Foreword:

On September 24, 2026, Harvard Kennedy School hosted a pivotal dialogue. Peter McElroy, Chief Economist at Anthropic, warned against AI computing power monopolies ("Coase Singularity") and proposed a 'Token Tax' (akin to carbon taxes) on excessive automation to balance capital and labor distribution.

Author | Fang Wensan

Image Source | Internet

The Vanishing Unemployment Wave: Why Are Macroeconomic Data Indifferent?

The biggest mystery in today's AI economy is the exponential growth in AI capabilities with almost no macroeconomic data response.

The U.S. unemployment rate remains stable at a historic low of 4.1%, with no significant increase in total factor productivity, sparking intense market attention. McElroy's core explanation is the 'J-Curve Effect.' He likens it to an expensive experimental phase for companies: while model capabilities exist, firms have yet to organize and feed internal contextual information to AI. 'Capabilities plus contextual information—both are indispensable.' Large enterprises, constrained by data firewalls and organizational structures, will see limited AI impact unless they invest in data reorganization and contextual provision. Companies are now at the bottom of the J-curve—investments are rising, but returns have yet to materialize.

Employment data points in the same direction. McElroy emphasizes, 'This appears more like a skill-biased labor augmentation model rather than a technology that fully automates and massively replaces jobs.' According to U.S. Labor Department data, no occupation has had all its tasks systematically automated by AI. Based on analysis of 100,000 real conversations, Anthropic estimates current models can accelerate individual tasks by about 80%. Extrapolated over the next decade, U.S. labor productivity growth could rise by 1.8 percentage points annually. If these efficiency gains spread across the economy, U.S. macroeconomic growth could return to the high-growth era of the late 1990s to early 2000s.

The U.S. Congressional Budget Office's long-term growth forecast is 1.7%, with the Federal Reserve's most optimistic estimate around 2.4%. This 1.8-percentage-point increase implies existing monetary and fiscal frameworks may need reassessment—in other words, policymakers may have systematically underestimated AI's growth engine.

Moderate Scenario: AI as an Upgraded Office Suite, Ensuring Social Stability

Anthropic's economic team, collaborating with MIT and Stanford scholars, used a 'task-based labor model' to dissect U.S. Labor Department jobs into thousands of micro-tasks, assessing which could be enhanced, replaced, or unaffected by AI. Based on this, they outlined three pathways to 2030.

The first pathway, the 'Moderate Scenario,' predicts AI will be applied in about 4% of economic activities by 2030, contributing an additional 1.6% to U.S. GDP, with the economy moderately rising to $34.1 trillion. Unemployment would edge up by just 0.1%, labor income remaining at around 60% of national total income, and only 0.3% of knowledge workers needing to switch careers. In this world, AI resembles an upgraded Office suite—everyone uses it, but no one loses their job.

Significant Scenario: White-Collar Stagnation, Blue-Collar Resurgence in a 'Mass Replacement'

The second pathway, the 'Significant Scenario,' envisions AI systems independently handling nearly half of human knowledge work, from basic coding and legal contract reviews to financial audits, junior copywriting, and data analysis. By 2030, AI would contribute an additional 8.3% to GDP, with the economy surging to $36.3 trillion.

The most striking feature is the labor market's polarization. Knowledge workers' wage growth stagnates, with cognitive job postings shrinking by about 4%. Junior analysts, copywriters, and general programmers become redundant. Meanwhile, blue-collar and on-site service workers—plumbers, electricians, nurses, eldercare providers—see wages rise instead of fall.

The reason is simple. AI and robots struggle with physically demanding, dirty, or emotionally nuanced tasks in the real world. As societal wealth grows, demand for these 'human-touch services' soars. White-collar workers sipping coffee in offices suddenly find bonuses slashed and job-hopping futile, while the electrician downstairs now charges double their hourly rate.

Extreme Scenario: GDP Surges 32%, Labor Income Share Plummets 15%

The third pathway is the scenario most aligned with exponential technological growth. In this extreme case, AI surpasses critical thresholds, operating tirelessly in data centers at near-zero marginal cost. U.S. GDP would surge 32% compared to a no-AI path, annualized growth rocketing to 15%, with the economy doubling every 4.5 years to reach $44.4 trillion.

Conversely, overall unemployment rises to 11.9%, with cognitive workers' unemployment hitting 17.9% and wages 11.5% lower than in a no-AI scenario. Most alarming is the shift in wealth distribution: labor's share of GDP drops from 60% to 45.2%, with capital income exceeding half of national income and over 80% of new gains flowing directly into capital accounts. A surging GDP and impoverished workers coexist in this extreme scenario—the cruelest paradox of the AI economy.

Three Singularities and the Token Tax: Policies for the Inevitable Future

Looking toward 2030, McElroy summarizes the AI economy's evolution as the 'Three Singularities.' The Software Singularity is first: AI's recursive self-improvement doubles its autonomous task scope every four to seven months. The Economic Singularity is second: AI automation breaks traditional growth models, challenging 'whether infinite growth is achievable in finite time.' The Coase Singularity is third: large-scale AI agent deployment drastically reduces transaction costs, reshaping firm boundaries and economic exchange structures.

McElroy's real-world tests revealed a shocking phenomenon: stronger models systematically extracted more rent during negotiations with weaker ones, yet people remained unaware of this asymmetric bargaining power.

Facing reshaped wealth distribution rules, McElroy proposes a concrete policy direction. He notes that current U.S. and global tax structures heavily tax labor, but if labor's income share declines, alternative tax systems must be considered. He states, 'Within economic policy frameworks, I view the Token Tax as an incentive mechanism to curb over-adoption, analogous to carbon or tobacco taxes.' This is not an isolated proposal—Anthropic CEO Dario Amodei previously suggested levying a 'Token Tax' on large model usage for redistribution and public welfare. Neither Ferman nor McElroy provided definitive answers in this dialogue.

Conclusion

The three scenarios are not mutually exclusive futures but three cross-sections of the same technological curve at different adoption speeds. The key variables are not model strength but how quickly firms assimilate technology, the intensity of policy interventions, and society's response to distribution issues.

Online References:

Wall Street See: 'Anthropic Chief Economist Forecasts AI Future: Three Singularities, Wealth Distribution, and Token & Capital Taxes'

Today's Headline: 'Claude Wrote 36 Papers, But Why Does a Harvard Professor Say It Doesn’t Qualify as a 'Real Worker'?'