If AI takes away people's jobs, who will pay taxes?
A few days ago, OpenAI released a report titled "Industrial Policy for the Intelligence Age". In this 13 - page document, OpenAI put forward a series of "people - centered policy proposals" designed to cope with the drastic changes brought about by super - artificial intelligence technology, covering multiple aspects from adjusting working hours to reconstructing the infrastructure system. Among all these policies, "levying an AI tax" is undoubtedly the most eye - catching one. OpenAI researchers suggest that in order to address issues such as the reshaping of work patterns and the erosion of the tax base caused by AI, the tax base should be reconstructed. "Increase the reliance on capital income, including raising capital gains tax for high - income groups, increasing corporate income tax, and implementing targeted taxation on continuous AI - driven revenues; at the same time, explore new types of taxes related to automated labor, such as AI tax and robot tax." This proposal has once again brought the discussion of "taxing AI" into the public eye.
With the rapid development of AI technology in recent years, many celebrities have proposed the idea of "taxing AI". For example, Bill Gates, the founder of Microsoft, has called on many occasions for taxing enterprises that use AI and robots to replace human labor at a tax rate equivalent to the tax burden level of the replaced employees' salaries, with the tax revenue used for social security, retraining, and employment stability. Geoffrey Hinton, the "father of deep learning", pointed out in an interview that the large - scale replacement of human labor by AI may lead to the shrinkage of the payroll tax base, so it is necessary to tax AI to maintain the finances of the welfare state. Dario Amodei, the founder of Anthropic, suggested levying a "Token Tax" on the use of large models for redistribution and public welfare.
However, so far, most countries in the world have not levied taxes related to AI. As far as I know, the only exception is Kazakhstan, which began to levy value - added tax on AI services such as ChatGPT in August 2025. In addition, other economies have adopted a relatively cautious attitude towards the AI tax.
01
Tax Challenges in the AI Era
The rapid development and popularization of AI technology will not only have a huge impact on the existing economic structure but also pose severe challenges to the current tax system. Specifically:
The first challenge is the erosion of the tax base and the depletion of tax sources that may be brought about by the widespread application of AI.
In traditional economic intuition, economic growth often means an expansion of employment, which in turn means a steady increase in tax revenue: more people work, pay more personal income tax and social insurance premiums, and the government will have more sufficient financial resources to maintain the public service and redistribution system. According to this logic, if AI can really bring about a significant leap in productivity as many tech optimists predict, the government should be able to easily obtain far more tax revenue than in the past.
However, many studies have shown that the situation may not be so. At least at present, the productivity improvement brought about by AI is far lower than the estimates of tech optimists. For example, according to the estimate of Daron Acemoglu, a Nobel laureate in economics, the annual productivity growth brought about by AI is only 0.06%, which is negligible. At the same time, the impact of AI on the tax base is quite serious. In reality, there are a wide variety of tax items, but the tax bases they correspond to are nothing more than the following: labor income, capital income, wealth stock, and consumption. For many economies, taxes based on labor income and consumption account for a relatively high proportion of their tax structure. At the same time, due to the need to protect property rights and encourage investment, the taxation of capital income and wealth is relatively restrained.
However, under the impact of AI, the two major tax bases of labor and consumption are severely eroded. We know that current AI is largely an automation technology, and its wide application means that a large number of labor forces will be replaced. In another widely cited study, Acemoglu and his collaborators found that automation technologies such as industrial robots will significantly reduce the market demand for labor. In the US labor market, for every additional robot per thousand workers, the proportion of the employed population in the whole country will decrease by 0.2%. When AI is combined with industrial robots, the squeeze on labor demand is even more imaginable. As the number of employed people decreases, the tax base of labor income tax will be directly impacted. At the same time, the market wage level and workers' income will also decrease, which will lead to a decline in consumption and a weakening of the consumption tax base. On the one hand, the growth is less than expected, and on the other hand, the tax base is damaged. The combination of these two forces will lead to a decrease in the tax revenue of labor income tax and consumption tax.
The second challenge is that in the face of the impact of AI, the government's public expenditure may increase significantly.
At least for some time, the impact of AI on the employment market may lead to relatively serious "technological unemployment". In this case, in order to provide training for the unemployed to help them re - enter the job market and provide necessary living guarantees for some people who are difficult to re - employ, the government needs to significantly increase its financial expenditure. To support the increase in expenditure, the government is required to develop more tax sources and obtain more tax revenue.
The third challenge is that in the AI era, the difficulty of levying traditional major taxes will increase significantly.
In the AI era, the production process will increasingly rely on intangible assets such as algorithm models, data resources, and computing power infrastructure. Traditionally, these assets are not taxed. At the same time, more and more labor activities exist in the form of gig work, freelancing, or digital platform tasks, and income is more fragmented and non - standardized, which makes it more difficult to tax labor income. Under the combined effect of these two forces, the cost of levying traditional taxes will become higher.
The fourth challenge is that the application of AI may cause a serious mismatch between economic activities and taxes in space.
