How governments might respond to the economic upheaval of TAI

Thanks to @SiobhanBall for reviewing and providing feedback on an earlier draft of this post. Any errors remain my own.

Summary

• Widespread labour displacement by transformative AI (TAI) will create severe fiscal pressures for governments who rely heavily on taxing labour income.
• Governments have various possible responses, such as spending cuts, tax-base shifts, inflation, or default. These options vary widely in their level of disruption as well as their welfare impacts.
• Which options a country ends up taking will depend on various factors, including how seriously policymakers take these risks now. Unprepared governments are likely to find themselves forced toward the most disruptive options.
• Productivity growth caused by TAI is unlikely to solve this: if AI increases profits at the expense of wages, tax revenues fall even as GDP rises.
• A more equal distribution of resources in a post-TAI world could help mitigate near-term instability as well as longer term risks around extreme power concentration.

The Problem

The economic benefits of Transformative AI (TAI) will not automatically be distributed evenly. As AI becomes a better substitute for human labour, we are likely to see a greater share of returns in the economy going towards capital holders and a smaller share going towards workers.

As others have pointed out, this labour displacement will pose challenges for governments in developed countries, which rely heavily on labour for tax revenues. At the same time, support for the growing number of people made redundant will strain government finances. Meanwhile, debt levels in many developed countries are still high from the Covid-19 pandemic, and aging populations have put further pressure on national budgets.  

This post explores what these dynamics might actually look like and the possible responses governments could unilaterally take. I focus in this post on countries that derive a significant share of government revenues from income taxes on labour (which is almost all developed countries).^[1] 

Government revenues might sound like a dry topic that people consider irrelevant once we get a superintelligent AI. However, work to manage widespread labour displacement seems sorely neglected, with only a few organisations directly working on this^[2] and the mainstream discourse lagging behind. I believe that managing the economic transition well is crucial to mitigating risks like extreme power concentration. Put another way - I would feel a lot better if AGI were developed by Norway than by Saudi Arabia.

Assumptions and caveats

Assumptions:

1. The timeframe I am thinking about is roughly the next 10-20 years.
2. I assume we get TAI, which may or may not involve AGI (AI capable of handling all cognitive tasks at least as well as humans). I am not assuming ASI (an AI agent or agents with intelligence far exceeding humans).

3. Humans have roughly as much control over AI as they do now. I am assuming AI as normal(ish) technology in terms of the amount of human control.^[3]

4. Labour displacement somewhere above 10% but less than 100%. This is a very large range and will depend on things like how rapid advancements in robotics are, and how rapidly AI diffuses throughout the economy. The number doesn’t really matter for the purposes of this post which is pitched at a higher level of abstraction.
5. As a result of assumption 4, I assume inequality will increase and jobs will be much less viable as a path to upward social mobility.

Caveats:

• My background is in tax and fiscal policy, not in monetary policy. I have tried to qualify my predictions about monetary policy accordingly, but I welcome views from people who have more experience in this area.
• Please flag any factual inaccuracies as you see them.

Possible governmental responses

There are various steps that governments could unilaterally take^[4] to respond to their inequality pressures. Some of these steps will be much less disruptive than others.

At a high level, the steps that a government could unilaterally take include:

• Shifting away from taxes on labour towards other tax bases. Other possible tax bases include consumption,^[5] land,^[6] inheritance,^[7] wealth,^[8] and capital income.^[9] It seems likely that TAI will generate enormous economic rents (i.e. profits exceeding the “normal” returns to capital), even in countries without any AI companies. To the extent that countries can tax these economic rents effectively, they should be in a good position to manage the labour displacement challenges of TAI.^[10] 

• Cutting spending. There is only so much that governments can cut before this starts to have real welfare impacts. Moreover, while spending cuts can (in theory) be done in a thoughtful, planned way that minimises welfare impacts, governments caught unprepared by the speed of labour displacement are more likely to engage in hasty, reactionary spending cuts, some of which may end up costing the government more in the long-term.

• Financial repression (capital controls, interest rate caps). Interest rate caps limit how much savers can earn. When inflation is high and interest rates are capped at low rates, this can result in negative real interest rates. In a liberalised financial system, savers would move their money elsewhere, but capital controls restrict this, giving the government a source of cheap funding. Financial repression played a large role in financing the industrialisation of Japan, Korea, Taiwan and China, before they developed robust income tax systems.^[11] 

• Printing money (e.g. through Quantitative Easing (QE) or helicopter payments). Printing money can cause inflation,^[12] which is effectively a tax on savers. There are definitely downsides to doing this, as inflation is politically unpopular and very high levels of inflation (hyperinflation) have very real costs. For this reason, many developed countries have independent central banks, so that politicians cannot control the money supply for short-term political gain. That said, independent central banks can still expand the money supply, and are likely to do so if they see high levels of unemployment. Note that this solution is far less effective for governments whose debts are denominated in a foreign currency or who don’t control their own currency (e.g. countries using the Euro or who peg their currencies to the USD).  

