The full article is available on The Information.
Major technological revolutions should force a renegotiation of society’s basic economic contract. AI will significantly diminish the requirement for labor at any given level of gross domestic product, and if we aren’t honest about that, we’ll stumble into a chaotic and suboptimal transition rather than engineer the optimal one. It’s not surprising that AI is scary to most people surveyed. We need to create a strong social safety net, strong core government services and a focus on preserving the human element in the emerging world if American citizens are to accept AI. But we must protect people and transitions, not necessarily existing jobs. Human value need not depend on outperforming machines at intelligence. Meanwhile, we must maintain a strong safety and security posture while holding on to a lead against our AI nation-state adversaries.
I’m a technology optimist. I’ve spent four decades devoting myself to and studying disruptive innovation, from the microprocessor, the internet and mobile phones to LLMs. On this point, I’m as certain as I’ve been about anything: AI will do at least 80% of the economically valuable work humans do today in 80% of jobs—in other words, AI will do 64% of all human labor. And that shift will happen faster than most people in Washington believe. The question isn’t whether mass underemployment arrives by 2035 or 2040—I’d put my guesstimate of 30% to 50% underemployment or unemployment with very wide error bars—but whether we have a coherent policy framework ready when it does and a strong safety net for those impacted. And yes, in this world, production of goods and services will grow fast, and many alternative better jobs as speculated below will be created. The number 30% to 50% is unimportant. What is important is preparing for a highly variable, dynamic scenario.
Right now we don’t have a plan.
The current U.S. tax code was designed for a world where labor income gets a fair share of economic value creation and capital needs encouragement. In that world, preferential treatment for capital gains made some sense: Capital formation drives growth, and that creates employment. But roughly $15 trillion of U.S. GDP in 2024 came from labor. When AI and robotics compress that figure between 2035 to 2040 by a significant percentage, and juice capital returns and reduce labor income share, that necessitates changes to the tax code from favoring capital to favoring labor. Otherwise, we will end up taxing at higher rates the thing we most want to boost, human labor, while subsidizing capital. AI systematically favors capital over labor. We need to not disadvantage labor over capital when AI is likely to cause substantial unemployment or underemployment and boost returns to capital.
And in the world of AI and the inequality it might cause, we need to sacrifice some capitalist efficiency or additional capital investment to reduce inequality and political discontent. Today’s capitalism isn’t level-playing-field capitalism with its lower tax rates for capital than labor and hundreds of tax breaks for special interests. We want a level playing field and a fair but government-deficit-neutral policy over 10 years. Some material share of the $15 trillion traditional labor income component will be displaced but replaced or complemented by microentrepreneurship, reduction in the cost of essential services and faster growth in GDP, leading to great abundance. We need sufficient funds to manage the transitions in a deficit-neutral way and to protect people that are displaced.
The right response requires radical simplicity and something close to political achievability, if anyone has the courage to propose it. I’d suggest tax changes should be deficit-neutral tax-and-transfer packages to be politically palatable. I contend the changes, to be feasible, should have the support of 100-plus million U.S. voters given that roughly 77 million voters elected Donald Trump in 2024.
Post-2028 election, the U.S. should eliminate preferential capital gains treatment in a deficit-neutral way, by taxing it as ordinary income but avoiding the usual charges levied at Democrats of tax and spend. No principled argument survives in a world of AI-boosted capital returns for special treatment for capital gains. Further special tax breaks like tax loss carry-forwards, pass-through deductions, stepped-up bases for inheritance, accelerated depreciation and business income tax gaps compound the preference. Simplifying the tax code and making it more fair makes sense, closing the many loopholes, including the really crazy estate planning only very rich people use. Of course, there are trillions to be generated if one addresses inheritance taxes above a high limit, affecting mostly rich kids, but that isn’t necessary here. There are plenty of special interest goodies inserted by lobbyists and campaign contributions, protecting asset holders who need no protection and distorting true capitalism. Critics will invoke corporate-plus-shareholder taxation, inflationary gains, realization lock-in and risk-taking as justification to favor capital gains, but these arguments no longer justify existing preferences given the magnitude of the AI balance tilt toward capital. This proposal is not unprecedented: A 1986 reform removed the 60% exclusion for long-term capital gains given preferential treatment. To be clear, most of my income is capital gains, and I would be significantly impacted by this change.
Capitalism is by permission of democracy, and I prefer that capitalism not be revoked by the socialist tendencies emerging in our society among voters. Reasonable capital gains carve-outs will be necessary, like gains on a primary residence up to a sensible threshold, and farm gains on legitimate family-owned working agricultural land, both important voting constituencies. This can generate approximately $400 billion per year in tax receipts, roughly $300 billion in capital gains tax increases and $100 billion in receipts from eliminating special tax breaks. Realizations matter: Investors can delay selling, change the timing of gains or use other tax-planning strategies, so policies to circumvent these strategies will be needed. The greater the deviation from deficit-neutral transfers needed in either direction, the more or less radical we will need to be in the changes we institute. We should through policy affect normal investment returns and genuine costs last and preferential treatment of extraordinary gains, inherited appreciation and economic rents first and get increasingly more radical based on need.
