How not to tax AI.

Bill Gates is worried about artificial intelligence (AI). He's still a pretty influential guy, so when he writes a 5,000+ word essay on a pressing issue, it's usually worth a read. In this case, Gates homed in on what is likely to be a political hot potato in the coming years – whether AI is coming for people's jobs (so far there's no evidence for this), and what can be done about it:

"Robots and AI combined can create a vicious cycle. After one company adopts them and uses the savings to lower its prices, its competitors will feel immense pressure to do the same. If existing companies don’t adopt them, then start-ups will. Many people will shift to other jobs, but the turmoil of losing work, getting retrained, and finding other work will be significant. Market forces will make adoption go faster and faster and, unless we intervene, there will be fewer good jobs available and the benefits will accrue to a small group."

Note the use of "we". Who, exactly, is going to "intervene"? It won't be some benevolent, all-wise technocrat. It will go through the same political sausage-making process that, in Australia at least, just delivered us a bailout of an aluminium smelter. It will also be the same series of governments that built the massive black market for cigarettes, and the organised crime syndicates that service it.

Gates' own essay proves the point. The tax, he writes, must be "targeted so it does not slow down the purely beneficial uses of AI, like making medicine and education cheaper".

So there are the first two carve-outs. No doubt every business in the world will be sending their lobbyists to their respective capitals to get their own exemption, until the tax base is so narrow that the only way it'll raise any serious revenue is with a high rate, making it one of the most inefficient taxes in the entire system (a tax's deadweight loss grows with the square of the tax rate).

But even if the political process could be cast aside, Gates' preferred tax still fails:

"I believe we should tax AI tokens and robots. Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings. But if you buy a robot, you can usually write it off right away as a business expense. The tax system nudges you toward replacing people with machines."

What Gates is describing is a tax on an intermediate good, which is a big public finance faux pas, because it will cascade down the supply chain the same way tariffs on car parts become a tax on the finished car. It's like "taxing steel during the industrial revolution — a self-defeating policy that could slow the productivity growth needed to fund public priorities".

But there are two other problems with Gates' plan that might be even more damaging. The first is that it's poorly targeted. A token tax would charge every metered token, not every displaced job. The bakery's timesheet software, the hospital's screening model, and the chatbot replacing a contact centre: they all pay the same rate. If job displacement is the problem, the tax completely misses it. It would also be incredibly easy to evade; how, exactly, would you tax self-hosted open-weight models?

The second is Gates' claim that the tax system's different treatment of labour and AI amounts to a bias in favour of AI. While it's true that hiring a worker attracts payroll tax and buying AI doesn't, that's not likely to be the only reason firms substitute one for the other, and the reason matters a lot.

For example, AI might genuinely produce more per dollar, in which case the gap is the wealth creation itself and taxing it shrinks the pool Gates wants to redistribute from (a big reason why capital generally shouldn't be taxed much, if at all). Or policy might have pushed labour's price above its value, in which case the fix is to remove the push, not add a new tax on AI. Gates never distinguishes between the two: his tax takes the gains in the first case, and patches a labour-side distortion with a machine-side tax in the second.

Gates is right that, should AI drive some serious economic disruption through unprecedented productivity growth, the transition will be rough and the gains should be shared. But his proposal would tax the gains directly, thereby slowing displacement, if at all, only by dragging productivity down with it.

If there's one word I could give for how best to prepare for a large-scale productivity boost, should AI deliver, it would be: flexibility. If labour is over-taxed relative to machines, then stop over-taxing and over-regulating labour. Cut the wedge and let prices do their work.