Can Artificial Intelligence Save the US Finances? Anthropic's analysis indicates that AI can enhance TFP total factor productivity to stabilize finances

In the world of economics, there is a metric often regarded as the “ultimate answer” used to measure the true driver of long-term growth in an economy, which is “Total Factor Productivity” (TFP). It reflects whether an economy can continue to generate more output through technological progress and efficiency improvements while holding capital and labor inputs constant. Economists believe that TFP almost determines a country’s long-term fate. Nobel laureate Paul Krugman once pointed out that whether a nation can improve its standard of living over time depends almost entirely on its ability to increase per capita output. Technological progress is at the core of all this. This article is excerpted and compiled from The Blockworks.

Stable US debt-to-GDP ratio can ensure fiscal stability

The importance of TFP is not only reflected in abstract growth theories but also directly related to the sustainability of government finances. A recent study by the National Bureau of Economic Research (NBER) in the United States indicates that if the US government can keep its debt-to-GDP ratio stable, then an annual increase of just 0.5 percentage points in total factor productivity growth is sufficient to stabilize US fiscal conditions.

0.5% may seem insignificant, but its impact is profound. According to the study, if such productivity growth can be maintained for ten years, the baseline forecast of US government debt would decrease by about $2 trillion; over 30 years, the debt-to-GDP ratio would be 42 percentage points lower than the baseline forecast, even 80 percentage points lower than a pessimistic scenario.

AI-assisted human productivity could boost total productivity

In the context of record-high US fiscal deficits and debt levels, such results are nearly unbelievable. However, researchers at AI company Anthropic believe that the potential of technology may be far greater. Recently, Anthropic analyzed approximately 100,000 actual interactions with Claude.ai, attempting to estimate the time difference required for humans to complete the same tasks with or without AI assistance, and further infer its impact on overall economic productivity. The study concluded that AI assistance has the potential to increase total factor productivity by about 1.1 percentage points.

This figure has drawn widespread attention. If a 0.5% productivity increase is enough to stabilize government finances for decades, then a 1.1% increase could theoretically have a disruptive impact on the economy and public finances. Of course, such optimistic estimates are not without skepticism in academic and policy circles. Anthropic’s study also admits that its analysis heavily relies on model assumptions. For instance, the study shows Claude can complete a course design within 11 minutes, saving teachers about 4 hours of work, but whether “saving time” necessarily translates into “increased output” remains highly uncertain.

Critics point out that the saved time may not be invested in higher-value economic activities but instead could be used for entertainment or consumption, such as scrolling social media or reading reports. In such cases, AI indeed enhances people’s welfare and leisure time but may not increase overall wealth, and its help in solving government debt problems is relatively limited. However, Anthropic also believes that their estimates may actually be conservative. The research does not account for the speed of AI adoption nor the ongoing evolution of model capabilities that could bring additional productivity gains. In other words, the study assumes humans will continue to use existing language models in the same way over the next ten years. Given the rapid advancements in large language models every few months and the fact that humans are still quickly learning how to apply them, Anthropic believes 1.1% might only be the “lower bound” of AI-driven productivity effects.

More importantly, this study only measures the impact of AI “accelerating the completion of existing tasks” and does not consider the fundamental reorganization of workflows and production methods brought about by technology. Anthropic points out that major productivity leaps in history, from electricity and computers to the internet, were not just about doing old things faster but fundamentally changing how things are done.

Such structural changes are difficult to model but often bring the most profound impacts. Even so, researchers maintain a cautious approach, listing detailed limitations and assumptions of their methodology. They also acknowledge that even if AI truly creates greater fiscal space for governments, future legislators might still increase spending and accumulate debt again. However, given the widespread perception that fiscal risks are imminent, even a small chance that this optimistic scenario might come true makes it worth looking forward to.

This article “Can AI Save the US Fiscal Situation? Anthropic’s Analysis Shows AI Can Boost TFP and Stabilize Fiscal Conditions” was first published on Chain News ABMedia.