These collections should flow first to those whose employment is impacted by AI, compensating a material share of their lost income, with how much depending upon collections. Any remainder should rebate the federal income tax paid by the roughly 75 million Americans earning under $75,000 and paying about $75 billion (around the number who voted for Trump). Around 40% of capital gains is paid by those earning over $10 million a year, so there is very little sympathy for them getting preferential tax rates. With some tweaking, these two moves, plus the elimination of rich people’s deductions, together can be made deficit neutral. Equalizing the treatment of capital gains and eliminating special goodies generates roughly enough revenue, based on first-order calculations, to allow us to write checks to those workers impacted by AI and those earning less than $75,000, the most needy earners, when fear of AI is a major issue. This isn’t redistribution. It will rebalance a tax code that was already redistributing, just upward. For comparison, extended benefits during the mass unemployment of the pandemic were estimated to cost about $250 billion per year. And this is the voter constituency we need to get behind this policy.
In 2030 and beyond, the U.S. should consider adding token taxes on AI compute and robotic labor substitution revenue of 20% if needed—and if, and only if, AI’s negative impact is significant. I’m generally not a fan of innovation taxes. But if AI systematically displaces the payroll base, a carefully designed compute-usage levy becomes a logical corrective. The exact nature of this token tax, be it on internal or external models, compute silicon or by other methods, will become clearer by 2030 and should be enacted only if displacement becomes significant. Given the past three-year growth rates of 200% to 900% annually in AI model revenue, token taxes according to ChatGPT estimates at 20% could be a trillion-plus in rapid-adoption scenarios and hundreds of billions in slow-growth (and less impactful to labor) scenarios.
Applied to all sources of AI, token taxes will raise prices of AI services but increase acceptability as a trade-off. The proceeds should flow first to unemployment and underemployment relief. If AI is displacing the payroll tax base, the least we can do is tax the displacement factors and return the proceeds to those bearing its impact. Later, if unemployment and underemployment expand and AI impact and profits grow, the token taxes can be more broadly distributed. This adds an estimated $100 billion to $300 billion in token taxes at a 20% rate by 2030 and much more by 2035 to the AI remediation kitty. Current investment in AI data centers—which I’d guesstimate to be about $4 trillion to $8 trillion by 2035, though current announced plans say it could be larger—is an indication of the level of inference revenue investors expect. For every dollar invested in data center capital, we can expect greater than 25% of the dollars as a 20% token tax when generating revenue, if we estimate data center revenues as four to eight times the invested capital over five years. Calculations have to be necessarily vague, given the ambiguity here. If the AI token revenues are less than is likely, the remediation revenues needed for job displacement will also be less, and the inverse will be true if revenues are larger. And if AI is accepted by 100 million voters, U.S. data center investments become safer.
In 2030 and beyond, we should make basic needs services based on nearly free expertise for everyone. AI will do this with or without government—the question is whether the government accelerates it or not. I’d suggest open procurement, measurable outcomes, provider competition and conflict-of-interest safeguards for AI governmental services. A federal initiative covering 24/7 virtual primary care and multispecialty care, mental health services and chronic disease management, personalized AI tutors for K-12 and college education, AI-based access to government services and legal aid could cost just a few dollars per citizen per month in token costs by 2030, delivering better than current services at a fraction of existing spending in these areas. These basic health and education services should be nearly free to everyone. When expert knowledge becomes nearly free, the wealth advantage of affording doctors, tutors and lawyers collapses—no redistribution scheme has ever achieved this kind of equalization. Providing these AI-based services nearly free will save far more in downstream healthcare and education costs and will likely have negative total costs. I would estimate a deflationary economy by 2035, reducing costs of living further.
A bold presidential candidate for 2028 could adopt this deficit-neutral platform: Accelerate AI and U.S. AI competitiveness, help most Americans and energize a frightened voter base with a strong safety net. Sometime in the 2030–35 timeframe, we could be approaching a kitty of a trillion dollars a year to lessen AI’s impact and additionally have very low-cost basic services to improve affordability. And making AI adoption for the people’s benefit in the form of accessible AI-based healthcare, AI-based education, AI-based transportation and more a reality will make AI more widely accepted, instead of a fight against socialists who will try to push back against AI restructuring with employment protection schemes that will cause America to lose the AI race and a chance at a utopic level of services for citizens.
