Artificial intelligence for international economists (by an international economist): Part 1 of 5

Richard Baldwin 07 December 2018



On the morning of 18 May 2018, I opened my window in Lausanne and in came a gentle breeze. On the morning of 18 May 1980, a landslide opened a volcanic window on Mount St. Helens and out came a superheated ball of gas that killed everything within an eight-mile radius. Fifty-seven people died. 

At some level, you could say both were examples of wind, but that would clearly be missing something. Revolution is not just evolution on fast-forward.

For technology, there is a world of difference between normal progress and explosive progress. What we are experiencing today is digital technology, or digitech for short, that is advancing at an explosive pace.  

I’m choosing the word 'explosive' on purpose. 

By one definition, an explosion is the input of pressure into a system at a pace that is too great for the system to absorb smoothly. The result is a chaotic disruption of the system. Our ability to gather, transmit, process, and store information is advancing this way. 

This explosive advance should matter for international economists for at least two reasons. 

1. Digitech is creating new forms of globalisation.

For the last couple of centuries, globalisation has been viewed as being mostly about goods – things we could send across borders on planes, trains and ships. Goods trade will continue to be important, but future globalisation will also be about the things that we do (services), not just the things that we make (goods). Specifically, digitech is creating a new form of globalisation that could be called 'telemigration' – people sitting in one country while working in offices in another, as I point out in my forthcoming book, The Globotics Upheaval: Globalization, Robotics, and the Future of Work (January 2019). 

2. We're not ready for it.

The nature and velocity of future globalisation means that few are prepared for what’s coming. It seems likely that our socioeconomic systems will struggle to absorb the new pressure. The result may well be an upheaval in some advanced economies – especially those like the US and the UK where social safety nets are lacking or are set too low. 

But why is this time different? Couldn’t we have said the same about steam power or containerisation? One explanation lies, strangely enough, in the physics of digitech.

These two key changes – the shifting focus of globalisation from goods to services, and the speed at which it is happening – can be explained by the physics of digitech. 

Past globalisation was governed by the laws of physics that apply to matter, since globalisation was mostly about goods. These laws set natural upper bounds on how fast things could change and thus created presumptions about what is ‘normal’ when thinking about globalisation’s advance. 

Future globalisation will be governed by a very different set of laws, since trade in services is mostly about electrons and photons. This difference matters. 

A different kind of physics and doubling times

The original globalisation was launched when steam took the horse out of horsepower and put it into the hands of manpower. This power radically lowered the cost of transporting goods. But, perforce, things moved more slowly. US imports and exports, for example, grew at about 3% per year in recent decades. At that pace, imports and exports double every 23 years – as does the competition, and the opportunities that come with expanded trade. 

That’s what globalisation looked like when I enrolled in the University of Wisconsin-Madison in 1976. In the time it took trade flows to double, I switched majors four times, got a BA, went to the London School of Economics, got a Master’s, worked at the UN, got a PhD at MIT, worked in the White House, and taught at universities for 13 years. 

When it comes to digital globalisation – the globalisation facing today’s workers – things work quite differently. My daughter, for example, enrolled in university in 2015. In the three years it took her to complete her BA, data flows and processing speeds doubled almost twice. The cross-border flows of economically relevant electrons and photons will double yet again before she finishes her Master’s. 

And it is not just the growth rates that matter. The absolute increments have gotten so large that they are hard to comprehend. Economists instinctively think about change in growth rates, but increments also matter. For example, if you could double the length of your stride with each step you took, your 10th step would be more than a kilometre long, and your 17th step would let you do a marathon in one stride. That’s a constant growth rate, but it is the absolute increments that makes that simple calculation shocking. 

Or consider the increments in computer processing power. When it was introduced in 2015, the iPhone 6s had 120 million times more power than the mainframe computer that guided the Apollo 11 mission to the moon and back in 1969. While that’s amazing, try this. The extra processing power that iPhones got between 2015 and 2017 was equal to over 240 million times the power of the Apollo mainframe. Here I’m not talking about the level of power, I’m talking about the increment in processing power in just two years. And the 2017 to 2019 increment in processing power will be almost a half billion times more than the speed of the Apollo computer.

Whether it is the length of strides or increments in processing power, those sort of increments in progress just don’t seem reasonable. They don’t seem plausible. And in the world where the physical laws of matter govern, they aren’t. But in the world of digitech, they are not just reasonable, they are inevitable. 

Exponential growth meets a brainbug

It is difficult to get one’s mind around this sort of explosive growth for a very good reason. Our brains, no matter how smart we are, are not really fit to deal with this type of growth at an instinctual level. 

Our brains evolved to understand motion in a walking-distance world, so we struggle to think through the implications of explosive growth. Here I don’t mean the sort of thinking economists do when they are calculating things – what Kahneman called “thinking slow”. I’m talking about the kind of thinking we do when considering the plausibility of predictions – what Kahneman called thinking fast, or intuitively. I’m talking about our gut reactions to seemingly alarmist predictions. 

