OPINIÃO. Tributação de robôs, AI e dados: existe um caminho?

Bill Gates recently reiterated his idea of taxing robots, which seemed eccentric nearly a decade ago. As artificial intelligence advances in both intellectual and physical tasks, Gates now proposes taxing AI usage, including through tokens used by generative models to process information. This proposal is far from isolated, as economists and tax scholars have long debated how systems built around work, income, consumption, and property should adapt to an economy where algorithms, data, and computational capacity increasingly generate value.

Gates focuses on employment, noting that a company hiring workers in the U.S. or Brazil pays taxes on wages, while purchasing AI-operated machinery that replaces labor can eliminate payroll taxes and potentially deduct investments, reducing corporate tax. Studies suggest that current tax designs often favor capital over labor, potentially leading to excessive automation that may outpace economically efficient levels.

Automation can boost productivity, lower costs, and create new activities, yet it can also displace workers and reduce labor's share in income.

Thus, the question is whether the tax system should remain neutral concerning capital versus labor choices or if treating these bases differently might artificially stimulate or discourage job replacement by machines. However, equating a new tax to a simple solution is misleading. Defining a robot for tax purposes, or why a token would quantify economic capacity, presents challenges.

Models delivering similar utility may consume varying processing units; the same token amount could replace an entire department or assist a doctor in interpreting an exam. An easily countable base isn't necessarily a fair tax base.

The discussion intensifies when considering data, a primary value source in the digital economy alongside software, know-how, algorithms, and other intangibles. Many seemingly free services involve a trade: users receive content or AI tools, while platforms receive data to aggregate, process, and monetize. The question is how to capture this value: through platform profits, user consumption income, or large companies' assets.

Each approach faces specific obstacles. Taxing income finds it difficult to identify and measure data-generated value; a domain public information, valuable in isolation, may acquire immense worth when combined and processed at scale. Its acquisition cost doesn't reveal future income potential, and often, no comparable market exists.

Determining how much profit results from data and to which jurisdiction to attribute it proves challenging due to unrecognized intangible assets in accounting. Consumption taxation offers a potentially simpler path, as it doesn't require identifying each data contribution to profit formation. In Brazil, the IBS and CBS were designed with a broad base to capture immaterial goods and services transactions, sometimes involving non-monetary consideration.

However, measurement difficulties resurface. When a user receives free AI access and provides data in return, what's the value of this consideration? Is it the service received's value, the data provided's value, or does the market value exist for information becoming useful only when combined with platform technology and database?

Property taxation faces an even more fundamental obstacle. Identifying the taxable asset, who controls it, its location, and its value at a given moment would be as difficult as with traditional intangibles. These questions are even more complex given reproducible, non-exclusive, and sometimes publicly available data. Historically, the modern corporate income tax emerged in the early 20th century U.S. to address the concentration of wealth and power among newly emerged corporations like railways, oil, and steel industries.

Today, Big Tech evokes similar concerns. Proposed solutions include taxing data, processing, or tokens; broadening consumption taxation; revising territorial division of taxable profits; or correcting automation and AI externalities. Defining data or tokens in relation to income, wealth, or contributive capacity, however, proves far more difficult.

Addressing these difficulties is necessary for fair taxation, yet it's essential to recognize that the digital economy is already partially within current tax systems. A new tax may correct distortions but can also add to existing ones, raising costs and, when passed to consumers, increase regressivity. As history suggests, the tax system adapts to new wealth generation forms. The real path lies in doing so without transforming taxation into an innovation penalty.