NYU Tax Policy Colloquium: Chye-Ching Huang (plus coauthors), How to AI-Proof the Tax System

Yesterday’s colloquium paper was by Chye-Ching Huang and seven coauthors (most at the Tax Law Center, and including my colleagues Lily Batchelder and David Kamin). Its title is How to AI-Proof the Tax System. Since it’s still a work in progress that the authors didn’t want posted yet, I won’t zoom in here on aspects of the paper that might conceivably change before it is disseminated.

An initial point the paper makes is that, even assuming AI ends up generating large profits even relative to the huge investments that are being made, we don’t yet know how the industry will shake out or where those profits would shake out. (This being a point not just about the identity of the particular companies that might end up as big winners, but also about just where in the production chain the big profits might chiefly stick.)

That said, the paper takes the view, I think correctly, that from a (non-Pigovian) tax standpoint two particular points are true. First, there is nothing particularly distinctive about AI – compared, say, to the rise of digital firms or the rise of online life generally. We simply get the usual (recent) concerns about problems such as enormously concentrated wealth creation. But second, AI, like several of the recent storms to pass through, such as the rise of digital firms, raises big concerns about tax planning and avoidability. For example, it may offer opportunities to keep taxable income in tax havens, avoiding both residence-based and source-based taxation, while also, at the owner level, avoiding realization and being transmissible dynastically tax-free.

In consequence, the paper proposes that what I’d call the “usual suspects” – base-broadening, tax preference curtailment, and so forth – are entirely adequate to the task here. Thus, it offers a list of reforms that both the authors and I would probably favor even without the rise of AI.

At this stage, the paper offers what is basically an all-star cast of longstanding income tax reform ideas, some decades-old, a few of more recent vintage – e.g., the global minimum tax, & responses to the use of ETFs given their recent growth. The idea is that adopting these proposals would help to keep more AI income in the tax base, just as they might do so for the profits of digital firms and their owners. But one’s buy-in rate for the proposals (such as eliminating basis step-up at death) is likely to vary, depending on the degree to which one is pre-sold, because the paper can’t profitably include long sections arguing for each one – even though many of the coauthors have written elsewhere on a number of the areas covered.

While I won’t run through these here, one issue I do want to mention because it involves a bone I have to pick with the conventional wisdom on which the paper draws. The paper notes that, even if one favors expensing for all or nearly all investment outlays, pairing it with interest deductibility is a mistake because it leads to a negative tax rate, rather than merely to exempting the normal return. But I would argue that the conventional wisdom fails to frame the issues optimally.

In a case where a taxpayer borrows to make an investment that it expenses, its interest deductions (giving it a tax loss) make no difference overall to the tax system, if the same amount of interest income is included by the lender at the same marginal rate as that of the borrower. (For a chain of borrowing and lending, one would need to look all the way through.) So why are interest deductions plus expensing deemed a problem?

Proponents of the conventional wisdom have a good answer to this question. Lenders often are at least effectively tax-exempt, so what you get is deduction with no offsetting (in terms of net tax liability) inclusion.

But here is an answer to that. While that is definitely a serious problem, it’s not about expensing! It’s about the use of loans and interest (as well as other counterparty payments such as royalties) to create tax rate arbitrages that strip profits out of the tax system. That problem likewise exists under economic depreciation. So the thing to do is address interest deductibility in light of the tax rate arbitrage.

Here I would now offer The Solution, if there were one that I distinctively knew. In practice, it’s most common to deny interest deductibility in specified circumstances where it is feared to be a problem. The theoretically cleaner solution might be requiring a match between the two sides (looking all the way through), but this is presumably unfeasible.