No, AI probably won't take your (whole) job, but the future workforce will look different. Fortunately the labour force is already shifting in mostly the right direction. Policy and social changes are nonetheless needed, says Westpac's Luci Ellis

By Luci Ellis*

AI – generative or otherwise – is simply another technology for automating previously manual tasks. If you can do something more productively, you need fewer people to produce current output.

Understandably, people worry that their jobs will therefore disappear. Some commentators argue that AI will induce mass unemployment, requiring governments to introduce a Universal Basic Income. These messages have even come from voices in the tech industry, which hardly seems like a good way to garner public support for the industry’s direction.

The good news is that with every wave of a new, transformative technology, we find new things to do. While there might be a bumpy transition, the end result is not lower employment. Consider that 300 years ago, around two-thirds of English men worked in agriculture; higher figures applied 200 years ago for agrarian colonial societies, including what became Australia. Now, fewer than 2% do, but we have more and more varied food to enjoy. The other 98% of the workforce do other things. The same process occurred when the Industrial Revolution mechanised spinning. The share of the non-agricultural workforce in England and Wales that was in the textiles industry fell even as production of clothing and other textiles boomed. And rising manufacturing productivity in the 1900s enabled the expansion of many desirable services.

Contrary to the claims of the more tech-oriented commentators, AI does not reduce the value of human labour. Quite the opposite. The supply of things ‘robots’ (AI) can do will expand massively, lowering their relative price. By contrast, the supply of workers available to do things that only humans can do will not increase as much, even allowing for some displacement through automation. Thus, the relative price of things only humans can do will in fact rise. The human touch, the in-person service, will become the prestige item, and a greater share of people will work in those fields.

And there will be many things that only humans can do, even as the AI models get better. In particular, knowing what the right question to ask is (which, along with “and how to ask it”, is all that people mean by “prompt engineering”) is a uniquely human activity.

The automation effect will not affect every job equally, of course. A range of recent reports have sought to quantify which tasks are most susceptible to full automation, which will be augmented, and thus which jobs will need to change the most. There are limitations to this work. For a start, jobs are not just a collection of disjoint tasks. The original ILO methodology also uses ChatGPT to work out which tasks are impacted and might not fully capture agentic AI’s current and future capabilities.

Still, the pattern is clear – clerical and administrative work is most susceptible to automation, while for most other occupations, AI will augment rather than fully automate. Technical and building trades, along with ‘caring’ occupations where the human touch is so important, are the least impacted.

Another thread in the concerns about the impact of AI on the labour market is that it will be uneven across different experience levels in the same occupation. Put simply, the tasks most prone to automation are the routine ones usually performed by the less experienced members of an occupation. We already hear this concern from contacts in industries such as management consulting, tertiary education and elsewhere. If all the routine ‘donkey work’ in these white-collar occupations is eliminated, these contacts worry, what will the new graduate staff do? How will they learn?

We suspect these fears are overstated. The amount of ‘donkey work’ needed to teach the new graduates how to be a more senior member of the profession was surely lower than the amount that needed to be done. My own observation of the economics profession accords with this. The workload for a young public-sector economist three decades ago involved a lot of spreadsheet grunt-work that needed to be done but did not teach the young economist much after the first couple of times they did it. Nowadays, the data have become more abundant, the models more sophisticated and the expectations elevated. And despite not cutting their teeth on as much ‘donkey work’ as their predecessors, the young economists and other analysts of today certainly meet those higher expectations. A steeper learning curve was possible. (And perhaps, with less time spent learning and doing the ‘donkey work’, more time is available to develop the all-important judgement, soft skills and other things only humans can do.)

There is a less optimistic wrinkle to the issue of impact by experience level, though. Even if the skill expectations for new entrants to white-collar roles increase, these entry-level roles are still the most impacted. Academic research shows that new entrants are also the most impacted in economic downturns. It is much easier to just not hire an unknown quantity with no experience than to sack longstanding, experienced staff. These early-career experiences also seem to have a ‘scarring’ effect on people, long into their careers.

It is therefore no wonder that people are worried about the prospects for new entrants to the workforce in the AI era. In the US, signs of a weakening in this segment could be a signal of a weakening economy, but researchers from Stanford University, the St Louis Fed and elsewhere suspect that AI adoption is already disrupting the jobs market. (It could also be a bit of both, or an unwind of some pandemic-era over-hiring.)

There does not seem to be evidence of this effect in Australia just yet. As Westpac Economics colleague Ryan Wells noted last week, youth unemployment has been volatile and broadly picking up. But we are not yet seeing evidence of a generalised deterioration in hiring for entry-level roles. JSA notes that fill rates for vacant jobs have been improving, including in the June quarter (the latest report). If the bottom had fallen out of the entry-level jobs market, we would expect to see the average number of applicants per vacancy rise materially in key industries, but the number of qualified or suitable applicants remain steady or fall. But the JSA data show the opposite – rising average numbers of qualified/suitable applicants with total applicants steady or a little lower than a year or two ago.

The research on differential impact by occupation does suggest that some policy and societal response is needed, though. As noted above, the roles most impacted by AI are expected to be white-collar occupations with a lot of routine tasks amenable to automation. Those least impacted are the ‘caring’ occupations where human contact is needed, and technical and trades occupations, where physical processes predominate. We have previously highlighted the industry-level implications of this.

The glass-half-full view of these likely shifts is that trends in labour supply in most western countries, including Australia but not the US, are shifting in the required direction. As Ryan and I noted in a report a couple of months ago, population ageing is, outside the US, leading to higher overall labour force participation, with the workforce becoming older and more female. A higher share of older, more experienced workers, who know what the right question is, are exactly the kind of workers who will be best placed to thrive in the AI era, perhaps with a little training and practice.

The glass-half-empty view, though, is that Australia already struggles to train enough technical and trades workers. A strong gender skew also limits supply: the construction industry has the lowest proportion of female workers of any industry. Even mining has a much better gender balance. Meanwhile, ‘caring’ occupations have expanded lately for other reasons, but public funding has its limits. Policy action is clearly needed around both these occupation groups. But society also needs to re-evaluate the social cachet accorded to some jobs over others. 

Luci Ellis the chief economist at Westpac Group.