• Confiscation of property without fair compensation. In developed countries, this probably won’t take the form of government officials seizing physical belongings, but there are precedents of financial assets being frozen, forcibly converted,^[13] or subjected to one-off levies.^[14] 

• Default/debt restructuring. If the other methods above still aren’t enough, a government might default on its debts. But every country's debt is another country/person's asset, so it is not possible for everyone in the world to default all at once. Looking at the list of “creditor nations” with the most positive Net International Investment Positions, it seems like Asia and Europe will feel the most pain.^[15] However, creditor nations are also more likely to hold shares in companies that benefit economically from TAI in the first place, so may enjoy a compensating windfall.^[16] 

Which steps are more likely?

Which steps any particular government is likely to take will depend on many things. I suggest the following non-exhaustive list of factors in no particular order, but please let me know any other factors you think could be important:

• How seriously policymakers are taking these labour displacement risks - countries that are better prepared will have more options available, because some options (like creating new taxes) simply cannot be done in short timeframes.  
• Whether the country currently has extractive or inclusive institutions - countries with more extractive institutions are less likely to shift taxes from labour to capital, and more likely to respond by cutting spending.
• How rapid labour displacement is - this depends not only on the speed of diffusion but also things like employment law - longer notice periods and labour protections will slow the rate of displacement.
• How much fiscal headroom the country has - countries with high existing levels of debt (especially where the debt is held by foreigners) may have less time to act and fewer options available.
• How quickly taxes can be raised - it is much easier to raise rates on existing taxes than it is to create entirely new ones. For example, VAT is highly efficient at raising revenue, so countries already with VATs in place may be able to raise a fair amount of revenue just by increasing their existing VATs. New taxes are slower not simply because of the time required to design them, but also to set up the administrative capacity to enforce them - a new tax that only exists on paper doesn’t raise any revenue.
• Random quirks like who happens to be in power at that time.

For these reasons, I expect that different countries will take different approaches. The world has many different countries at different points on the extractive-inclusive spectrum. Those that are already closer to the ‘inclusive’ end (which is most of the developed world) might be persuaded to distribute the gains from TAI broadly. Countries where AI diffusion is slower may also be able to learn from the successes and failures of countries that experienced faster diffusion. 

How quickly will countries need to respond?

I don’t know how quickly AI will displace jobs. But governments can be pretty swift at times:

• Emergency powers. Governments moved quickly in previous wars and during Covid.
• Tax rate increases can be fast. Establishing a new tax takes a long time, but existing taxes can usually be increased quickly - sometimes even overnight.
• Monetary policy. Monetary policy can help smooth shocks. As unemployment rises, central banks are likely to loosen monetary policy (lowering interest rates and perhaps printing money).

Perhaps the bigger point is that governments don’t have to respond instantaneously to shocks. Lots of events, including Covid and bog-standard recessions, trigger sharp declines in government revenues at the same time that government spending increases. Governments don’t usually try to balance their budgets during these periods, and just come out the other end with more debt.^[17] So governments could have several years to rebalance their tax bases, even if widespread job displacement occurs within months.

That said, I don’t think most governments are taking these risks seriously enough, and the above should not be construed as an excuse for complacency. Most of the better options I’ve described - like taxing capital and economic rents - will take time to implement properly.^[18] If a government is not prepared, they may be left with the much more unattractive options like hyperinflation or default. Helping governments prepare for the labour displacement so that they aren’t forced to take the more disruptive steps could therefore be an impactful and tractable near-term intervention.

Could productivity growth be the solution?

In the mainstream discourse, people worried about high government debt and fiscal pressures from an aging population often point to “AI” as a possible productivity booster that will solve things. However, increased productivity doesn’t translate directly into higher tax revenues. It only translates into higher tax revenues through higher wages, salaries and profits. Yet, because labour income is often taxed at higher effective rates than capital income,^[19] if profits increase at the expense of wages and salaries, overall government revenues would fall.

If TAI increases productivity in the government sector, that might result in some cost savings that allow for spending cuts. But I am skeptical this will be enough, as wages and salaries usually make up a modest share of government spending.^[20] Handing over core government functions to AI as a cost-savings measure should also be highly concerning to those worried about misalignment.