Mnuchin’s mistake

Take for example Steve Mnuchin, US Secretary of Treasury. Mnuchin is a man whose foresight made him rich. In 2009, at the depth of the Global Crisis, he bought a failed mortgage lender and pocketed $1 billion in profit when he resold it in 2015. 

Mnuchin is so rich that, in the financial disclosures he had to fill out to become Treasury Secretary, he left off $100 million in wealth. By accident. When pressed at his senate hearing, he explained: “I think as you all can appreciate, filling out these government forms is quite complicated.” 

But when asked in March 2017 whether artificial intelligence (AI) would replace workers, he responded: “I think that is so far in the future. In terms of AI taking over American jobs, I think we’re like so far away from that, that, uh, not even on my radar screen. Far enough that it’s 50 or 100 more years.” 

There is a very good reason that even sophisticated thinkers like Mnuchin fail to process explosive growth, why well-informed, well-advised corporate leaders find themselves and their companies ‘disrupted’ by digitech that is – in a thinking-slow way – entirely predictable. Blame evolutionary psychology. 

Projecting the future with walking-distance brains

Our brains, the key bit of equipment we use to deal with predictions about the future, evolved for a very different job. Think of this this way.

Evolutionary psychologists believe that animal brains (including ours), evolved to track motion and react to it. Most motion in the natural world is more or less linear over short distances, and so our brains are hardwired to understand things that move in a walking-distance world. That is why we tend to think that tomorrow will be like yesterday, only a little bit more so, or a little bit less so. We think of the future in straight lines. This is why our brains struggle to think instinctively about digitech. Our brain evolved in a world where ‘really fast’ meant a spear in flight. 

‘Holy cow’ moments 

To visualise this, consider at schematic drawing. The straight-line way we think about the future is show in Figure 1 as the straight line, rising steadily. But that's not how digitech works. Digitech tends to follow the lazy-S curve. It starts at almost zero so the increments are – for decades – small, almost imperceptible. But then it hits an explosive increments phase. Eventually, the law of diminishing returns dominates and digitech turns back into straight-line growth. But consider what happens with exponential meets straight line. We get what I call the ‘holy cow' moment.

Figure 1 Holy cow!

Amara’s Law

During the early years of exponential growth, the promise of digital technology seems thrilling, but actual progress is imperceptible. People – using instincts embedded in their walking-distance brains – make predictions in this phase that prove to be alarmist. And they do it time after time. Roy Amara, a futurist who gave his name to Amara’s Law, has another way to think of ‘holy cow’ moments: “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run." Eventually, just around the holy cow moment, the overestimates turn into underestimates. 

Deep down, this mismatch between perfectly predictable technology progress and disruption arises from the fact that we tend to use our instincts, our gut feelings when evaluating claims about the future. And our instinctual mind evolved to deal with a walking-distance world, not the digital world. 

These two key points – the incredibly large increments and our systematic failure to think straight about the increments – are why digitech is not your father’s technological progress. It is really something quite different. And it is something that international economists should understand. 

What is driving these insanely large increments in digitech? As it turns out, there are four laws that explain them.

Four digital laws 

This series of blog posts will cover these all in more detail, but here's a summary.

The explosive rates of growth in processing and transmission are so remarkable that they have been given names – Moore’s Law for processing and Gilder’s Law for transmission. 

  • Moore's Law. The information processing power of microchips doubles every 18 months or so. This allows computers to get faster, cheaper, or smaller (or some combination of all three).
  • Gilder's Law. Transmission capacity – bandwidth – doubles every couple of years. Communication between people, computers, and machines gets faster, cheaper, and more reliable at an incredible pace 

(There are a few other, similar laws that are complements to these two; for example, Kryder’s Law on the cost of data storage.)

The other two major laws suggest that when you double computer power and the size of the internet, the economic and social impact more than doubles. These laws explain why the usefulness of being able to process and transmit grows even faster than the ability to process and transmit. These are Metcalfe’s Law, and Varian’s Law. 

  • Metcalfe’s Law. Being connected to a network gets more valuable with the square of the size of the network, while the cost of joining falls. As a result, online networks tend to get very big, very fast. 
  • Varian’s Law. Digital components are free, while digital products are highly valuable. Innovation explodes as people try to get rich by working through the nearly infinite combinations of digital components in search of valuable digital products.

Taken together, these laws are transforming the world of work at digitech speed. They are bringing automation and globalisation to the service sector at a pace that is faster than you think. 

My next post will look at Moore’s Law and the lawmaker, Gordon Moore.

Read the next blog in the series here.



Topics:  Productivity and Innovation

Tags:  artificial intelligence, technology, AI, processing power, holy cow moments

Professor of International Economics at The Graduate Institute, Geneva; Founder & Editor-in-Chief of; exPresident of CEPR

Vox eBooks

CEPR Policy Research