Trade Adjustment Assistance Failed Workers—Lessons for AI Displacement

The Trump administration quietly ended Trade Adjustment Assistance in summer 2024, closing a program that had supported workers displaced by trade since 1962. The timing is notable: as artificial intelligence threatens job losses across income levels and industries, policymakers are citing TAA as a potential model for helping workers navigate AI-driven displacement.

That would be a mistake. The program's six-decade record reveals both how difficult it is to help displaced workers and how critical it is to get the response right.

Why it matters

AI could displace workers at a scale that dwarfs manufacturing's decline in the 2000s—which affected only a small subset of the labor force yet carried profound political consequences. Business leaders have warned of unemployment reaching 20 percent within five years. Unlike trade, which primarily hit manufacturing, AI could affect rideshare drivers, software engineers, accountants, and nearly every occupation. Building effective support systems takes years, making it essential to act before displacement patterns become clear.

A program designed to fail

Trade Adjustment Assistance had an impressive origin story. Proposed by the United Steelworkers and championed by President Kennedy, it was created to smooth passage of the 1962 Trade Expansion Act by addressing worker concerns about lowered tariff barriers.

But the program was hobbled from the start. Opponents worried it would help the "wrong" workers, leading to safeguards so strict that the U.S. Tariff Commission approved zero applications during the program's first seven years. Even after eligibility loosened in the 1970s, the approval process remained needlessly complex, requiring adjudicators to examine company records and survey customers to confirm imports caused layoffs. Workers often waited months or years between job loss and receiving aid—if they qualified at all.

During the China shock of 2000-2007, which eliminated up to two million manufacturing jobs, only about 130,000 workers per year were deemed eligible for TAA assistance. Fewer than 50,000 annually enrolled in training.

What actually worked

Recent research shows workers who received TAA training earned $50,000 more over the following decade than those who didn't. A 2002 wage insurance program for older workers paid for itself by accelerating their return to work. In areas where TAA reached workers, it reduced political backlash against trade and anti-incumbent sentiment.

But these benefits were sharply limited by narrow scope. A program that helps only a minority of those it's meant to serve fails its mission. At worst, TAA bred complacency, letting policymakers imagine displaced workers had support that didn't exist.

Building better systems

Any AI displacement program must avoid TAA's mistakes. Rather than requiring workers to prove exactly why they lost jobs, future programs should extend comprehensive assistance to all displaced workers. Extended unemployment leaves lasting effects on earnings, health, and well-being—programs must treat every additional day out of work as institutional failure.

Future efforts should grant "presumptive eligibility" for services before layoffs occur. If accountants or programmers face high displacement risk, retraining could become available before job loss.

Workers must also have real agency in shaping their path forward. The 1960s "automation funds" created through collective bargaining offer one model—partnerships between labor and management that supported research into training approaches, boosted severance pay, and gave workers shares of automation-generated profits.

Political leaders need to act now because effective programs take time to build and because the public rightly senses that sufficient safety nets don't exist. The fact that earlier efforts largely failed should spur policymakers to do better this time, not abandon the effort.

These details were first reported by Foreign Affairs in an analysis of worker displacement programs.