Without a defensible connection to economic reality, the proposed tax on computational processing remains a thin proxy for a direct and effective AI regulation. As the United States navigates the complexities of the digital transformation in 2026, the “Make AI Work for Americans Act” has emerged as a central pillar of the legislative agenda to protect the domestic workforce. Introduced by Senator Mark Kelly, the bill attempts to capture a portion of the immense economic value created by machine learning and redistribute it to communities facing automation-related job losses. However, the tripartite tax framework—targeting computational units, advertising revenue, and excess profits—reveals a significant conceptual gap. While the intention to safeguard workers is clear, the mechanism for achieving it relies on metrics that do not fundamentally correspond to the underlying drivers of labor market disruption. This misalignment creates a fragile foundation for what is intended to be a robust social safety net.
Technical Measurement and Policy Alignment
Evaluating the Metric: Computational Processing
The most technical flaw within the current proposal involves the definition of a processing unit as ten kilobytes of binary data. In the realm of computer science, a byte represents a measure of information storage or transfer, which is distinct from the actual computational effort required to process that data. For example, a generative model might ingest a small text prompt and perform trillions of operations to produce a complex visual output, while a basic database might move gigabytes of data with minimal algorithmic intervention. By conflating data volume with computational intensity, the legislation fails to account for the reality of modern hardware execution. This distinction is vital because a tax based on data volume penalizes the movement of information rather than the intensity of the AI activity itself. Such a mismatch ensures that the tax burden is distributed based on peripheral technical characteristics rather than the actual scale of the automated intelligence being deployed.
The Correlation Gap: Hardware Workloads and Job Loss
Beyond the technical definitions, there is an absence of empirical evidence linking high-intensity computational processing directly to the specific problem of labor displacement. The legislation assumes a linear relationship where more “compute” equals a greater loss of human employment opportunities, yet this causal link remains largely theoretical in the 2026 economic landscape. In many industrial applications, massive computational power is used to optimize supply chains or improve safety protocols, which can actually support human workers rather than replace them. Without a verified correlation between hardware workloads and wage depression, the proposed levy functions more as a general excise tax on technology than a targeted intervention. This lack of specificity undermines the bill’s moral and economic justification, as it asks the tech sector to pay for social harms without demonstrating that the taxed activity is the primary source of those harms. Effective policy requires a tighter data-driven connection.
Establishing a Nexus: Policy Goals and Fiscal Realities
A fundamental requirement for any durable tax policy is a strong nexus between the base being taxed and the policy objective the revenue is intended to serve. In legal and economic theory, the most defensible taxes are those that internalize a specific negative externality, such as a carbon tax designed to mitigate environmental damage. To illustrate this, one might consider a factory being taxed on its water usage to fund local reservoir maintenance; the connection is direct and logically consistent. However, the “Make AI Work for Americans Act” operates more like a tax on the number of times a factory’s front door opens to pay for water infrastructure. While the two variables might occasionally fluctuate together, one does not cause the other. By using observable technical quantities like data packets as proxies for the abstract impact of AI on society, the bill risks creating a system that is both inefficient and legally vulnerable. The distance between the tax and the harm is simply too vast.
Assessing Revenue Streams and Legislative Strategy
The Evidentiary Burden: Justifying Technical Proxies
This attenuation between the tax and its stated purpose places a heavy evidentiary burden on policymakers to justify why a specific metric, such as a 10-kilobyte unit, should represent the socio-economic adjustment costs of AI. When a tax base is disconnected from the harm it addresses, it creates significant friction for both regulators and taxpayers. Companies are forced to track arbitrary metrics that do not reflect their business models, while the government struggles to explain why one firm pays more than another despite having the same impact on employment. The choice of a data-based unit appears to be a matter of administrative convenience rather than a reflection of true economic reality. For the legislation to be effective, it must move beyond these thin proxies and identify a measurement system that bears a more defensible connection to how automation actually alters the value of human labor. Without this refinement, the bill may face significant legal challenges regarding its fairness and logical consistency.
Revenue and Energy: Analyzing Indirect Taxation
The secondary pillars of the act, namely the digital advertising tax and the excess profits tax, introduce further layers of administrative complexity without resolving the underlying issue of specificity. The proposed 5% tax on digital advertising revenue is targeted at large platforms, operating on the assumption that these entities represent the vanguard of the AI revolution. While advertising revenue is a clear financial stream, the tax does not distinguish between revenue generated through traditional means and that which is derived from advanced AI models. Similarly, the excess profits tax component relies on a combination of profit margins and electricity usage to trigger a 50% levy. However, high energy consumption and profitability do not serve as reliable indicators of worker displacement. A highly successful firm might achieve massive returns through a breakthrough in energy-efficient hardware that expands human capabilities rather than replacing them, yet it would still be penalized under this broad framework.
Strategic Recommendations: The Path Toward Durable Regulation
Ultimately, the path toward effective AI regulation required a clearer definition of what constituted labor-displacing technology and a more sophisticated measurement of its economic footprint. Legislative efforts during the 2026 to 2028 period focused on developing transparency requirements for how AI models were integrated into specific industrial sectors. This allowed for a more granular assessment of where automation created true efficiencies and where it merely eroded the bargaining power of the workforce. Instead of universal levies on bytes, the focus shifted toward targeted transition programs funded by broad-based corporate contributions that did not rely on tenuous technical proxies. The primary takeaway was that effective taxation had to be grounded in the economic reality of the industry rather than in convenient but unrelated technical metrics. By aligning the tax base with the actual mechanisms of wealth creation, the government established a more sustainable framework for managing the workplace.