Traditional production and consumption activities usually have a clear geographical attribution, so it is easy to levy corresponding taxes in specific countries or regions. However, algorithm - driven production is naturally cross - border. A model can be trained in one place, deployed in another, and serve global users, and its value - creation process is divided among multiple jurisdictions. This makes tax collection face an increasingly prominent problem: the generation of the tax base is global, while the tax - collection power is still regional. Under this mismatch, the problems of profit transfer and tax - base erosion are further magnified.
02
Reasons for Levying the AI Tax
Under the superposition of the above multiple challenges, "whether to tax AI" is no longer just a policy option but gradually shows a certain institutional necessity. In fact, as AI plays an increasingly important role in the economy, if corresponding tax arrangements are not established for the new type of production factors represented by AI, not only will the financial system become increasingly overburdened, but also an unfair "scissors gap" will form between the two major factors of labor and capital. In this sense, taxing AI is not a simple intervention in technological progress but a institutional response aimed at repairing the tax - base structure, maintaining financial sustainability, and ensuring social fairness.
First, taxing AI is not only a compensation for the reduction of the tax base but also a measure to respond to the reconstruction of factor returns caused by changes in the production structure.
With the wide application of AI, value creation is shifting from being mainly labor - based to a composite system that relies more on capital, algorithms, and data. This transformation directly leads to changes in the income - distribution pattern: the income that was originally presented in the form of scattered wages is increasingly transformed into corporate profits and capital returns. In this case, if the tax system still mainly relies on labor income and consumption as the tax base, even if the total economic volume keeps growing, the fiscal revenue may have a structural shortage, and at the same time, the distribution imbalance between capital and labor will be magnified.
Therefore, the significance of taxing AI can be understood as an institutional arrangement of "realigning the tax base": it attempts to bring the value that has been transferred from labor to the capital and technology system back into the taxable scope, so as to maintain the stable operation of the financial system. At the same time, this adjustment also has important distribution implications. Driven by AI, the change in the relative returns of factors may lead to a continuous increase in capital income and a relative decrease in labor income. If the tax system fails to respond to this, inequality will become an endogenous result. By appropriately taxing AI - related income and using the relevant income for public expenditure and redistribution, this trend can be alleviated to a certain extent, so that the benefits of technological progress will not be overly concentrated. It can be seen that taxing AI is not simply increasing the tax burden but a necessary correction of the existing tax system in the context of changes in the production function and income structure.
Second, in the context of the increasing difficulty of tax collection, taxing AI can also reduce the overall tax - collection cost by "changing the tax object".
One of the important reasons why the traditional tax system relies on labor and consumption is that they have a high degree of observability. However, as mentioned above, after the rise of the platform economy and the gig economy, this advantage is weakening. On the contrary, although AI - related activities are more complex in form, their key elements are often concentrated in a few enterprises or platforms, and thus have a higher degree of concentration and monitorability. Therefore, if a reasonable tax - collection method can be designed, such as levying taxes based on computing power or usage, it is possible to cover a large - scale economic activity with a relatively low tax - collection cost. In this sense, the AI tax is not only the development of a new tax base but also a structural optimization of the tax - collection method.
Third, in the face of the problems of spatial mismatch and cross - border flow, taxing AI can also play an important "anchoring role".
In the current system, multinational enterprises can transfer taxable income to low - tax regions through profit transfer, intangible - asset pricing, etc., thereby weakening the tax - collection ability of each country. The value related to AI often depends more on specific technological infrastructure and market demand. If tax rules can be designed around these elements, such as linking them to the location of users or the place where computing power is used, it is possible to alleviate the problem of tax - base loss to a certain extent. Although this process highly depends on international coordination, the reform of digital - economy taxation in recent years has shown that in the context of a highly mobile tax base, it is not completely impossible to redistribute the tax - collection power through rule innovation.
03
Theoretical Disputes over the AI Tax
It should be noted that although taxing AI can effectively help the government cope with the tax challenges brought about by AI, there are still many theoretical disputes around issues such as whether the AI tax should be levied, when to levy it, and how to levy it.
The first dispute is whether taxing AI will affect economic efficiency and inhibit technological innovation.
In classical public - finance theory, there is an important conclusion: to avoid distorting capital allocation, taxes on capital income should be avoided as much as possible. Among the existing "AI tax" proposals, the most popular one is the so - called "Robot Tax", that is, taxing the AI equipment used by enterprises. Many scholars believe that if this proposal is adopted, it may violate the above - mentioned conclusion of public - finance theory, distort the capital allocation in the economy, and thus damage economic efficiency. In addition, some scholars believe that as a general - purpose technology with strong spillover effects, the wide application of AI not only directly improves productivity but also brings about broader economic growth through industrial linkages. In this context, imposing an additional tax burden on AI or automation is equivalent to artificially increasing its use cost, which may slow down the speed of technology adoption.
