer questions differently, depending on the context into which it is introduced and the problem it is designed to solve.’

them. As Jasanoff puts it, ‘In all of its guises, actual or aspirational, technology functions as an instrument of governance.’ It moulds all of the ethical, legal and social systems on which we depend.13 We cannot prevent this moulding happening, but we can direct it.

When water turns to steam, we can still make use of its power. But we need new tools to handle it – or we risk getting scalded. And so we need a new set of tools to manage our phase transition into the Exponential Age.

nyc-yellow-cabs-in-just-six-years-2019-08-09> [accessed 7 January 2021]. 49 Kelle Howson et al., ‘Platform Workers, the Future of Work and Britain’s Election’, Media@LSE, 11 December 2019 <https://blogs.lse.ac.uk/medialse/2019/12/11/platform-workers-the-future-of-work-and-britains-election/> [accessed 7 January 2021]. 50 James Manyika et al., Connecting Talent with Opportunity in the Digital Age (McKinsey & Company, 1 June 2015) <https://www.mckinsey.com/featured-insights/employment-and-growth/connecting-talent-with-opportunity-in-the-digital-age> [accessed 6 October 2020]. 51 Neil Munshi, ‘Tech Start-Ups Drive Change for Nigerian Truckers’, Financial Times, 26 August 2019 <https://www.ft.com/content/c6a3d1f2-c27d-11e9-a8e9-296ca66511c9> [accessed 21 September 2020]. 52 ‘Upwork Reports Fourth Quarter and Full Year 2020 Financial Results’, Upwork Inc., 23 February 2021 <https://investors.upwork.com/news-releases/news-release-details/upwork-reports-fourth-quarter-and-full-year-2020-financial> [accessed 21 April 2021]. 53 Lijin Yeo, ‘The U.S. Rideshare Industry: Uber vs. Lyft’, Bloomberg Second Measure, 2020 <https://secondmeasure.com/datapoints/rideshare-industry-overview/> [accessed 23 September 2020]. 54 ‘Gig Economy Research’, Gov.uk, 7 February 2018 <https://www.gov.uk/government/publications/gig-economy-research> [accessed 21 September 2020]. 55 Ravi Agrawal, ‘The Hidden Benefits of Uber’, Foreign Policy, 16 July 2018 <https://foreignpolicy.com/2018/07/16/why-india-give...