I hope this AI thing works out: Part 2

Last week, I wrote about the high hopes priced into AI stocks in the US and why I hope these companies can continue to deliver on these hopes. But my real concern about AI is that it has to work out in lifting our national productivity because if it doesn’t, we are getting into increasingly hot water.

You see, we have a growth problem everywhere in the developed world. The demands on our education and healthcare systems are rising, we must spend more on defence to counter the aggression of Putin’s Russia and other countries, and we are already suffering from tax burdens that in many countries are at a multi-decade high.

For three decades, we have managed these issues by cutting defence spending to finance our welfare system without having to raise taxes. Alas, that is no longer an option. So, unless we find a magic money tree (and believe me, politicians everywhere have been looking hard), we either have to increase taxes or increase growth, so that the government receives higher tax revenue without hiking tax rates and we can fund rising government obligations.

The problem with growth is, however, that in the long run, it is pretty straightforward to determine. It essentially depends on three factors:

• The supply of labour. If we have more workers, we can produce more stuff and grow our economy faster. Unfortunately, most high- and middle-income countries suffer from declining population growth and – even worse – declining working age population. Which is why we have increasing demands on our welfare system to begin with. And demographic trends aren’t going to change anytime soon. The only fixes I can think of is to work longer (as is the case in Japan) or increase immigration (and we all know where the debate stands on that one).

• The supply of capital. If we invest more, we can replace labour with capital and machines and get out of our quagmire with declining working age population. This is essentially where AI and other forms of automation and digitalisation can help and where I really hope we can make progress in the next five to ten years. However, our track record on investments is pretty poor in recent decades (not years, decades). I came across the below chart from Bloomberg that shows not only that the UK has invested less than its G7 peers for 24 out of the last 30 years, but also that most countries have reduced investments as a share of GDP over time or at best kept it stable. The only G7 countries with a marked increase in investment-to-GDP are Canada and France.

Investment-to-GDP in G7 countries

Source: IPPR, OECD, Bloomberg

• Productivity growth. In this case total factor productivity (TFP) or the efficiency, with which capital and labour are employed and the technological progress that increases this efficiency.

When it comes to productivity growth, generative AI promises large efficiency gains, but we also have to be aware that there are dissenting voices. I recently wrote about Daron Acemoglu’s estimate that generative AI might increase productivity by 0.66% over ten years, or a mere 0.06% per year. I find this rather low, but then again, I have to remember that we have been here before, in a way.

I started my career at the height of the tech bubble of the late 1990s and back then, we were promised the internet would revolutionise the world and trigger massive productivity gains. But, as one economist apparently said: “You can see the influence of the internet everywhere but in the productivity statistics”.

Just take a look at the five-year rolling average TFP growth for the UK and the US below. The internet, just like the PC have given us a boost in productivity growth that lasted for about five years before it evaporated again. And that boost was not very strong. Rather in the order of 0.5% a year. That isn’t enough to solve our fiscal problems in the long run. It is barely enough for a politician to kick the can down the road to the next election.

Productivity growth and technological innovations

Source: Penn World Tables 10.0

Looking back in history, there have been instances when innovation led to a clear step change in productivity and economic growth. The most important of them was the invention of the steam engine that drove the industrial revolution. Other innovations that had a long-lasting impact on productivity were the internal combustion engine which enabled a more efficient use of energy than the steam engine and the electric light bulb which literally lit up the night.

If you look at these inventions, an idea that comes to mind is that lasting improvements in productivity are driven by our ability to harness different forms of energy and deploy this energy cheaply and universally. Recently, the always recommendable Matthew Syed of the Times and Sunday Times wrote about this in his Sunday column.

His argument is that lasting productivity gains are only possible if we find technologies with a high Energy Return on Investment (EROI). This leads him to argue that we should focus our investments mostly on energy infrastructure like the electricity grid and nuclear power.

In my view, Matthew is coming to the right conclusion for the wrong reason. I think we need to invest heavily in our electricity grid to enable us to decarbonise our economy and become energy independent from countries like Russia, Saudi Arabia, Qatar, and others. The shift towards nuclear power and renewables has three key advantages for our economies: (i) it reduces the negative impact from climate change, (ii) it increases our geopolitical independence, and (iii) it creates a ton of jobs and boost GDP growth through an increased capital intensity in our economy.

But in my view, it won’t increase our productivity growth beyond what is possible with new technologies like AI. Indeed, the gains in GDP growth from investment in our energy infrastructure are likely to be much lower than in the past for a simple reason: We no longer live in an industrial world where the manufacture of goods drives economic growth and prosperity. It does so to some extent, particularly in middle-income countries. But in high-income countries we live in a post-industrial society where services are the key driver of economic growth and prosperity.

Just look at the energy intensity of the US, the UK, and the world overall. It has declined by two thirds in the last fifty years. Or to put it differently, even if we find technologies with high EROI, their impact on GDP is likely to be only one third the impact they had fifty years ago. And this is why we need to find technologies that increase efficiency and productivity in a services economy. Generative AI promises to be this technology. And yes, we need a lot of electricity to run this, which is why we need to invest in energy infrastructure, but as I said, the productivity gains come not from energy inputs but from technologies that allow us to deliver services to everyone cheaply and efficiently. I think generative AI can be that technology. I just hope that AI doesn’t end up like the internet…

Energy intensity of the economy

Source: OECD