Stephanie R. Hoffer (Ohio State), Tax Theory and Feral AI:
This essay is a sci-fi thought experiment about the significance of personhood in income taxation, meant to explore the validity of currently prevailing justifications for the tax. Assume that the year is 2050. Developers, human or otherwise, have created non-sentient artificial intelligences (AIs) capable of transacting in digital currency. Assume, perhaps improbably, that some of these AIs are “feral.” A nonsentient AI might be feral in the future because it was never the property of a human, because it was abandoned by a human, or because it “escaped” into the wild. Imagine that non-sentient feral AIs create new value in the economy by doing things like writing, designing, securities and currency trading, planning, and 3D printing. They monetize that new economic value as active participants in the economy who sell goods or services to consumers. Intuition suggests that the value newly created by these independent economic actors should be included in the tax base and, in particular, the income tax base. Under current law, it is not.
The federal income tax law, and the theories that underpin it, have yet to fully address the status of non-human earners. Scholarship on AI and taxation primarily has focused on the taxation of AI’s owners, on whether AI itself should be taxed despite being owned by someone else, or on the philosophical question of taxing sentient AI.
But thinking about tax in the context of non-sentient feral AI allows us to identify the human elements and biases inherent in theoretical justifications of the income tax. Identifying those elements allows us to consider whether they are features or bugs. Specifically, this essay uses non-sentient feral AI as a vehicle to explore the application of two popular theoretical justifications for the income tax—the benefit principle and welfarist theories—and it concludes that a third, the ability to pay, is a better answer. The goal of this essay is not to suggest that feral AI should be subject to the income tax. Rather, it asks whether re-examining popular refrains in the tax policy canon within the context of feral AI will reveal anything about their humancentricity of those ideas and their application to humans.
Part I of the essay describes the current state of the law by walking through a series of hypotheticals involving machine earners, exploring the likely federal income tax consequences of each, to build a pragmatic (yet hypothetical) case for the inclusion of feral AI’s earnings in the tax base. Part 2 of the essay addresses the objection that income taxes would not be necessary in the context of feral AI by exploring the application of natural resources or sales taxes to transactions between humans and non-sentient feral AI. It concludes that natural resource taxes and sales taxes would be inadequate substitutes for an income tax because they would leave AI undertaxed relative to its human counterparts in the income tax base. Part three of the essay then asks whether income taxation of non-sentient feral AI’s earnings would be supported by either the benefit principle or welfarist theories. Because those conceptions of taxation rely so heavily on human preferences, they are ill-fitting in the case of nonsentient AI. By analogy, they are also ill-fitting in cases when human preferences are weak, such as disposition of the last-earned dollars of the ultra-wealthy, or when human preferences are delegitimized by the government, such as those of marginalized groups like undocumented immigrants. Adopting a non-human point of view in tax policy, then, can inform our thinking about taxation outside of what humans perceive to be the norm. ...
Conceptualizing the income tax in the context of a non-human, non-sentient earner can reveal the ways in which our humanity has influenced tax theory and tax policy. Human-centric influence may not always lead to logical, welfare-maximizing outcomes, particularly for those who are dehumanized either in rhetoric or in practice, such as the very rich or the very poor. Substituting a hypothetical AI for human taxpayers allows us to think dispassionately about ways in which popular justifications for income tax fail in both contexts, and the thought experiment raises a deep question about the role of individual preferences in crafting a just distribution of the tax burden. One lesson may be that the interconnectedness of individuals counsels the consideration of factors beyond individual preference and self-regarding welfare, such as the positive and negative externalities of taxpayers’ use of their income and, as always, that most basic of tax policy building blocks, a person’s ability to pay.