What about global inequality?

This post deals only with national inequality and labour displacement pressures, and focuses primarily on developed countries. I have written about global inequality in a separate post. However, I do believe the issues are linked - if developed countries do a reasonable job at managing their internal inequality pressures and preserving inclusive institutions, I expect things will go better for the Global South.   

Possible notes for optimism

Labour displacements are very disruptive. But I will leave with two possible notes for optimism.

First, the disruption is not the same as a physical disaster. Physical disasters like earthquakes, hurricanes, pandemics and war destroy value. If AI displaces labour, that means it should be creating value. The challenge is one of distribution - a very real challenge for sure, but tractable at least in theory.

Secondly, interventions are possible. Luckily, we are starting from a position where most developed countries have reasonably inclusive institutions, rather than the opposite. I do not think highly unequal outcomes such as those described in The Intelligence Curse are inevitable. Norway was able to escape the resource curse because it had already developed inclusive institutions before it discovered oil in the 1960s. Botswana similarly sidestepped the resource curse in part because its institutions were relatively inclusive before it found diamonds in the 1970s. My hope is that those who live in countries with relatively inclusive institutions can push for solutions that will preserve, and maybe even broaden, that inclusivity.

Politicians and policymakers are highly concerned about their tax bases eroding and should also be highly motivated to avoid the most disruptive options I’ve listed, like hyperinflation and default. The problem is that I don’t think most have been taking the labour displacement risk as seriously as they should - but that is changing, and we may be able to help that along in constructive ways.

1. ^{^}

    Exceptions may include financial centres like Hong Kong, Singapore and Luxembourg and the Gulf States.

2. ^{^}

    Convergence Analysis and Windfall Trust being two I know of.

3. ^{^}

    As regards the question of human control. I am less bullish than the authors of that article about the ability for displaced workers to find other (decent) jobs.

^{^}

 I set aside the possibility of high levels of global coordination to share the benefits of TAI, as I find them highly unlikely in the near term.

^{^}

 Contrary to popular belief, consumption taxes are not necessarily regressive, depending on how the tax raised is spent. Broad-based consumption taxes tend to be very efficient revenue raisers, so a consumption tax that raises revenue to be spent in a progressive way can make the overall tax and welfare system more progressive.

^{^}

 Even if the land in question does not form part of the AI supply chain, productivity increases can still give existing landholders a windfall gain because the supply of land is fixed.

^{^}

 In practice, inheritance taxes do not tend to raise a lot of revenue because rates are low, and evasion increases at higher rates. Revenues also tend to be quite lumpy and unpredictable because wealth tends to be concentrated and deaths can be unpredictable. In larger countries like the US, the law of large numbers may help smooth out some of the lumpiness.

^{^}

 Wealth taxes are hard to enforce unilaterally, except at very low rates. They are also inefficient because they undertax economic rents and overtax the normal return.

^{^}

 Taxing capital more heavily could mean an increase in the corporate tax, or increasing the tax rates on dividends, interest, royalties and capital gains. Some have also suggested novel taxes like robot or automation taxes, which come with various implementation challenges.  

^{^}

 This is easier said than done, because identifying economic rents can be quite tricky. Various countries are able to tax their location-specific resource rents through royalties and resource taxes. Norway is the only country I know of that has successfully levied higher taxes on general economic rents using their shareholder model.

^{^}

 See How Asia Works by Joe Studwell.

^{^}

 Printing money may not cause inflation if the amount printed matches real output growth.

^{^}

 In 1933, the US passed Executive Orders that forced people to convert their gold at a fixed price of $20.67 per ounce, with non-compliance punishable by up to 10 years imprisonment. The market price for gold was estimated to be more like $30 per ounce during this period.

^{^}

 For example, in 1992 Italy applied a 0.6% overnight levy on all bank deposits.

^{^}

 Note this isn't exact - the NIIP isn’t limited to government debt, and includes household debt, corporate debt, and even equities. But it can give us a rough idea of where the global imbalances lie.

^{^}

 Of course, the people who hold the shares and get the windfall may not be the same as the people who hold the debt and suffer the default.

^{^}

 The idea is they’ll pay this debt off over time, but this doesn’t always happen.

^{^}

 Unless a government already has them set up.

^{^}

 The main reason for this is that capital is often mobile and more easily hidden, so high tax rates may just cause capital to move offshore. Some countries like Norway cleverly manage to tax location-specific economic rents at very high rates, but most countries do not.

^{^}

 Average 16% of government spending in high income countries and 20% in upper-middle-income countries as at 2023 according to Our World In Data.

Show all footnotes