October 15, 2020

An initial policy concern from rapid automation was that if robots continue to substitute for human workers, then a fiscal policy crisis may result as tax revenues decline during a period of rapid automation. That first problem arises because the tax system has been intentionally designed not to tax capital assets, such as robots, or at least not to tax them to the same degree as human labor. A second problem also exists: advanced artificial intelligence (“AI”) may soon have the ability to engage in factual structuring as a means of direct tax avoidance. This direct tax avoidance planning by advanced AIs could further erode tax receipts because an advanced AI also has the potential to formulate its own version of “social norms” in respect of tax compliance. Furthermore, an alternative method to tax ideology to formulate tax policy may also arise from AI referred to here as “tax actualing,” where an advanced AI with a sufficient set of data in respect of cash flows through the economy uses data to make accurate predictions and to thereby supersede current methods of economic modeling. Various critiques of proposals for robot taxation are also addressed here including supposed: (1) productivity losses on taxing robots, (2) additional complexity inherent to all of the robot tax proposals, (3) difficulty in identification of “robots” as capital, and (4) inability to capture benefits from capital assets. Finally, an advanced AI is likely to prefer a tax system which maintains its ability to obtain tax deductions for incremental capital investment. Since higher income tax rates are strongly associated with rapid economic growth in nearly all human societies—past, present, and by all indications, future—it is likely that artificial intelligences will voluntarily choose to assess income taxes upon themselves at high rates as a means to encourage capital re-investment. 

Author:  Bret N. Bogenschneider, PhD, JD, LLM 

PDF: http://ncjolt.org/wp-content/uploads/sites/4/2020/10/Vol.-22-Issue-1_Bogenschneider_Final_1-56-1.pdf

Volume 22, Issue 1