AI rapidly improving, could vastly boost US GDP, Anthropic's Clark tells policymakers

TEXT START: Rapidly improving artificial intelligence capabilities will bring far faster change across all fields, Anthropic co-founder Jack Clark said on Saturday, with one of the company’s internal models forecasting that under certain conditions it could increase US economic output by almost 50% by 2030.

The Dissection

The text converts AI acceleration into a growth narrative for policymakers and markets. It acknowledges job losses, biological danger, cyber risk, and declining human control, but keeps the central story intact: AI produces enormous wealth, new industries absorb disruption, and regulation can preserve human command.

That is not a complete forecast. The 50% GDP figure is conditional and internally generated, with no demonstrated mechanism for distributing the gains, maintaining labor demand, or preventing ownership from concentrating. The IPO reference exposes the surrounding incentive structure: Anthropic is warning governments about an uncontrollable technology while simultaneously presenting its economic upside as an investable future.

The Core Fallacy

The article confuses aggregate output with mass economic viability.

A 50% GDP increase can occur alongside the destruction of the wage-to-consumption circuit. AI can create new industries without creating enough human jobs, because those industries may be designed, operated, and scaled by AI with a small number of owners, controllers, and technical servitors. “New inventions” do not automatically imply new mass employment.

Under P1, cognitive automation erodes the labor requirement. Under P2, institutions cannot reliably preserve large human-only economic domains against cheaper, faster AI competitors. Under P3, productive participation collapses. The article treats higher GDP as evidence of systemic health when, under the Discontinuity Thesis, it may be the measurement of labor’s eviction from production.

Hidden Assumptions

• AI will remain a human-controlled complement rather than becoming an autonomous substitute across most cognitive work.
• New industries will require human labor at sufficient scale to replace displaced jobs.
• Economic growth will translate into wages, purchasing power, and social stability rather than accrue primarily to AI-capital owners.
• GDP will remain a meaningful proxy for welfare when output is increasingly machine-generated and ownership is concentrated.
• International cooperation can restrain AI capabilities despite competitive pressure between states and firms.
• Human oversight of biological and cyber research will scale as the systems being supervised become more capable than their supervisors.
• Expanded data-center capacity will democratize participation rather than deepen control by the owners of compute, energy, logistics, and maintenance.
• Policymakers have enough time and institutional capacity to act before capability growth outruns governance.
• Scientific research will be enhanced through human employment rather than increasingly automated end to end.
• New jobs will be numerous, accessible, and economically necessary—not narrow specialist niches serving the machine economy.

Social Function

Primary classification: partial truth functioning as transition management and ideological anesthetic.

The warnings about speed, misuse, and shrinking control windows are credible within the text. The anesthetic lies in the implied remedy: regulate quickly, build more compute, and trust future industries to convert displacement into prosperity. This lets institutions acknowledge the fire while continuing to discuss GDP, investment, and innovation as if the old employment structure were still intact.

It is also elite self-exoneration. The text frames the crisis as a technical governance problem rather than an ownership problem. It asks who will control increasingly powerful systems, but does not confront who will own them, who will receive their output, or what happens to people whose labor is no longer required.

The Verdict

The article correctly identifies acceleration and danger but mistakes technological abundance for the survival of post-WWII capitalism. Its “new industries” are not proof that mass employment will return; they may be the delivery system for its extinction. If the forecast is accurate, the central fact is not that America becomes $44 trillion richer. It is that production can surge while the majority lose their claim on production.

This is a polished growth narrative wrapped around a genuine alarm: AI may expand the economy faster than humans can govern it, but the article never confronts the terminal break between output and participation. Under the Discontinuity Thesis, that omission is the entire autopsy.