No, AI Will Not Fix Soaring Budget Deficits

Congress grasps for the next miracle cure for its irresponsibility

(Originally published at Reason)

Silicon Valley has a new pitch for Washington: Artificial intelligence will solve all your budget problems. Industry leaders claim the technology will turbocharge productivity growth, generate a flood of new tax revenue, slash the cost of delivering government services, and perhaps even bend the curve on Medicare and Medicaid spending. Deficits will shrink. The likelihood of a debt crisis will melt away, as technology accomplishes what politicians couldn’t. In the words of Elon Musk, AI and robotics are “the only thing that can solve for the debt situation.”

Politicians, desperate for any exit from their self-imposed fiscal catastrophe, are listening. The “easy answers” of simply cutting waste, defunding immigrants and foreigners, taxing the rich, or slashing defense spending are woefully insufficient to address budget deficits on track to hit 9 percent of GDP within a decade and 14 percent of GDP within three decades under current policies. Closing deficits of this magnitude would require an aggressive combination of Social Security and Medicare reforms with new broad-based taxes—unless Silicon Valley can produce an AI-based economic miracle that lets Americans continue collecting exorbitant government benefits at relatively low tax rates.

Unfortunately, like all the other “easy solutions” to budget deficits, AI is highly unlikely to produce the trillions of dollars in annual fiscal savings necessary to avert the hard decisions. Betting America’s fiscal future on AI is wishful thinking.

To analyze the prospects, we can examine the fiscal and economic categories AI is most likely to affect.

A Productivity Miracle Still Leaves a Massive Deficit

The late 1990s technology boom drove annual productivity growth roughly 1 percentage point above its long-run trend from 1995 through 2005. That extra growth generated healthy tax revenue and broadly raised living standards. If AI delivers a comparable shock—and some credible economists believe it could deliver more—the revenue windfall would be real and substantial. A sustained 1-point boost to annual productivity growth rates would raise tax revenues by $143 billion annually in 2028, rising to $834 billion (1.8 percent of GDP) annually by 2036.

While such revenues are significant, they would shave only a fifth off the $4.4 trillion projected budget deficit under current policies. As exciting as it may be to nearly double productivity growth rates, this would barely offset the economic growth declines already occurring due to labor force reductions driven by falling fertility rates, retiring baby boomers, and immigration restrictions.

Yes, some AI enthusiasts may suggest that long-term productivity growth could leap by 2 or 3 percentage points (or more)— essentially doubling or tripling its baseline annual growth rate. This would provide additional deficit reduction but still not come close to balancing the budget. Moreover, the productivity-to-wages-to-tax-revenue chain can weaken. If AI gains accrue disproportionately to capital owners rather than boosting worker wages, added income and payroll taxes may not grow as quickly. On the other hand, AI could help the IRS identify tax cheats and collect unpaid taxes.

All together, even an optimistic scenario delivers only $1 trillion annually in new net revenues by 2036.

The Hidden Price Tag of Automation

The more AI revolutionizes work and expands economic productivity, the more disruptive it will likely be to the workforce. AI is highly unlikely to permanently raise jobless rates—the economy has evolved from hunting and gathering to mass agriculture, the industrial revolution, and the information revolution without that occurring, because there are always new consumer demands to satisfy with new industries and jobs. But a full AI workforce revolution without significant and expensive job displacement costs is highly unlikely, and the adjustment costs for existing workers could be painful and expensive.

Economists have identified two potential job displacement scenarios. The first scenario, which has not borne out so far, involves significant layoffs in AI-heavy industries, bringing long spells of unemployment for mid-career professionals. The second scenario, which is beginning to play out, is a significant reduction in entry-level job openings for younger workers in selected industries. The first scenario produces perhaps 2 million to 3 million jobless workers at a given time requiring unemployment benefits, Medicaid, food assistance, and job-retraining assistance. The second scenario suggests perhaps 4 million to 5 million younger workers entering the labor force in lower-skill jobs at lower wages (or not at all) and thus reducing tax revenues and pushing up low-income benefit costs.

Depending on which scenario ultimately plays out, and to what degree, the total displacement costs could consume anywhere from 15 percent to 40 percent of all new tax revenues (or more than 100 percent if accompanied by a universal benefit of at least $3,000 annually).

