AI, macro and ice cream parlours

A long time ago, I worked as a salesperson in a then well-known ice cream parlour on the Leidseplein in Amsterdam. I had a colleague there who was studying Artificial Intelligence. I thought it was a fascinating subject, although I couldn't put my finger on what it meant exactly. Years later, I still don’t really know what it means, although I have a slightly better idea. And it still fascinates me, not least because the use of Artificial Intelligence (AI) also has macroeconomic implications. That's what this blog is about.

Gen AI

Let us first make it clear exactly what we are talking about. AI is a branch of computer science involved in "developing systems that perform tasks that normally require human intelligence", according to ChatGPT. A typical example of an AI application is a spam filter for your e-mail. Generative AI (hereinafter, “Gen AI”) is a sub-set of AI. This is a form of AI capable of creating new content such as text, image, sound, but also computer code, for example. Gen AI has come under intense scrutiny, including from investors and economists, since the launch of ChatGPT in November 2022.

The first question from an economic perspective is whether Gen AI will lead to increased productivity growth and hence economic growth. The idea behind this is that Gen AI can automate some of the tasks performed by people, enabling more products and services to be created per working person. It does look as though Gen AI will lead to productivity growth, but estimates vary widely: from 1.5% additional productivity growth per year to just 0.07%.

Cost reduction

That last estimate comes from the economist Daron Acemoglu. Acemoglu is sceptical about Gen AI resulting in a spurt in productivity growth. While he does think Gen AI could automate some (20% according to Acemoglu's estimate based on previous studies) of the tasks currently performed by humans over the next decade, he estimates that only 23% of these can be automated cost-effectively. For the rest, the cost of automation will be higher than the savings it will bring. The conclusion is that only 4.6% (20%*23%) of the tasks performed by people can be automated cost-effectively over the next ten years. As a result, Gen AI's productivity gains are limited, in Acemoglu’s view.

Labor market

Another question that arises when thinking about automation is whether it is going to lead to a wave of layoffs and unemployment. This could be the case. But historically, the labour market has always adapted to new situations over time. Disruptive technologies tend to create new jobs, as well as eliminate obsolete jobs. The internet and the computer are examples of such disruptive technologies. For example, before the arrival of the Internet there were no web designers, and before the arrival of computers there were no software developers. A study by the economist David Autor shows that 60% of the jobs people currently have did not exist 80 years ago. This does not alter the fact that it takes time for a labour market to adapt. As the speed at which people find a new job or re-skill can also vary enormously, unemployment could indeed temporarily increase.

Inequality

In addition to effects on the labour market, the arrival of Gen AI could also increase inequality in various areas. The IMF recently developed an index that measures countries’ readiness to reap the benefits of Gen AI. Richer countries score significantly higher on this AI preparedness index than poorer countries. That's partly because richer countries have the digital infrastructure to take advantage of Gen AI, such as good internet. And partly because the people are more highly educated, making it easier for them to change jobs when their own job disappears. Thus, richer countries will benefit more from Gen AI than poorer countries, which could widen the gap between the two. Within countries, too, the gap between theoretically and practically educated people and between older and younger people for example could widen, because one group is more skilled at using Gen AI than the other. This is undesirable, both in a moral sense and from an economic perspective. This is partly because rising inequality can contribute to increasing populism, which can lead to higher government debt, more policy uncertainty and higher interest rates.

A lot of energy

Finally, an important aspect of Gen AI is that it consumes a lot of energy. Asking a question to ChatGPT costs ten times as much energy as searching for something on Google (0.3 watt hours compared to 2.9 watt hours). In addition, it costs a lot of money and energy to ‘train’ Gen AI models such as ChatGPT 4.0 on data. For example, The Economist recently calculated that ‘training’ the current Gen AI models costs around USD 100 million. The emergence of Gen AI does not have a positive effect on climate, at least in the sense that it will not contribute to less energy consumption. If that high energy consumption and the high cost of training is passed on to customers, Gen AI could, in the short to medium term, lead to higher inflation. This also has implications for interest rates, as these reflect expected economic growth and inflation. In combination with the effect on economic growth (higher), it looks as though Gen AI is more likely to push interest rates higher rather than lower in the short to medium term.

What are the implications for investors?

For starters, the US economy could continue to grow faster than the European economy for the time being, partly because the US is higher on the IMF’s AI preparedness index. That could make a case for US equities, although it has to be said that these already have high valuations and are ‘expensive’ in that respect. In addition, the concentration risk of the US equity index is high, as almost a third of the value of the S&P500 index consists of only seven companies. However, Gen AI increases the chance of a ‘growthflation’ scenario (high growth and high inflation). If this happens, it would benefit marketable securities such as equities and higher-risk corporate bonds. Furthermore, demand for training and maintenance and the use of Gen AI could increase demand for data centres, potentially creating opportunities in certain real estate segments. But Gen AI could also increase inequality in the long run, fuelling populism. And its effects are more likely to be inflationary than deflationary. Both are bad news for government bonds. Finally, the high energy consumption of Gen AI is a concern. We have to try to find a solution to this problem together. Maybe we can ask my former colleague at the ice cream parlour for their advice.