Gates says the tax code pays employers to swap workers for machines

Payroll taxes attach to a flow of wages. Expensing attaches to the timing of a capital purchase. Gates calls the gap between them a nudge toward replacing people with machines ^[3], and the description holds, but it is not a loophole anyone drafted. It is what happens when a levy earmarked to labor sits beside ordinary capital cost recovery and nobody reconciles the two. Closing it means naming the thing taxed, and Gates names robots and AI tokens ^[4]. A token is a billing artifact of model vendors rather than a statutory category, and Fortune's account of the nearly 6,000-word essay carries no rate and no definition of what would qualify ^[13].

The essay then asks one instrument to do two jobs that subtract from each other. Gates wants the levy to slow the move away from human labor and to raise money for retraining and a stronger safety net ^[7]. He also wants it because income tax receipts fall when fewer people work, while demand for retraining and social security rises ^[5]. Deterrence shrinks the base. Revenue at scale means the substitution went ahead regardless. His own wording says which way he expects it to land: the tax would slow the rush "a little" ^[7]. That is a revenue measure wearing a behavioural label, put forward on the heels of news that US government borrowing reached $40 trillion ^[6] and argued on the grounds that the funds have to come from somewhere while budgets are stretched ^[11].

The 2017 objections have not been answered so much as repriced. Summers, writing before the current AI boom, called Gates seriously astray and the idea protectionism against progress ^[8]. Robert Seamans of NYU Stern made the narrower and more useful version: on the data then available, a robot tax would dissuade investment, lower growth, and, where machines complement labor, reduce hiring and wage growth ^[9]. Read the hedge. "The data we currently have" was 2017 data, and nothing in this exchange updates it. Gates does not contest the efficiency loss; he says accelerated innovation means a little inefficiency is affordable if it keeps people employed ^[10]. The disagreement has moved from whether substitution is cheaper to who should be made to pay the difference.

Pew supplies the politics rather than the evidence. Seventy-one per cent of US adults expecting fewer jobs against 5% expecting more is a ratio of roughly fourteen to one ^[1][14], seven points up on 2024 ^[15], and young adults are no calmer, with 73% expecting fewer career opportunities over 20 years ^[2]. That measures expectation, not displacement. It tells a legislature how much room it has and nothing about how many jobs moved. The harder problem sits after the money is raised: Gates says it has to reach workers who lose jobs, people whose hours or wages fall, and the communities where the losses concentrate ^[12]. No tax design does that part.