Under a midpoint assumption of taking back 25 percent of the budgetary gains from productivity, worker displacement costs would total $250 billion annually by 2036.

Longer Lives, Bigger Bills: AI Could Create a Fiscal Paradox for Social Security and Healthcare

Any AI-produced improvements in worker productivity and wages will automatically expand Social Security’s spending liabilities, because initial benefit levels rise with lifetime earnings. These effects would be partially offset by job displacement effects that reduce both near-term payroll tax revenues and long-term benefit liabilities.

If these effects roughly cancel each other out, then Social Security’s tie-breaking factor may be AI-assisted advances in medicine meaningfully extending average lifespans, allowing retirees to collect benefits for more years—while simultaneously driving up Medicare costs as well. This would likely mean a modest worsening of Social Security’s long-term shortfalls that manifests in the 2030s and 2040s.

AI’s fiscal effect on healthcare spending is heavily disputed. The optimists paint a rosy picture in which technology would dramatically slash the one-quarter of national health spending currently allocated to administrative functions. New technologies would diagnose injuries and diseases earlier and treat them more quickly and effectively, extending lifespans and improving quality of life. Many of these advancements would be accomplished with at-home technologies and less time spent in doctors’ offices and hospitals. Federal Medicare and Medicaid costs, along with economy-wide health spending, would plummet in this scenario, solving the leading driver of long-term deficits.

All of these exciting outcomes are possible—but probably not the dramatic savings for the federal budget. Even if AI substantially reduces healthcare administrative costs, there is no guarantee the productivity gains would accrue to the government through lower payment rates; many past productivity gains have accrued to health providers and insurers instead. Aggressive diagnostic improvements can save treatment costs for many, but they can also encourage overdiagnosis, which brings unnecessary utilization costs for others.

Adopting new technologies requires substantial capital investment. And while automating existing procedures can lower costs, developing expensive new technologies that people want to use can drive utilization costs higher. Overall, it is not clear that AI’s efficiency savings would outweigh the costs of faster and more aggressive diagnoses plus the higher demand for expensive new technologies.

Moreover, even AI’s markedly improved health outcomes can only delay, not prevent, the inevitable cost of disease, end-of-life care, and death (with higher Social Security and Medicare spending in the meantime). Thus, it is not clear that even the rosy AI scenario described above would notably reduce per-beneficiary healthcare costs over these (likely extended) lives.

So an AI boost could lead to modest economy-wide health savings, but these would largely accrue to health providers and insurers rather than the federal government. Medicare and Medicaid spending would keep rising with an aging population and rising health care demand.

AI in the Military: More Missiles, Not More Savings

AI has the potential to streamline military logistics, intelligence and surveillance, and autonomous systems. At the same time, this would likely trigger an expensive investment surge in AI-enabled weapons systems. Historically, military “efficiency” gains have tended to get recycled into capability expansion rather than applied to deficit reduction.

These scenarios do not appear imminent in any case, as less than 1 percent of the Defense Department’s 2024 budget request went toward AI. Thus, AI-related savings in the defense budget, currently taking up 13 percent of federal spending, would be small to nonexistent.

The Computing Revolution Didn’t Notably Shrink Government. Why Would AI?

AI should streamline government administration and reduce the need for federal workers. Yet the federal personnel savings may disappoint. The federal bureaucracy’s widespread adoption of computers and internet technology in the 1980s and 1990s merely froze nondefense federal employment. That precedent, along with strong civil service job protections and current AI-based job trends, suggests that eliminating perhaps a tenth of the 2.2 million federal civilian workforce would represent an ambitious target. In this area, the federal budget might see annual savings of $30 billion from the $300 billion in salary and benefit costs for current civilian employees.

Looking at the federal government’s broader administrative expenses (which are a much smaller share of government spending than benefit costs) a fully scaled up AI-based systems may be able to shave $20 billion to $50 billion annually from the estimated $186 billion in fraud and improper payments. At the same time, deploying AI across federal agencies will require significant upfront acquisition, integration, and training costs. For roughly a decade, there would likely be no net savings as upfront investment costs swallow any efficiency savings.

CLICK HERE for the full article at Reason