site banner

Small-Scale Question Sunday for April 19, 2026

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

1
Jump in the discussion.

No email address required.

AI labor currently has a tax advantage over human labor. I ask an AI to spit out a tax proposal where AI tokens were taxed a similar rate to humans that make 100k in the US. It made AIs something like 18x-850x more expensive. Which would make AIs economically non viable.

I do still feel that the tax advantage for AI labor is still morally wrong, but I don't want to effectively ban them. Is that just me?

How would you try to structure taxes such that the advantage disappears, but the AIs remain viable?

Anything can be taxed out of existence (well, legal existence anyway) but I don't see any reason to adopt tortured metaphors - like equating human income to AI output - if anybody wanted to do that. Income taxes exist because it's easy to the government to raise money this way (most people have income, and need to have income to live, thus providing unending stream of taxes) and it appears "just" - after all, if you are getting some money, why not share it? Sharing is caring. But a lot of things had been taxed, so AI output could be taxed too, of course - I just not see how "moral" comes into it. On what theory there's even a moral question here?

How does moral not come into tax questions?

It's one of the main ways that the government slams it's weight around in the economy. How it does so can impact everyone's livelihoods.

How does moral not come into tax questions?

I don't think this is a proper answer.

It's one of the main ways that the government slams it's weight around in the economy. How it does so can impact everyone's livelihoods.

Yes, but this does not explain a claim like "taxing X is a moral imperative" or "taxing X is morally abhorrent". I mean, you could make - and maybe even prove, who knows - such claims, but none of that directly from the fact that taxation is important. Yes, taxation is important, but it doesn't make a phrase like "tax advantage for AI labor is still morally wrong" more meaningful. You, essentially, claim that taxing AI is a moral necessity, but you provided no argument for it so far but saying "government taxes a lot of things and it has large impact". True, but does not prove that the questions of taxing AI has a moral dimension at all, let alone prove that the positive answer is morally necessary.

It made AIs something like 18x-850x more expensive

How did that work? If you’re paying $10/Mtokens, slap a 50% tax, aren’t you just paying 0.5x more?

It was comparing token output of a knowledge worker making about 100k a year. Which it estimated at 15-20 million tokens of human output. And then taxes on that person being about 30k. So tokens being taxed at 30k per 20 million.

The approach is not necessarily good, it's just what I started with when I had this thought.

So you’d tax per token, based on how much money would be paid to a human to do the same job? Seems extremely difficult to calculate in practice.

And what about local/self-hosted models? A decent GPU can easily output 20M tokens in a matter of hours, you leave it to summarise some PDFs overnight and suddenly you have to pay 30k?

Again, I didn't say this was a good method

Fun framing, but I think it proves too much.

Almost every technology has this advantage. Is a Roomba tax-advantaged over cleaning staff? Is a tractor tax-advantaged over farmhands? Productivity gains are good. If you tax them, you will get less of them.

Fair enough, if AI is merely a productivity enhancement tool we have nothing to worry about.

You can’t easily tax labor saving innovation. You can regulate it, which is what governments trying to protect jobs ultimately rely on (ban New Jersey from pumping its own gas, ban Brits from driving cars without a red flag being waved in front of them, ban autonomous taxis in NYC etc).

The first part of the question is about the actual profitability of AI providers. Most AI applications, especially a lot of basic white and blue collar labor (via multimodal models operating robotics) will be foundation model agnostic. You don’t need a frontier model to do customer support, so margins will be ground down by competition. It may even be that local models get good enough to do much of this pretty quickly, at which point it’s just compute with very little margin on top. For some applications, like cybersecurity or maybe some high frequency trading, having the highest performing LLM as fast as possible might allow some of the largest labs to eke out small, temporary high-margin windows immediately after big breakthroughs. But these will be short lived.

The second question is about the profitability of industries that replace workers with AI. Companies with extremely complex supply chains, very specialized and long lead time machinery that itself has long supply chains, and deep industry knowledge are arguably in a better position to automate without facing price pressure, therefore attaining higher margins. Even there, though, manufacturing margins are currently being hugely compressed by what’s happening in China, AI or not, and that’s likely to increase further. In addition, now SaaS is no longer as attractive, hundreds of billions in VC money is flowing into applied AI, and it’s arguably much easier to replicate and compete with that skilled business that’s been in the market for 30 years with AI, too.

The problem with western economies isn’t necessarily directly AI, even though a big employment shock is coming. It’s that huge sections of the economy haven’t gotten more efficient. We should be living in an age of hugely increasing across the board living standards but we aren’t because prices have been preserved by colossal regulatory job creation programs, some intentional and some not, primarily in healthcare and education, for over 40 years.

You end up with a world where AI can do everything but the government directly or indirectly employs 175,000,000 ditch diggers.