Artificial Intelligence replacing employment means dangerous drops in government revenues. Taxing AI isn’t the solution.
Whether as a labour efficiency tool or complete displacement of workers, Artificial Intelligence (AI) is set to undermine the established economic model. At stake is lost income, or even jobs, putting huge strains on public finances and risking public unrest. It’s not like this is a new story: ancient Rome’s slave labour replaced employment for the working classes (‘free to be poor’). The welfare bill – bread and games – to keep the proletarii subdued, and resulting tax rises, helped bring down the Empire.
Ironically, the earliest AI adopters are tax authorities themselves looking for tax evaders.
But Governments are now anxious about the prospect of shrinking income tax receipts from as AI suplants workers. In most modern economies, labour taxes are now the largest single source of receipts – in the EU is accounts for 53% of tax receipts. So governments are now exploring how public spending can continue even as the pool of taxable income from human workers shrinks. This includes taxing AI. But how?
A number of AI taxation frameworks have already been proposed. But the practicalities of their application, negative effects on productivity and ease of avoidance, may undermine any of these. Any new tax model must meet the accepted tests of good taxation: neutrality; simplicity; certainty; flexibility and efficiency.
Watch VATCalc’s new AI VAT Advisor deliver tax advice (legislative references and tax cases) on complex VAT questions. Saving hours on tax research, with the accuracy of our global VAT/GST tax database. It’s a working example of how hours of work by a deeply qualified tax expert – today charged by the hour – is done by AI in seconds.
UK and France open debate on AI taxes
In anticipation of lost income, governments are already looking at options to tax providers of AI. The UK’s Government Office for Science last month speculated on the tax regime responding to high financial gains from using AI to automate work.
France’s National Assembly recently proposed a law to establish a taxation regime for works generated by AI using works, whose origin remains uncertain. Its aim is to redistribute earnings from AI services which are based on original work, for example, sourced from the internet. And then ensure the creators are rewarded.
This follows the likes of South Korea, the first country to launch a ‘robot tax’ in 2017. Rather than taxing entities directly, the law reduces tax breaks that were previously awarded to investments into robotics.
Proposed AI tax frameworks
So while actual AI taxes in practise are rare, a number of models have been proposed:
- AI as taxable persons. Since the technology is replacing the income-earning and taxable human, AI chould be taxed on the same level. This idea has had the backing of Bill Gates. And not just for income tax – the AI function could be considered as taxable party of a VAT chain to the final consumer.
- Assessing the AI owner. This model would require determining an income for the AI function, and levying income taxes on the licensee or owner of the technology. Or there could be a flat tax, perhaps fixed for a period of time, say three years, when AI is first adopted.
- Taxing automation redundancies. Charging employers a levy on the relatively levels of headcount and redundancies made due to increased AI.
- Corporate income tax measures. Using existing corporate tax regime tweaks to extract extra taxes from adopters of job-replacement AI. This could include:
- reduced or nil capital allowance deductions of AI expenditures against tax bills. Although this would by overwhelmed by the growing preference of jurisdictions to give tax tech investment breaks to boost productivity;
- higher corporate income tax rates on heavily automated companies; or
- lower employer taxes to stimulate those that retained larger workforces relative to their sector.
Failure to meet the tests of good taxation and fear of stifling innovation may sink tax proposals
But other jurisdictions are slower to react as it’s not clear how to design a sound taxing framework. The main challenge is documenting AI activities within the wider manufacturing or service provision. One of the key components of any tax is a clearly definable supply. And also how to calculate the taxable benefit – a further basic requirement of any tax. Would the tax be transactional, like VAT, and therefore charged on the customer? This all means taxing AI would fall at the basic simplicity and certainty tests for a new tax.
Even if the nature of the taxable activity or transaction could be defined, it would be hugely complex to legislate for all the various business models and their future variations to prevent easy avoidance or even evasion. This would create a huge compliance burden and costs on businesses and policing tax authorities, undermining the efficiency and cost-benefit of such taxes. And any heavily codified laws would lack flexibility to adopt to emerging business modes.
Introducing unilateral taxes would also threaten to make countries uncompetitive, and scupper home-grown innovation. And could be easily avoided by multi-nationals relocating work or digital processes abroad. These two factors would mean any tax measure would fail the neutrality test of taxation. This sets the principle that taxes should do as little as practical to distort business behaviour, and apply consistently between traditional and digital commerce.
Fiscal levies for adopting AI would also dissuade improvements in productivity that power countries’ growth. Again, this would be highly unattractive for governments and businesses anxious to adopt and develop new technologies.
Tax won’t fix the AI income gap
It is clear that governments will have to invent new revenue sources as AI undermines their current tax revenues. But simple taxes on providers or users will not be to practical. Hence why numerous proposals in the past five years have failed any serious interrogation.
For them moment, the effect of AI on pubic revenues leaves us as exposed as Rome.
