History suggests that waves of technological innovation take time to show up in productivity, with users not producers reaping the spoils. Productivity pessimism in Australia should not last forever, says Westpac's Luci Ellis

We are frequently asked how Australia might boost its productivity growth and so its living standards. We are also frequently asked our view of the implications of genAI and large language models. These questions are mostly about the medium term but the underlying beliefs also have implications for macroeconomic policymaking in the short term.

A little historical background is in order here. Australia, and most of the rest of the Western world, experienced a boom in productivity growth in the late 1990s associated with the adoption of computers and the internet. That episode held two lessons. First, it took time for a new wave of technology to be fully adopted and embedded in business models and processes. That is why it initially seemed that the computer revolution had not boosted productivity – recall leading US economist Bob Solow’s 1987 quip that “You can see the computer age everywhere but in the productivity statistics”. Second, most of the gains went to the users, not the producers of the new technologies. Indeed, productivity growth was higher in Australia than in the US over the late 1990s, as the then RBA Governor Ian Macfarlane pointed out at the time.

As that wave of technological innovation crested and matured, productivity growth slowed globally. The gains from adoption had been reaped. Moreover, the new technologies introduced in the first couple of decades of this century – specifically social media – were seemingly more likely to distract us than make us more productive. They are also more prone to the network effects that direct the gains to the platform operator rather than the users, limiting the boost to productivity.

Even when the earlier generation of innovations that culminated in what we now know as AI were introduced – machine learning models and data science techniques – the productivity gains were hard to see. Part of the issue is that that generation of technologies was essentially an exercise in combining people with PhDs in physics or maths with computers to generate a machine-learning model to replace, say, an insurance adjuster with a high-school or undergraduate education. A technology that requires rarer skills than the ones it seeks to replace is rarely successful in gaining broad adoption.

It was therefore perhaps not surprising that measured productivity growth slowed around the world starting from the early 2000s. The economics profession was worried about the possibility of ‘secular stagnation’, and some believed that “diminishing returns in the digital revolution” were causing the slowdown in productivity growth, at least in the US. (There were other, demand-side, causes proposed as well. But the lack of technological innovation did seem to be a large part of the story.)

Roll forward to the last couple of years and we are now starting to see the next generation of machine learning applications – LLMs and other approaches based on transformer architectures. These are more accessible to end-users than their predecessors and hold out more of an opportunity to remodel business models and processes to take advantage of this. For this reason, most observers, including the IMF, expect at least some productivity boost from this new technology wave. Like the previous technology wave from PCs and the internet, this boost could take a while to come through. It might not be as slow as that previous wave, though, given that internet distribution itself speeds adoption.

The RBA’s revised assumptions about trend growth in labour productivity, and so potential output growth, released in their August Statement on Monetary Policy and elaborated on in a recent speech by its chief economist, need to be seen in that context. Effectively, what the RBA has done is use a 20-year average of productivity growth as the trend to which actual labour productivity growth is assumed to converge over the next two years. A 20-year horizon for this average cuts out all of the 1990s productivity boom from the calculation: it is a pure ‘secular stagnation’ era average.

And that might well be the right assumption for the next couple of years. If it takes a few years for AI to boost productivity across the economy, then this will not be clearly evident until the period beyond the RBA’s current forecast horizon. But for it to still be the trend rate of labour productivity growth much beyond that, one must believe one of two things. Either one must believe that AI will do nothing to boost overall productivity growth – in which case, sell your Nvidia stock now! (This is of course not investment advice, simply the logical implication of that belief.) Or, one must believe that AI will boost productivity growth elsewhere, but for some reason not in Australia. This seems like a stretch, or at least an argument that needs to be justified explicitly.

The RBA has assured the public that it does not assume that the slower productivity growth assumption applying to the next two years will remain the case over subsequent years, and that it will update its view as the data evolve. Observers are entitled to ask how the RBA proposes to do this, noting the backward-looking nature of many of its models for estimating these ‘star’ variables, and whether it will be nimble enough in updating its view. We recall the decline in estimated rates of feasible unemployment (the NAIRU) over the 2010s in a range of advanced economies and reflect that these are hard calls to make in real time. 

An implication of the RBA’s view on growth in productivity, and so potential output, is that for the time being at least, signs of stronger GDP growth will by default be interpreted as demand outstripping supply, and so a reason to keep policy a bit tight. Such an interpretation will be vulnerable to any pick-up in productivity stemming from AI or other technological or structural shifts. It will also be vulnerable to the RBA’s apparent assumption that the five-decade upward trend in labour force participation – and so labour supply – is at an end. As Westpac Economics colleague Ryan Wells and I noted last month, we think the trend still has some way to run in Australia.

Given the difficulties of assessing things like potential output growth, a more robust way to judge whether output is running faster than capacity is to watch inflation. Likewise, a pick-up in wages growth is likely to be more of a sign of a tight labour market than whether the rate of wages growth exceeds some rule of thumb based on an assumed rate of trend productivity growth. Without these price-based signals, it is hard for policymakers to be confident that a shift in a quantity-based measure like output is a sign of a shift around trend, or a shift in the trend.

Indeed, while we wouldn’t want to make too much of one month’s figures, this week’s labour market data was entirely consistent with our medium-term view that there is a bit more trend growth in labour supply available than some observers assume. As has so often been the case over the past couple of years, it will again all come down to the quarterly underlying inflation print later this month. And while that print will be a high one, this might not continue in following quarters. The RBA’s guiding assumptions make it more prone to productivity pessimism and structural hawkishness that could be repeatedly corrected by lower-than-expected inflation outcomes. It’s a heck of a way to manage the economy.