Will AI create a gap in government tax revenues?

AI is increasingly taking work off our hands, including paid work. If large groups of people were to lose their jobs due to AI, what would that mean for government tax revenues? We called Thijs Knaap, Chief Economist at APG, to discuss.

Is there a historical parallel for what we are seeing with AI today?
“What we have certainly seen before is the fear that a new technology will cause large numbers of people to lose their jobs. We saw that during the wave of automation in the previous century, and even earlier in industries like textiles.

Those concerns were not always unfounded. Think of major national shocks such as the closure of shipyards or coal mines. These led to prolonged disruption, with entire regions, and sometimes generations, struggling to find work. So that parallel is definitely there. What we don’t know is to what extent AI will have a similar effect and how quickly it might happen. But that uncertainty itself is nothing new. Each time, you see it emerge again.”

So what is different about AI?
“What may be new is that AI is now affecting occupations that were previously considered relatively safe. It is no longer just about industrial labor or routine tasks, but about work that used to be seen as a stepping stone to the middle or upper class. For some groups, that comes as a shock, especially since it affects different people than in the past.

Research by Statistics Netherlands (CBS), for example, shows that people with college or university degrees are more likely than average to believe that AI could take over parts of their work. In the long run, there is usually still work available, but the mix of tasks changes. The key question is what new jobs will emerge. That is what makes the short term particularly uncertain.”

If AI replaces labor, does that also mean government tax revenues will decline?
“About one-third of government tax revenue in the Netherlands is directly linked to labor, roughly €105 billion. If you include social security contributions, it amounts to more than half. So the initial assumption is indeed that if labor shrinks, a large part of the tax base disappears as well.

However, it is a bit more complex than that. According to CBS figures, the public sector itself relies almost entirely on labor. If AI makes labor more efficient, government could potentially become smaller and require less tax revenue.

At the same time, the largest areas of government spending, such as social security and healthcare, are not easy to automate. Welfare benefits do not become cheaper with AI. And if unemployment rises, those costs will actually increase. So in a scenario where AI leads to significant unemployment, the overall impact would likely be negative for public finances. In that case, governments would have to look for alternative sources of revenue.”

Could taxing capital, or perhaps AI itself, be a solution?
“To some extent, yes. If labor becomes less important, capital would need to be taxed more heavily. But that is more difficult. Labor is relatively immobile, employees cannot easily relocate. Capital, by contrast, is highly mobile. Profits can be shifted to countries with lower tax rates.

You could also consider taxing AI itself, for example per interaction or per unit of use, per ‘token,’ as it is sometimes called. In that case, you would tax the use of AI in the same way labor is taxed. But that raises practical challenges. How do you measure it? And would you want the government to know exactly how people are using AI? That also raises privacy concerns.

The biggest issue is that much of this technology is developed outside the Netherlands. That makes effective taxation even more complicated. It is therefore important for European governments to invest in their own AI capabilities and actively support their development. This reduces dependency on U.S. technology. There was already a case for that based on strategic autonomy. Now there is a second reason: if you want to tax AI, it helps if the technology falls within your own jurisdiction. Governments could, for example, require certain services to be developed or used within Europe. That effectively acts as a subsidy and gives you more control.”

Is this scenario of mass unemployment caused by AI realistic?
“The reassuring view among economists is that things tend to work out in the long run. Historically, that has been the pattern: jobs disappear, but new ones emerge. Not always for the same people, though. And in the short term, you could certainly see a period like the 1980s, with prolonged unemployment due to restructuring and automation.

This can also create a self-reinforcing dynamic: people lose their jobs and reduce spending, companies sell less and cut back even further. You can end up in something resembling a recession or even a depression, despite rising productivity. At that point, it is up to government to step in and keep the economy moving.”

Could redistribution, for example through a basic income, be part of the solution?
“That is one of the options. If labor taxes generate less revenue, but you succeed in taxing AI or capital, that money still flows into the public coffers and can be redistributed through benefits. That is not necessarily a basic income in the strict sense, but the mechanism is similar. If there is structurally less work, the current system of temporary benefits could come under pressure. In that case, you might need a model that guarantees a minimum income at all times.

But again, we do not know whether that will be necessary, because new jobs are also likely to emerge. You could even argue that in the future, demand for labor will increase due to aging populations. A growing number of retirees will require services. In that sense, AI could actually help meet that demand. At the macro level, AI does not have to be a negative development. But the transition could be quite uneven, and that is where the real challenge lies.”