Unions and Shorter Workweeks: Proven Tools for AI Prosperity

Postwar policies offer a blueprint for AI gains

As debates intensify over whether artificial intelligence will trigger mass unemployment, economist Dean Baker argues the solution isn't novel—it's historical. Writing for the Center for Economic and Policy Research, Baker points to the 1947-1973 period, when U.S. productivity grew at roughly 3 percent annually without causing widespread joblessness or deepening inequality. Instead, strong unions, rising minimum wages, and evolving work standards ensured gains reached most Americans.

Baker, co-founder of CEPR, remains skeptical that AI will produce the productivity surge many predict. But even if it does, he contends, the policy toolkit already exists. The challenge is political will, not economic theory.

Why it matters

Current policy discussions often treat AI-driven productivity as unprecedented, ignoring decades when similar growth rates lifted living standards broadly. If AI does accelerate output, the distribution of those gains depends entirely on institutional design—not technology itself. Without deliberate policy choices, productivity increases flow to capital owners and executives, not workers.

Unions provided bargaining power during the golden age

In the 1950s and 1960s, roughly one-third of private-sector workers belonged to unions. That collective bargaining power translated productivity improvements into higher wages and benefits across the economy. Today, union membership has collapsed to just over 6 percent of private workers—not because employees lost interest, but because laws made organizing far harder.

The 1947 Taft-Hartley Act allowed states to ban union security agreements, enabling workers to benefit from union representation without paying dues. Enforcement of labor law became toothless; companies that illegally fire union organizers face trivial penalties. Canada, with a similar economy but different labor policies, has maintained far higher unionization rates.

Baker advocates for first-contract arbitration, requiring companies to negotiate in good faith or submit to binding arbitration. This would prevent the common tactic of stalling until organizing momentum collapses.

Minimum wage once tracked productivity

From 1937 to 1968, the federal minimum wage rose roughly in step with productivity growth, reaching levels that would equal close to $28 per hour today. That floor lifted wages not just for minimum-wage workers but for those considerably higher on the income ladder. The 1968 unemployment rate remained below 4 percent, contradicting claims that higher wage floors destroy jobs.

If the minimum wage had continued tracking productivity, it would now exceed the current median wage, directly raising pay for more than half the workforce and indirectly boosting another 20 to 30 percent through spillover effects.

Shorter hours spread work more broadly

Baker emphasizes a point he considers obvious but often ignored: work hours are not fixed. The same total labor can be distributed among twice as many workers if each works half the time. The 40-hour workweek became standard in 1937, nearly 90 years ago. If AI genuinely multiplies productivity, reducing the standard workweek to 36, 32, or fewer hours would distribute employment more widely.

Mandated vacation time, standard in other wealthy nations, achieves the same goal. Fewer annual workdays mean more workers share available jobs.

Policy choices, not technology, determine outcomes

Baker argues that hand-wringing over AI's labor impact ignores straightforward solutions. Beyond strengthening worker power, he calls for ending policies that redistribute income upward: extended patent and copyright monopolies, bankruptcy laws favoring private equity, and special legal protections for dominant social media platforms.

The analysis was published by the Center for Economic and Policy Research, which first reported these details in Baker's regular "Beat the Press" series.