Whither AI? UK productivity growth, a longer-run perspective

Explore how AI impacts UK productivity growth in a historical and future context, comparing past technological innovations to the present.

In his best-selling new book on AI, Co-Intelligence, Ethan Mollick, a specialist in innovation and entrepreneurship at Penn’s Wharton School, describes AI as a revolutionary technology: “we have never [before] built a generally applicable technology that can boost our intelligence”. We think this description is a bit hyperbolic, but only if you take a long-run perspective: the invention of writing 5,000 years gave us the ability to store and recall information almost without limit, the printing press made the reproduction of information cheap and – combined with mass literacy – available to almost anyone, and information and computer technology (ICT) enables all of to access near-instant search and recovery of vast amounts of information. LLMs (Large Language Models) are the next step – in Mollick’s nifty turn of phrase, they are “connection machines”, computer models trained on vast volumes of data that point us to relationships and ideas that we might not have thought of on our own.

We asked ChatGPT to write an opening paragraph for a blog that links AI to 5,000 years of human history (see the image). How different is it to ours?

Prompt: Give me an opening paragraph for a blog that links AI with 5,000 years of human history? Generative AI: Throughout the annals of human history, spanning over 5,000 years, humanity’s relentless pursuit of knowledge and innovation has been a defining characteristic. From the early days of agriculture and the birth of civilizations to the Industrial Revolution and the dawn of the digital age, each epoch has marked a significant leap in our collective capabilities. Today, we stand on the brink of another monumental shift with the rise of artificial intelligence (AI). As we delve into this new era, it’s fascinating to draw parallels between our past achievements and the transformative potential of AI, exploring how this cutting-edge technology continues the timeless human tradition of using tools to shape our destiny and redefine the future.

It’s … OK. It sounds “human” and is stylishly written. But our introduction queues up one of our main points: the previous innovations we mention – writing, printing, mass literacy, ICT – took decades to millennia to see their full effects play out. There’s no hint of this forward-looking thinking in the ChatGPT answer. But one of our points is that this is very much early days.

It is a well-established tradition within the dismal science to miss the importance of technological revolutions. Thomas Malthus’ model of economic growth was a conceptual breakthrough in 1798 but it explained the historical record only up to the point of the Industrial Revolution. A more recent example is Paul Krugman’s prediction that by 2005 the internet would have less of an impact than the fax machine. As the saying goes, it’s difficult to make predictions, especially about the future.

UK productivity: a long-run perspective

One such attempt at prediction is by MIT economist Daron Acemoglu: he estimates that AI can lead to a 0.53 to 0.66 percent increase in total factor productivity (TFP, a commonly-used overall productivity measure in economics) over the next 10 years; in other words, an increase in the annual rate of productivity growth of perhaps 0.06 percentage points per year. This is pretty small, though not quite as small as it first sounds given the UK context.

A recent account of British economic history sees low TFP growth in the early Industrial Revolution in England & Wales, followed by an increase in the mid-19th century (Crafts 2021). In the diagram below we show UK TFP growth from 1760-2022 (based on McLaughlin et al. 2014). Underlying TFP growth peaked in the 1960s, in line with a view that sees the frequency of major technological innovations peaking in mid-twentieth century (Gordon 2012). UK productivity has been in decline since the 2007-8 financial crisis and productivity growth over the past decade has been almost nil. If the actual contribution of the AI revolution is anywhere near Acemoglu’s pessimistic forecast, we will have to look elsewhere for solutions to the UK productivity problem.

UK TFP growth 1760-2022

Limits to AI growth?

In Acemoglu and Johnson’s book Power and Progress, a key issue is whether AI will complement or substitute for labour. For example, it could help a low-wage tradesperson by giving them access to bespoke knowledge that helps improve job performance. But if AI replaces a task that humans already did quite well, such as self-service tills, then by substituting for productive labour AI will not lead to any productivity improvements. The direction of technological change will clearly have broader social implications such as inequality (see this recent IMF report ).

Perhaps the most profound challenge relates to what is known as “model collapse”. The original LLMs were trained on “clean” data uncontaminated by AI chatbots, but LLMs post-2021 will be trained on web-scraped text that has in it output generated by AI models. The quality of AI output degrades as they start to mimic themselves as well as humans. This scenario is most applicable to general LLMs but less so to specialised LLMs such as AIs that search for patterns connecting genes and diseases, because the training data is under the control of the investigators.

What is the likely impact of AI on productivity growth? We are cautiously optimistic (we’ve not gone down the existential rabbit hole that preoccupies one of us) and sit somewhere between the extreme pessimism of Acemoglu and the uber optimism of Mollick. Why don’t we give ChatGPT the last word?

Prompt: Can you write a 100 word summary to question: will AI increase productivity growth? Generative AI: AI has the potential to significantly boost productivity growth by automating routine tasks, enhancing decision-making, and driven innovation. In various sectors, AI can handle data entry, process automation, and customer service efficiently, freeing human workers for more complex tasks. It also provides valuable insights for better strategic planning and operational efficiency. However, the integration of AI poses challenges such as potential job displacement, ethical concerns, and high implementation costs. Despite these challenges, with proper management and upskilling of the workforce, AI can unlock substantial productivity gains and create new economic opportunities.

This is rather better. Maybe we should have asked ChatGPT this question before we wrote the blog?

Professor Mark Schaffer FRSE, Professor of Economics, Heriot-Watt University 

Professor Eoin McLaughlin, Professor of Economics and Head of Research (Department of Accountancy, Economics and Finance), Heriot-Watt University 

The RSE’s blog series offers personal views on a variety of issues. These views are not those of the RSE and are intended to offer different perspectives on a range of current issues.

You might also like