Generative AI: gimmick or game-changer?

The imaginations of futurists and investors alike have been captured by recent developments in AI. Parallels have been drawn to other catalysts of industrial automation, such as railways and electricity. But AI has also drawn warnings that its potential has been over-exaggerated and we risk repeating the mistakes of the dotcom bubble in the early 2000s. Which chapter of history is most likely to repeat?

It was inconceivable just a few years ago that a computer could be trained to interpret almost everything ever published and use it to create new material. But that is what we have with generative AI – so called because it can generate responses. It can give coherent and contextually accurate answers to almost any query and in any language. This includes most spoken languages – as well as an ever increasing range of visual ones, including charts and graphs, pictures and videos. Unlike automation tools from previous industrial revolutions, AI is able to perform intellectual tasks rather than purely manual ones. This opens up use cases across service sectors such as education, law, healthcare and finance.

AI may well be adopted more quickly than previous industrial breakthroughs. It does not depend on the construction of costly infrastructure, as was the case with railways, electricity transmission and even broadband. We already have access to AI via our laptops and iPhones and it can be switched on in an instant.

We are seeing clear evidence of this very rapid uptake, suggesting tangible demand behind all the excitement. Over 100 million people registered to use ChatGPT in the two months after its launch, a record pace of adoption. Technology firms are using the “large language models” underlying ChatGPT to build other tools that can be integrated into existing software or provide new services. This is driving huge demand for “the picks and shovels” companies that power AI. Chip designer Nvidia has raised its sales forecasts by 80% for 2023 and 100% for 2024, for instance.

Productivity: the magic elixir of economic growth

Productivity is a measure of the output per hour worked in a given economy and is one of the key determinants of long-term prosperity. More output for less time and effort ultimately results in higher incomes and better economic outcomes. Unfortunately, in many Western economies, productivity has been making little contribution to growth over the past 20 years.

Productivity can be a volatile measure, often spiking after recessions as production ramps up before firms have rebuilt their workforces. However, the last time that we saw a sustained, multi-year rise in productivity was in the late 1990s. This coincided with the launch of Windows 95 – the basis of the PC operating system that most of us still use today – leading many to suggest that the PC era unleashed a step change in corporate productivity. There were probably other contributing factors, but the period was a very good one for corporate margins and earnings. Over the second half of the 1990s, S&P500 earnings doubled and the index delivered a return of 181% (from 1995 to 2001).

Could we be on the cusp of a similar surge in productivity as AI causes a boom in workplace efficiency? We may find out relatively soon. Microsoft has announced plans to integrate large language models into its suite of office products used by over a billion people. Writing, working with data and creating presentations could soon become much less time consuming.

If AI can lead to improvements in productivity, it would come at a particularly opportune time. Since the pandemic, the supply of labour – another key determinant of long-term economic growth – has been under significant pressure as a result of demographics and other factors. Over time, this reduces economic output and raises costs for companies. Higher productivity could offset the negative impact and, as in the 1990s, we could see economic growth and corporate profits ahead of expectations.

Productivity surged in the late 1990s

Average level of US non-farm output per hour (year-on-year % change)

^{Source: Bloomberg Economics}

Past performance is not a guide to future performance and may not be repeated.

Social costs

Much has been written about the threat that AI poses to employment in certain industries. It is possible that there will be some job losses as companies find new and more efficient ways of operating. As part of our broader conversations with companies around AI, we encourage them to handle any such transitions as responsibly as possible.

However, this is not the first time that the global economy has been through a wave of automation. Historically, we have seen some forms of work go away – but other kinds of work and new jobs have been created. Today, UK firms employ thousands of people and spend billions of pounds on digital marketing. Twenty years ago, those jobs were probably unimaginable for most of us. Who would have expected to see social media influencer as a career possibility?

Are markets unrealistic?

Investors have been hugely enthused by AI’s potential. The seven largest US technology companies have added roughly $3 trillion to their market capitalisation this year, as investors anticipate their revenues will rise in line with spending on AI-enabled services. Incredibly, this increase in value is greater than the total value of all the constituents of the FTSE100. One of these US tech giants is chipmaker Nvidia, which currently dominates the market in the powerful microchips required to run large language models. It has seen its share price more than triple this year. Its vertiginous rise has led many to suggest it’s overvalued and overhyped. But the numbers probably suggest otherwise. Nvidia currently trades at a multiple of around 40x its expected earnings for this year. That may be too rich for the taste of some investors – but it’s nothing compared to the peak multiple of the largest infrastructure player of the dotcom boom: in 1999, Cisco traded on a staggering multiple of 200x. Meanwhile, Nvidia is expected to make $31 billion of profit this year. So by itself, it will make almost half the entire profit pool of the US technology sector in 1999. History may rhyme, but it never exactly repeats.