Automation and Equity: Taxing AI to Safeguard Jobs and Government Budgets

Pervasive AI usage by businesses will inevitably lead to job losses. This could be outright replacement of a worker by AI or partial automation that reduces the overall number of workers required. The government must react to this by considering a tax on AI usage, in order to pay for the eventual large unemployment benefits bill that will come due.

Whether AI fully or partially replaces workers, it will lead to an increase in overall productivity and therefore a greater economic output for the country. However, it comes at the financial (and social) cost of a decrease in the human workforce.

Once AI increases the output of a company, there is no need to retain full staffing levels, unless revenue increases as well. However, if revenue does not increase, then payroll expenses will be decreased due to the over-supply of production without a corresponding change in demand.

The economics of imposing an AI-tax in response to lay-offs merits deeper consideration. An AI’s cost is negligible compared to a human’s. However, in many cases, while an AI could produce 100% of the output of a person quantitatively, it may only produce 80% of the same quality as a person. In such cases, replacing a person with AI requires either 1) accepting a lower quality product for a much lower price to consumers or 2) retaining a person to oversee the AI part-time in order to bridge the quality gap.

Imagine a $50k salaried worker is replaced by an AI, which costs about $500 in licensing fees. Assuming an AI’s output is only 80% of a human’s quality, the company has the two options listed above 1) sell a lower quality good, or 2) use a part-time employee’s input.

In the first scenario, the company replaces a $50k employee with a $500 AI, and produces around four-fifths of the quality of the ex-employee’s output, which works out to $40k of output. This decreases the company’s output (and therefore revenues) by $10k but reduces its expenses by $49.5k – a net benefit of $39.5k for each $50k employee removed.

In the second scenario, assume an employee works part-time to adjust the AI’s 80% quality up to the original 100%, which takes the employee 20% of their own time. Now, in order for the company to produce one full person’s output, it must pay a total of $10,500: the AI licensing costs of $500 plus $10k for the employee’s salary.

In theory, a partially or fully laid-off employee goes on unemployment benefits and requires less money than their original full-time net income. Their $50k salary had likely resulted in a net income of $40k. For a fully unemployed individual, they would require around $30k from government assistance (instead of the $40k they had while fully-employed). In the scenario where the employee works part-time to run the AI, the employee receives a $10,000 salary (assume no tax) and would require only $20k from the government.

Consider what it therefore costs the society to produce $50k of output. Pre lay-offs, the company had spent $50k in salary, of which $40k went to the employee and $10k to the government in taxes.

After the lay-off, in scenario 1, the company spends $500 on AI and nothing in salaries. The government has lost out on $40k: $10k of taxes from the company and $30k in unemployment benefits being paid out. Essentially, the company gains $49,500 at the cost of the government losing $40k.

In scenario 2 – where the employee stays on part-time – the company produces $50k of output at a total cost of $10,500, which consists of $10k in salaries and $500 in AI fees. The government loses out on a total of $30k, which accounts for $10k of taxes and $20k in unemployment benefits. In this scenario, the company has gained $39,500 at the cost of the government losing $30k.

There is clearly a large shift of wealth from the government to the company in the lay-off scenarios, even though in both scenarios the society overall gains $9,500 of economic output which was previously untapped. This means the overall economic pie has increased by $9,500 as compared to the pre-layoff era. The government’s role is therefore to determine how much of this gain should be split amongst between the government and the company.

These figures show a strong reason to tax the company for the whole $30k (scenario 2) or $40k (scenario 1) that the government has lost. On the whole, after such a tax, the company still stands to gain $9,500 for each employee replaced by AI, while the government breaks even. (The individual, for their part, will require a new source of purpose, which poses a teleological issue that is beyond our scope.) This tax ensures: the individual is not destitute, the government breaks even, and the company saves itself $9,500 (about 20% of the employee’s salary), while the society enjoys the same level of goods/services from the company.

Of course, some limits are required. An AI tax should roughly equal the difference between the new cost of licensing the AI and the savings produced by laying off the employees. If the tax is too low, the government will be in a net loss position and society will have effectively redistributed money to the corporation that should have been earmarked for the unemployed. There is also concern, on the other hand, of imposing a tax that is higher than the profits created by the lay-offs, which would penalize the company for using AI. If companies are penalized for using AI, they will opt to avoid AI, resulting in them retaining low-efficiency structures, which would leave money on the table. Without AI, the economic pie will not grow.

In order to maximize the overall economic benefit to society, AI usage and adoption should be encouraged, not penalized. While there should not be a very high tax, it must be sufficient to avoid bankrupting the government while enriching private entities in the process.