New York City Mayor Zohran Mamdani’s pledge to raise the minimum hourly wage to US$30 captures a dilemma that is haunting most advanced economies. Even as wage floors rise, workers feel less secure. Yet by responding with blunt tools like the minimum wage, policymakers are overlooking the deeper problem: The decoupling of human labor time from economic value.
For two centuries, labor markets were anchored in that relationship. Wages, contracts and social protections all assumed that time was a reliable proxy for output, but artificial intelligence (AI) has severed this link. Medical diagnosticians who spent years mastering pattern recognition now compete with systems that process cases in seconds. Lawyers with AI agents can complete tasks in minutes that would take hours otherwise. The question is no longer whether minimum wages risk overshooting, but whether time-based compensation remains meaningful at all.
What is to be done?
Illustration: Louise Ting
For starters, fiscal frameworks must recognize machine time as a distinct production input and price it accordingly. This is not to suggest a “robot tax,” but rather something closer to a metered input levy, using data that are already tracked in real time.
Trying to tax firms directly for their “degree of automation” has never worked, because such processes are organizationally opaque. Algorithmic decisionmaking embedded in software is nearly invisible and attempts to define a “robot” have proved legally absurd. Does Excel count?
Moreover, automation has long been deemed essential for maintaining global competitiveness, often posing a “problem” only for those offering manual labor.
Policies that penalize productivity tend to be non-starters. The most prominent effort came in 2017, when the European Parliament considered a proposal — embedded in Member of the European Parliament Mady Delvaux’s report on robotics and AI — to levy a “robot tax” on companies that replaced workers with automated systems. The idea was framed as a way to fund social protection and retraining, yet it was explicitly rejected in the final plenary vote amid concerns that the measure would undermine innovation and damage Europe’s competitiveness. Since then, robot-tax proposals have remained largely theoretical across EU member states, reinforcing the political reality that policies perceived as penalizing productivity struggle to advance.
The answer might be to tax “AI-hours”: the computational time consumed by AI systems performing economically productive tasks. As computational time is already precisely metered for billing purposes in the cloud industry, which underpins most enterprise AI services, AI-hours are one of those rare tax bases with built-in, automatically logged audit trails.
Consider a mid-size law firm in the US that automates contract review with an AI system, replacing 25 full-time paralegals earning an average of US$65,000 annually. The hourly rate for this displaced labor is US$32.50 (calculated as US$65,000 divided by 2,000 standard work hours per year). If the firm consumes 50,000 AI-hours annually in this automation, the taxable value of displacement is US$1,625,000 (50,000 hours times US$32.50). A 15 percent levy on this displacement value would generate US$243,750 in annual tax revenue. This framework taxes the economic value of human labor that AI systems replace, scales automatically based on wage levels across occupations and links the tax burden to the magnitude of labor displacement.
That might sound like a modest sum, but it would be meaningful when summed across all legal services, financial analysis and medical diagnostics. By 2028, enterprise AI infrastructure spending will reach US$200 billion globally.
True, edge computing, which processes data near their source, would complicate things, because the usage logs are local, but in those cases, the tax base would shift from data flows to proxies for capacity, such as chip specifications and energy consumption.
As regulators already use these metrics for carbon-footprint reporting and energy audits, firms would simply need to register AI-capable hardware above defined thresholds and submit quarterly usage estimates, cross-checked against capacity and power-usage telemetry.
Of course, there is a risk that firms will game the system to qualify for lower tax rates. Clearly, definitions will matter enormously. The two keywords requiring the most clarity are “augmentation” and “substitution.” Augmentation refers to cases where humans use AI, but remain the primary decisionmaking authority. Substitution occurs when an AI executes full workflows, with humans intervening only under exceptional circumstances. A radiologist reviewing AI-flagged anomalies, but making final diagnoses has been augmented. An AI system auto-approving 95 percent of loan applications is substituting for humans.
With the terms clearly defined, an AI-hours levy could be tiered at, say, 5 percent for augmentation and 15 percent for substitution. Workflow audits would determine classifications. If a firm claims augmentation, but humans meaningfully handle fewer than 40 percent of cases, the system would be reclassified. While the exact rates can be debated, the core principle mirrors the existing tax distinction between operating and capital expenditures.
