The Hidden Costs of Taxing AI: Why the Obvious Fix Might Backfire

The conversation between Alex Bores and tax expert Martha Gimbel points to a tension most AI policy debates miss: the same tax that targets AI-driven inequality could inadvertently slow cancer research while barely touching the call center it aims to protect. This isn't just a tax policy debate. It's a problem that requires thinking about the whole system, where the obvious solution creates unexpected ripple effects. Anyone designing AI policy, investing in AI-adjacent industries, or worried about their job's future should read this to understand why the first-order solution rarely solves the second-order problem.

Why Taxing AI Tokens Punishes the Wrong People

The conventional wisdom goes like this: AI is concentrating wealth at the top while workers get left behind. So tax AI, redistribute the proceeds, and problem solved. Alex Bores, a Democratic congressional candidate and computer scientist, proposes exactly this: a token tax on commercial AI usage, paired with reduced tax deductions for AI investments, funding an "AI dividend" for Americans.

But here's where systems thinking reveals the crack in the logic. Martha Gimbel, executive director of Yale's Budget Lab, points out that a token tax taxes business inputs, not outputs. And that distinction matters enormously when you map the full causal chain.

Consider two hypothetical companies. A call center replaces 100 workers with an AI chatbot. Because most customer calls follow predictable patterns, the chatbot doesn't need many tokens to handle them. The token tax generates minimal revenue from this company, even though it just eliminated a hundred jobs.

Now consider a biochemist using AI to model protein interactions for cancer research. This requires massive token consumption, generating and analyzing millions of molecular combinations. The token tax hits this company hard, even though it's creating value, not replacing workers.

"If you're taxing per token and you have a scientist in a lab trying to cure cancer and they're using a ton of tokens, they then bear the cost of the token tax, whereas the call center that doesn't require as many tokens to replace its workers doesn't."

-- Martha Gimbel

The system responds perversely. The tax targets the wrong behavior (heavy AI usage) when the actual problem is labor displacement. And because it taxes inputs rather than outputs, it distorts business decisions in ways economists generally dislike. (Think steel tariffs, but for neural networks.)

The 18-Month Payoff Nobody Wants to Wait For

Bores's argument for why AI needs special treatment is worth sitting with. He claims AI is fundamentally different from every previous technology:

"AI is the first technology ever developed where the makers of the technology are explicitly saying that they're trying to replace all human... every other technology before was by default a complement to human beings, not a substitute."

-- Alex Bores

This is a strong claim, and it drives his urgency. But Gimbel's counterpoint reveals a different timescale problem. She points to census data showing 96% of businesses haven't changed hiring because of AI, and 2% have actually increased hiring. The immediate crisis Bores fears hasn't materialized.

The implication is uncomfortable: we might be designing policy for a problem that exists mostly in projections. And poorly designed policy, once enacted, creates its own downstream effects: distorted investment, slowed innovation, competitive disadvantage against countries like China that Bores acknowledges aren't taxing AI.

Bores's response to the China question is revealing: "I don't primarily think so, but I also think that we shouldn't be incentivizing replacing humans with AI." He's prioritizing domestic labor protection over international competition. That's a defensible position, but it requires accepting the trade-off explicitly.

Where the Real Leverage Lives

Gimbel's alternative is less flashy but more systemic. Instead of designing a new tax for a specific technology, she argues for fixing existing policy gaps that would matter regardless of AI's trajectory. A reformed corporate tax system. Better capital gains treatment. A more generous unemployment insurance system that helps workers displaced by any cause, not just AI.

This is the systems thinker's move: address the underlying structure rather than the surface symptom. The current tax code already incentivizes replacing humans with AI. Bores himself notes that "we put huge taxes on hiring a human... and we put huge discounts on using AI." That's a policy choice embedded in existing law, not something AI created.

The question becomes: do you design a targeted intervention that might miss its mark (token tax) or fix the broader system that's creating the distortion in the first place (corporate tax reform)? One feels more responsive. The other is more durable. The tension between them is the entire debate in miniature.

Key Action Items

• Over the next quarter: Audit your organization's tax incentives around automation. Are you being subsidized to replace workers? Understanding your current position is prerequisite to evaluating policy changes.
• Over the next 6-12 months: If you're in a token-heavy industry (biotech, research, content generation), model what a token tax would cost your operations. The burden won't fall evenly across sectors.
• Immediate: Watch for state-level AI tax proposals, not just federal. New York is already considering token taxes, and state policy often precedes national.
• 12-18 month horizon: Invest in understanding the difference between AI as complement vs. substitute in your specific domain. The policy response will depend on which category your use case falls into.
• Ongoing: Push for unemployment insurance reform as a hedge against any automation-driven displacement. This is the policy that works regardless of which technology causes the disruption.
• Immediate discomfort for long-term advantage: If you're a policymaker, resist the urge to design AI-specific taxes until you can map the full causal chain. The token tax sounds targeted but creates perverse incentives that compound over time.
• Over the next 2-3 years: Track the census data on AI-driven hiring changes. If the 96% figure shifts dramatically, the policy window opens. Be ready with proposals that address root causes, not symptoms.