Not long ago, a paper published in the "Journal of International Studies" seemed to provide some evidence for this view. In this paper, the author simulated the economic consequences of levying a "Robot Tax" to increase tax revenue while reducing taxes to encourage employment. The results showed that the levy of the "Robot Tax" would not only significantly reduce the economic growth rate but also reduce employment, and the economic cost it generates may be higher than the tax revenue it brings.
The second dispute is whether it is too early to levy the AI tax now.
Some scholars point out from the perspective of AI development that at present, AI is still in a relatively early stage of development, and its industrial application is not yet sufficient. In this case, if the AI tax is levied rashly, it may hinder the normal development of AI and suppress its growth potential that has not been fully realized. Other scholars start from the perspective of fiscal theory and also believe that the levy of the AI tax should not be rushed. For example, although Professor Anton Korinek of the University of Virginia is one of the active advocates of the "AI tax", he also opposes the early levy of the "AI tax". In his view, when the impact of AI on the economy is still relatively small and the two traditional tax sources, labor income and consumption, are still relatively sufficient, the optimal tax structure should still be mainly based on labor income tax and consumption tax. Only when labor income has shrunk to a very small share and is difficult to form an independent tax source should taxing AI capital be adopted as a second - best option.
The third dispute is whether the AI tax can really effectively improve distribution?
As mentioned above, one of the important reasons for supporting the AI tax is that it helps to alleviate inequality. However, this is not self - evident in economics. In practice, the actual effect of taxation depends on the tax incidence. Enterprises may transfer the tax burden to consumers by raising prices or to workers by lowering wages. In the context of automation, the bargaining power of labor has already declined, and this transfer is even easier. Therefore, some researchers believe that "taxing AI" is not necessarily equivalent to "taxing capital", and its distribution effect may be much weaker than intuitive judgment. In addition, there is a deeper question: does the source of inequality really lie in AI itself? If the problem mainly comes from education, skill structure, or market competition pattern, simply relying on tax tools may not fundamentally solve the problem. This makes the effectiveness of the AI tax at the distribution level an open question.
The fourth dispute is how to design the structure of the AI tax.
For the design of the tax structure in the traditional economy, there is already very mature research and many consensuses have been formed. However, for the tax structure in the AI economy, the existing research is still very insufficient. What form should the "AI tax" take? How should different types of taxes be combined? How should the combination of tax types be adjusted at different stages of development? There are still great disputes about these issues. Behind each proposal, there are respective supporting and opposing opinions, and it is still difficult to reach a consensus in the short term.
04
Difficulties in Implementing the AI Tax
Compared with the above - mentioned theoretical disputes, the problems faced by the "AI tax" at the practical level may be more. Specifically:
First, and the most fundamental problem, is the difficulty in defining the tax base. The operation of the tax system depends on a clear definition of the tax object. For example, what constitutes "income", what belongs to "capital", and what is "consumption" are all relatively clearly defined, so it is relatively easy to tax them. However, AI itself is a highly vague and constantly evolving concept. From simple automation scripts to complex deep - learning models, from algorithm optimization embedded in the production process to generative systems for end - users, its technological forms present a continuous spectrum rather than discrete categories. In this case, any attempt to use AI as an independent tax base will inevitably face the problem of boundary demarcation: if the definition is too broad, almost all digital technologies may be included in the taxable scope; if the definition is too narrow, it is easy to be circumvented, thus weakening the policy effect.
Second, it is the problem of positioning the taxpayer. The traditional tax system can operate stably because taxpayers have a clear legal identity, that is, individuals or legal entities. However, the AI system itself does not have a legal subject status. It can neither own property nor assume obligations. Therefore, under any institutional design, the so - called "AI tax" ultimately needs to be levied on enterprises or users. This brings a structural dilemma: if the tax is still borne by enterprises, how should the relationship between it and the existing corporate income tax be defined? Should it exist as an additional tax or replace part of the existing tax types? If not handled properly, it is easy to lead to double taxation or the complication of the tax system, which will instead reduce the overall efficiency. Some more radical ideas attempt to regard AI as a "quasi - legal entity" and endow it with a limited legal subject status, so that it can become a direct taxpayer. However, this idea will quickly lead to deeper legal - philosophical problems: if AI can pay taxes, should it also enjoy rights and assume responsibilities? How should its legal status be defined? Obviously, these issues are far from reaching a consensus.
Third, there are information and measurement constraints in the tax - implementation process. Different from traditional labor income or physical capital, the value generation related to AI is often embedded in a complex production process and is difficult to be separately separated and measured. For example, in a company's profit, how much should be attributed to the AI model and how much comes from the brand, management, or market environment do not have a clear division standard. If there is no reliable measurement basis, any taxation based on the "AI contribution" may rely on highly subjective evaluations, thus increasing disputes and compliance costs. This is why some policy discussions in recent years have begun to turn to more observable indicators such as data usage and computing - power consumption: although they are not perfect, they are at least more measurable technically.
Fourth, the cross - border issue still makes the implementation of the AI tax very difficult. Different from traditional production, the development, training