The Cost of AI "Tax Revenue losses & Cost of Universal Income"

Methodology

Our study sets out to evaluate the prospective influence of Large Language Model (LLM) based chatbot technologies on labour markets in OECD nations, and the subsequent effects on taxation structures. This exploration is articulated through three key dimensions: Jobs-at-risk, the potential cost to the system, and the future of taxation.

Jobs-at-risk

By "Job-at-risk", we do not consider complete replacement of an occupation by an AI worker, but rather the opportunity for companies to trim workforces as a result of productivity increases resulting from the use of LLM-based chatbot technologies.

We consider productivity gains from LLM-based chatbot technologies with the following characteristics:

• ChatGPT4 level capacity for language understanding and reasoning; and the limited capacity for creative problem solving and complex task resolution.
• Access to the internet and to internal company documents and databases

We do not consider the following developments that may be coming to the field:

• Further advancements in robotics based on LLM technology advancements.
• Shifts to telemetry; that is, if the occupation today requires in-person contact, the model assumes this stays the same.
• Development of domain specific AI systems that combine LLM technology with traditional AI and Machine Learning specifically developed to allow higher-order creative problem solving.

The model is driven by an evaluation of productivity gains on the tasks core to each occupation, as defined by the occupation database ONET, given these assumptions. The productivity gains for the tasks were averaged for each occupation, weighted by the importance of the task per the ONET database, giving an aggregate productivity increase; and then converted to a redundancy percentage.

The final jobs at risk was estimated based on occupation/industry matrices for each country. These matrixes were estimated using the US occupation/industry mix as a model; ILO estimated breakdowns of employment by industry were multiplied with the equivalent US occupation by industry matrix achieving an estimated occupation/industry matrix for each OECD country. The final total number of employed persons per occupation per industry was multiplied by the redundancy percentage for each occupation.

Potential cost to system

To estimate the potential cost to the tax system as a result of redundancies, we combined both lost tax revenues due to loss of employment with the cost of supporting redundant workers.

As Universal Basic Income has been proposed as a solution to the existential threat of AI on labour markets, we took the approach of using the Universal Basic Income regime proposed by ILO to estimate the cost of supporting redundant workers.

Future of taxation

In the scenario considered in this study is limited to the use of LLM advancements to trim workforces–it ignores the potential for higher quality services and new services and industries altogether–the benefit of LLM is solely concentrated to the bottom line of corporations. Therefore, we present the required increase in corporate tax revenues in each country should tax systems *only* seek to cover the potential cost to the system through corporate tax increases. This increase is compared to the increase in overall tax revenues should the potential cost to the system be carried by tax revenue increases across the board.

Column notes

Population

The 2022 population of the country, in millions of persons.

Source: OECD

GDP/Capita (USD)

The 2022 Gross domestic product per capita, in 2022 exchange rate US Dollars.

Source: OECD

Employed Persons

The total number of employed persons (full-time or part-time) in the country in 2022, in thousands of persons.

Source: OECD

Jobs-at-risk

The total number of jobs at risk of being made redundant due to productivity increases due to LLM-based chatbot technologies, in thousands of persons. Estimates are provided on a sector basis, per the North American Industry Classification Standard.

• Agriculture: NAICS code 1
• Industry: NAICS codes 2-3
• Services: NAICS codes 4-8

Sources: ILO, ONET, ChatGPT (See methodology notes above)

Potential cost to system

Estimated lost tax revenues due to unemployment, per person

The annual value of lost tax revenues for each worker due to redundancy, assuming new employment is not found. Includes income taxes, payroll taxes and compulsory social contributions.

Calculated as total tax 2022 tax revenues from income, payroll and social contributions divided by *Employed Persons*.

Sources: OECD

Estimated average cost of universal income, per person

Per the ILO proposal of a minimum viable universal income, the base universal income in rich countries should be set to 50% of the median disposable income in the country. Using available OECD data, the universal income is calculated as 50% of the average disposable household income per capita.

Sources: OECD

Estimated total cost to system, per person

The sum of *Estimated average cost of universal income, per person* and *Estimated lost tax revenues due to unemployment, per person*.

Total potential cost to system

The product of the *Estimated total cost to system, per person* and total *Jobs-at-risk*, giving the total annual potential cost to the system under the assumption that:

• Companies use productivity increases only to reduce total headcount through redundancies
• Redundant workers are not able to find new employment
• AI displaced workers are provided with the ILO proposed Universal Basic Income to endure redundancy

Future of taxation

Required Tax Revenue Growth to cover cost to system

The % increase in required total tax revenues in order to absorb *Total potential cost to system*, assuming:

• constant government expenses

Sources: OECD

Required Corporate Tax Increase to cover cost to system

The % increase in required total corporate revenues in order to absorb *Total potential cost to system*, assuming:

• constant government expenses
• *Total potential cost to system* is only absorbed through increases in corporate taxes.

Sources: OECD