Four primary reasons tend to be put forward for why this might be the case.
One is that the data simply doesn’t capture the gains being made because of the plethora of free stuff that is made available over the internet, and their failure to measure intangible benefits not before encountered.
A case in point would be access to virtually the entire body of human music ever recorded where once you would have had to go to an expensively priced live concert to hear just a tiny fraction of it.
Another, related explanation is that traditional measures of real output have overestimated inflation and underestimated productivity because they do not take fully into account quality improvements of IT goods and goods in general.
A slightly more contentious explanation is that the gains have not been widely spread, but have instead been snaffled by the likes of Jeff Bezos and the investors and employees that cluster around these “robber barons” of the modern age.
The productivity gain has in other words been used to enrich the relatively few. This might particularly be the case in economies such as the US and the UK, where jobs are abundant and unemployment is basically a matter of choice. Let unemployment grow, and productivity, if not economic growth, will automatically increase.
And then there is the possibility that as economies become ever more dependent on labour-intensive, service-based industries for their wellbeing, it is just that much more difficult to show productivity gain, which tends to be concentrated in manufacturing sectors where greater output for less labour is easier to achieve.
Growth in the public sector, which appears virtually immune to productivity gain, may also be part of the explanation. And finally, it may simply be because it takes a long time, possibly decades, for these technologies to fully show up in the statistics.
What makes AI so potentially revolutionary is that its main impact is expected to be on employment in precisely those jobs in rich, service industries which hitherto have remained so stubbornly immune to productivity gain.
Radiography is but one narrow field where the impact is already being felt. Unlike a mere human, AI can analyse literally millions of scans a second and, moreover, it can spot things the human eye might miss or misinterpret. As the available data builds, it should also be able to revolutionise early diagnosis and treatment.
But just because a technology becomes available, doesn’t mean it will be used.
Taking radiography by way of example, doctors first have to be convinced and apply their skills elsewhere. Resistance from vested interest among medical professionals is one thing. Introducing AI more widely into the public sector, where it is perhaps most urgently required, will encounter all kinds of Luddite-style opposition.