AI will create enormous corporate wealth. The question is who will own it. A sovereign fund with ownership of AI companies makes every American a capital owner, not a bystander to the AI economy. But a sovereign fund decision can be made based on AI’s impact over time.
Many, especially economists, will quibble with specific estimates, but the arguments are directional, not dependent on incremental precision and not fundamental in the era of dramatic change that is likely. I am advocating for a coherent initial policy framework to get started. If we accept that these are not fixed one-time policies but ones that will need adjustment, maybe even major adjustment, in 2028, 2030, 2035 and beyond as our vision and estimates get less fuzzy, then we must not wait until it’s is too late or until we get a socialist government out of fear of unfettered capitalism and AI. Again, capitalism itself is by permission of democracy and may be at risk. I am advocating private actors to retain control of business and services while the government manages the disruption as needed.
I get asked: “What will people do?” On employment creation, AI will make it easy for every citizen to become a microentrepreneur. I can see tens of millions of American microentrepreneurs reemployed by 2035, not needing to know finance, legal, taxes or any other specialty, fully supported by AI to do whatever their unique contribution or skill is, be it wood carving, research, writing, sports both team and individual, music, painting, pottery, chefs, jewelry, custom furniture, physical art, collectibles, entertainment, custom fashion, childcare, children’s education, dog walking, baking, ceramics, travel guides, stylists, gardening, live performers, personal performances, co-creation with consumers and other human-centric endeavors. What will be valued? Human provenance: This particular person made or performed it. Physical embodiment: It was cooked, grown, fabricated or delivered in the real world. Live presence: The consumer interacted with an actual person. Taste and reputation: The creator acts as a trusted filter over infinite possibilities. Community: Buying the product provides membership and relationships, not just an artifact. Think a next-generation remix of a combination of Etsy, Shopify, Patreon, Roblox, Substack and an on-demand microfactory, but designed around one- or few-person consumer businesses, and exactly the opposite of cubicles in the Fortune 500, assembly-line labor, farmworkers in fields, what I call “servitude jobs.”
AI tools can make every American their own boss by making microentrepreneurship very easy for most people with particular skills. Additionally, eldercare and childcare will be massive consumers of human labor and preferred by humans, given current trends in global demographics. Humans will provide the human element of care to all humans, especially families. The elitism of intelligence will be valued less and human preferences much more so. The human touch will be valued over pure utility and supplied by microentrepreneurs. The human aspect, provenance rich, sometimes artisanal, will be part of the value of products and services, in ways not displaceable by AI but assisted by it. As AI provides basic needs and humans are free to pursue their desires, instead of servitude to survival labor, human labor and endeavor will be valued and human preference will be for human-made objects and services allowing for more microentrepreneurship opportunities. Status, social capital, reputation, competition, caring and lust for mastery will be what humans strive for.
Half of today’s jobs are in my view “servitude to survival,” a new form of voluntary slavery to just make an income. Farmworkers toiling in heat, bent over, 8 hours a day for 30 years, and assembly-line jobs mounting tires on a car for 8 hours a day for 40 years or until your back gives out, are the opposite of human dignity. The indignity in often coercive and monotonous conditions, injury, insecurity and lack of choice will decline, and humans will be freer to do what they want, not just survive. These servitude to survival jobs are a very large percentage of U.S. jobs today. Traditional economic frameworks like GDP metrics will be distorted in a deflationary economy, with many free services unaccounted for and with an increase in products and services without personal income growth. There will be much more equality, much higher minimum standards for every U.S. resident, basic assured services possible through governmental services, along with pockets of extreme wealth.
Corporations will still provide infrastructure and core products, but a much larger portion of consumer value than today could be created by identifiable people operating businesses with one to five workers. And for those who want them, there will be some cubicle jobs.
In conclusion, this requires manageable changes to current institutions or bureaucracies. Replacing a tax code written for the 1950s with one designed for the 2030s refounding of America gets a sufficient number of self-interested voters behind these initiatives. Capitalism operates by permission of democracy, and democracy’s voters are the 100-plus million people who would benefit if they need help from AI’s nearly free services, declining costs and deflationary economy and who are protected by funds in this kitty from income dislocations. Not all prices will be deflationary: Housing and possibly food may start to consume a larger share of people’s income and must be addressed, especially for the lower half of income earners. Again, this approach to capital gains should be deficit neutral to be more politically palatable. That arithmetic isn’t a coincidence. Enough voters is what makes this politically achievable. Transitions, especially rapid ones, are difficult, but we will have the resources to help those adversely affected. Preserving public support for AI requires giving people security and a meaningful stake in its benefits. The alternative is to do nothing and watch a democracy with huge structural underemployment or unemployment decide that capitalism itself is the problem. History doesn’t grade that outcome well. The future utopia is ours to invent. The dystopia is ours to choose.