Taxing robots: the good news and risks about AI and government revenue

AI may deliver a large economic dividend and disrupt the labour market, and the Prime Minister rightly had these in focus in his recent address on AI in Australia’s interests. But AI will also likely change how income is earned and where income is taxed – potentially exacerbating income inequality and shifting the tax base offshore. Given these risks, there has been a surprising lack of discussion about the interaction between AI and the Australian tax system.

Digging into the details, there is both good news and cause for concern. Australia’s tax system looks robust to some features of AI, with changes in how income is allocated between capital and labour unlikely to hurt government revenue by itself.

Current tax settings could be more vulnerable to where income is being taxed. AI may turbocharge an existing trend to importing digital services. But it remains unclear if this would reduce the tax base as a whole. Regardless, a rebalancing from income taxes to GST would make the tax base more robust to AI.

The tax base is not exposed if AI shifts wages to profits within Australia

Many expect that AI will bring substantial increases in productivity. And there is a scenario where the productivity gains from AI are ‘business as usual’: that a variety of workers remain integral to economic activity and wages grow with productivity, even as many tasks currently done by people are automated.

However, if AI replaces tasks and jobs more quickly than new valuable tasks can be created for workers, the wage share of income could decline. This has led to calls for a ‘robot tax’ on AI by a wide variety of individuals, from Bernie Sanders to Bill Gates.

But in practice this reflects a shift in income from wages to capital owners. Concern about the effect of this on the tax base is more relevant to the US than Australia, as Australia taxes capital income more heavily – with capital income (except for capital gains and earnings within superannuation funds) taxed at full income tax rates.

Using the PBO Build Your Own Budget tool, we can illustrate what will happen with Australian Government revenue if automation reduces wage growth by 1ppt per year and instead increases business profits – our ‘wage decline’ scenario. A transition from domestic labour to capital income does not generate a large hole in the tax base. The result is close to a wash, in fact it is slightly positive as the corporate tax rate is higher than the average tax rate paid by workers.

This is only an approximation and makes a number of assumptions that might not hold in the event. It assumes that the decline in wages is equal across the income distribution and has no broader economic effects. It also focuses on Federal Government revenues, while state government tax revenues will take a modest additional hit from payroll taxes.

Importantly, the ‘wage-decline’ scenario holds productivity constant while only changing the distribution of the pie. If instead we were in a ‘productivity increase’ scenario where labour productivity grew 0.6 per cent more per year – a little stronger than the PC’s conservative estimates of the effect last year – then this would scale up tax revenue, such that total revenue could rise even as the labour share of tax revenue declines.

It could matter if AI replaces Australian labour with imported AI services

Another risk is that a transition towards AI may reduce the income tax base within Australia by shifting the source of income – and thereby taxing rights – across countries.

This opens a new front in the tax challenges from the digital revolution. These challenges have been front of mind for tax policy since at least 2018, when the OECD’s first report into the issues was released.

AI, and previous to that business service platforms, allow Australian businesses to import rather than relying on domestically provided services. ABS estimates suggest that digital services imports of this nature have quintupled relative to the size of the Australian economy since 2015.

If AI uptake is substantial, then part of the capital income earned from this process will head overseas – adding to digital services imports and potentially reducing the Australian tax base if AI substitutes for Australian workers.

But this is too simplistic. Foreign labour and capital are being used to provide the AI services. As the global income tax system is origin/source based, the income of AI providers will be taxed where the labour and capital are used to provide them. Even if the country where the company is headquartered allows profits to be shifted overseas, this is a subsidy for Australian consumption – it does not provide a justification for why we should tax these companies.

In addition, the use of these imported services frees up domestic labour and capital to shift toward other uses, expanding the Australian tax base in other ways.

However, the AI services have value partly because they are embedded in processes in Australia, producing products sold to Australians, utilising Australian infrastructure, and having been trained on data about Australians. Furthermore, AI providers can reach into a market replacing other services and products without a taxable physical presence. In this way the destination country – where the final consumption occurs – may have a legitimate claim on some portion of the income earned by cross-border firms.

More specifically, some economic rent may be associated with access to the Australian market, data, or user base, which is taxed primarily overseas. How much economic rent will be associated with AI services remains unclear, but there are scenarios where it is large.

For this reason, the OECD has recommended a reallocation of taxing rights towards destination for digitalised services, production, and processes. But it would require coordinated international action, and in the near term the OECD’s proposal doesn’t have the global support necessary to progress.

Rebalancing from income tax to the GST may make the tax system more resilient

A shrinking tax base due to AI and broader digital services does not mean that AI companies aren’t paying their fair share. It is instead a debate about who receives taxing rights over the income generated by these large entities. This is really a debate between governments about who should collect tax revenue, which means that any policy change that tries to grapple with this issue will also generate geopolitical tensions.

These tensions are the reason why the OECD proposals have failed to take off. And it is why digital services taxes (DSTs) have proliferated as a substitute. DSTs introduce a new destination-based tax specifically targeted at select digital services.

However, Australia already has a destination-based tax that can be utilised – the GST.

A higher rate of GST would ask foreign firms to pay tax based on the destination they sell in and hence would be more robust to digitalisation than the current reliance on income tax. Although GST would insure us against revenue risks from digitalisation, the politics would be another matter.

However, beware a world in which all countries move towards shifting taxing rights to where final consumption occurs rather than at origin. This would not benefit Australia. As we have written about previously, a global shift to destination-based cash-flow tax (essentially GST with a credit for labour costs) would significantly reduce tax collected in Australia, largely because it would shift how the rents from Australian resource exports are taxed.

There is a legitimate risk to Australia’s tax base from AI and policymakers need to get on the front foot, as they have in other areas of AI policy. These are important issues to work through, and shifting some of the tax burden from income to GST deserves to be in the policy toolkit.