Per-token AI tax punishes cheaper models that use more tokens

The catchy “token tax” idea breaks once you look at how AI is actually priced. A literal per-token levy means charging for every chunk of text a model processes, which sounds simple but skews incentives immediately.

Dean Ball’s pushback is that this setup would punish cheaper models that use more tokens to get the same result, while pricier, more efficient models get a relative break. He argues the only coherent version is a sales-tax-style levy on total AI spending, though that still directly discourages using AI at all. Meanwhile, critics note the debate keeps ignoring the basic comparison: token costs versus paying human labor.