Even with a 15 percent levy, AI-hours would still be dramatically cheaper than human labor.
The purpose of such a levy would be to nudge firms toward hybrid systems where human judgement improves outcomes, rather than toward fully autonomous configurations. It would make labor displacement marginally costlier than augmentation without suppressing innovation.
The industry itself should welcome such a policy. The real threat to AI progress is not taxation, but political backlash. A predictable, rules-based levy could secure a social mandate for AI by ensuring that the gains are shared.
What about the global dimension? As AI-hours become central to production, goods and services created with unpriced machine time could undercut those that are not. Fortunately, we already have a way to deal with such arbitrage. A digital border adjustment mechanism (DBAM) could require firms to disclose AI-hour content for traded goods and cross-border digital services, mirroring carbon border adjustment mechanisms already in operation. Implementation would not be difficult, because transfer-pricing rules already require multinationals to provide comparable documentation.
An AI-hours levy would have broad appeal. Businesses would get predictable rules instead of ad hoc wage politicking; labor would get a mechanism for capturing technological surplus (without blocking adoption); fiscal conservatives would get stability as payroll taxes erode; and progressives would get a tool to reduce inequality without having to resort to constitutionally dubious wealth taxes.
Coordination remains the main challenge. Early adopters might face short-term competitiveness pressures until DBAM regimes mature, but plurilateral agreements — as with carbon pricing — could create a critical mass.
Debating whether minimum wages should be US$16 or US$30 misses the point. AI systems operate continuously at negligible cost. Without a new fiscal framework, we will keep drifting toward an economy with a diminishing tax base.
An AI-hours levy can convert machine productivity into public revenue without suppressing innovation. It is the best response available.
Sami Mahroum, founder of Spark X, previously held posts at INSEAD, the OECD and Nesta.
Copyright: Project Syndicate
The White House’s decision to take a 9.9 percent stake in Intel Corp is looking like very shrewd business indeed. Since the government bought in at US$20.47 a share last August, the US chipmaker’s surging stock price has delivered the US a US$43 billion return. One of the reasons the investment has so far proved so sound is that the White House has made sure of it. According to The Wall Street Journal, Howard personally pushed deals on Intel’s behalf with some of the most lucrative clients imaginable. They include Nvidia Corp, the company at the heart of the AI
A single photograph can cut through a lot of noise, but it can also be used to misrepresent the truth. At the very least, it can concentrate the mind on something that requires further investigation. On Monday last week, Ma Ying-jeou Foundation CEO Tai Hsia-ling (戴遐齡) and former National Security Council secretary-general King Pu-tsung (金溥聰) held a news conference in which they showed a photograph of former foundation CEO Hsiao Hsu-tsen (蕭旭岑), now Chinese Nationalist Party (KMT) deputy chairman. In the image Hsiao is seated next to Xiamen Taiwan Businessmen Association chairman Han Ying-huan (韓螢煥). The two men were holding
I first met Professor Ray Jiing (井迎瑞) as a film and documentary student at Shih Hsin University’s (SHU) Department of Radio Television and Film in 1988. The following year, he went on to become the director of the Chinese Taipei Film Archive — forerunner of the Taiwan Film and Audiovisual Institute (TFAI). Over his eight-year tenure, Jiing rescued and restored over 200 classic Taiwanese films. In 1997, he established the Graduate Institute of Studies in Documentary and Film Archiving at Tainan National University of the Arts (TNNUA), and I joined the program in his third cohort of students. Beyond a
A recent report concerning a student who is suing his teacher posed the question in its headline: Does failing a student in two subjects constitute bullying? The college student in Chiayi County apparently sought NT$2 million (US$63,603) in state compensation, but a court dismissed the case. The first reaction of many might have been to ask: What has happened to students nowadays? Some say that teachers have lost their authority, while others say students are overindulged. Some even start reminiscing over the days when “whatever the teacher says goes.” However, the real issue might be overlooked if emotional reactions like that are the
RSS