Artificial Intelligence as a general-purpose technology: an historical perspective

Artificial Intelligence as a general-purpose technology: an historical perspective Artificial Intelligence as a General-Purpose Technology: an Historical Perspective

Abstract

This paper looks at the impact on productivity of general-purpose technologies such as steam, electricity and ICT. It finds they had big effects but only with a lag which was substantial in the first two cases. The experience of the First Industrial Revolution is explored and it is found that this is not a template for a general-purpose technology having a major adverse effect on workers' living standards. The essence of that industrial revolution was not rapid productivity growth in the short run but the 'invention of a new method of invention' which increased technological progress in the long run. Since artificial intelligence is potentially a general-purpose technology that raises the productivity of research and development, it may be the basis for a Fourth Industrial Revolution.

One. Introduction

A General Purpose Technology (GPT) has been defined as 'a single generic technology, recognizable as such over its whole lifetime, that initially has much scope for improvement and eventually comes to be widely used, to have many uses, and to have many spillover effects'. As such, it can be expected to be pervasive and to have a significant impact on aggregate productivity growth, possibly for a long period of time and probably after an initial lag. The classic examples are typically considered to be steam, electricity, and information and communications technologies. Growth accounting is one way to estimate the productivity impact of these GPTs (section two).

Quite possibly, Artificial Intelligence (AI) will eventually also be seen as a classic GPT. Indeed, 'techno- optimists' would argue that today's productivity paradox of excitement about AI and robotics combined with slow productivity growth is explained by the delay before the potential of this new GPT is realised. Growth accounting estimates for earlier GPTs show that their impact on productivity takes time to develop.

At the same time, new technologies are feared by some for the pressure that may be put on the labour market with adverse implications for workers' living standards. British industrialization in the early nineteenth century is often seen as a prime example of such an outcome. It is well-known that real wages increased very slowly during this period of acceleration in technological progress which is often seen as an era when the share of wages in national income was squeezed in an early example of workers being replaced by machines. Informed by the account of 'Engels' Pause' in Allen, this interpretation has attracted renewed interest from economists in the context of worries about the impact of AI and robotics on the labour market. This experience is reviewed in section three to see whether it really is a worrying precedent.

Cheerleaders for AI expect it to be a GPT driving a 'Fourth Industrial Revolution'. Earlier Industrial revolutions are also often thought to be associated with classic GPTs; steam with the First Industrial Revolution, electricity with the Second Industrial Revolution, and ICT with the Third Industrial Revolution. For an economic historian, however, this does not do justice to the concept of an 'industrial revolution' which entails a significant change in the methods of generating advances in technology rather than faster technological progress and an acceleration in the rate of productivity growth per se.

As Alfred North Whitehead famously remarked, 'the greatest invention of the nineteenth century was the invention of the method of invention'. The First, Second and Third Industrial Revolutions each saw improved methods of invention and it is helpful to consider the prospects for the impact of AI as a GPT that delivers a Fourth Industrial Revolution in the light of this historical experience (section four). To achieve this, it needs to be not just a GPT but also the invention of a method of invention, as some writers think it may be.

In general, IMIs and GPTs are distinct, as Figure one describes. IMIs raise productivity in the production of ideas while GPTs raise productivity in the production of goods and services. However, a subset of GPTs also provide an IMI and have an important role in increasing the productivity of innovative effort. ICT was a case in point. The contribution to productivity growth as an IMI will typically not be attributed directly to the GPT by growth accounting but will show up as an increase in total factor productivity growth in the aggregate economy.