As economists often remind us, productivity is not everything, but in the long run, it is almost everything. That foundational economic truth, famously highlighted by Nobel laureate Paul Krugman, captures the essence of total factor productivity (TFP). TFP serves as the primary metric economists use to measure the contribution of technological innovation to broader economic growth, tracking an economy’s sustained ability to produce greater output using the same amount of inputs.
Because the continuous process of producing more with less is the fundamental mechanism by which human living standards improve, TFP is arguably the most vital measurement in modern economics. As Krugman explains, a country’s ability to elevate its standard of living over time depends almost entirely on its capacity to raise its output per worker. Technology is the primary engine that makes that happen, and TFP is the instrument used to gauge its success.
To understand the tangible, macroeconomic stakes of technology-generated productivity, consider a recent paper published by the National Bureau of Economic Research (NBER). The study estimates that an additional 0.5% of annual TFP growth would be enough to stabilize the United States government’s finances, locking in the debt-to-GDP ratio at today’s levels.
While 0.5% may sound modest, the long-term compounding effects are staggering. If sustained over the next decade, NBER estimates that a 0.5% boost to annual TFP would reduce the baseline forecast for US government debt by a remarkable $2 trillion. Over a 30-year horizon, that same sustained increase would drive the US government’s debt-to-GDP ratio down by 42 percentage points compared to the NBER’s baseline forecast, and an astonishing 80 percentage points lower than its pessimistic scenario.
Given the seemingly intractable and hopeless state of government finances today, simply maintaining the current level of indebtedness feels like a dream scenario that is far too good to be true. Yet, new research from artificial intelligence pioneer Anthropic suggests that the economy might actually be capable of outperforming those projections.
Anthropic recently conducted a comprehensive study analyzing 100,000 real-world conversations on Claude.ai. The goal was to estimate how long the tasks within those conversations would take with and without AI assistance, allowing researchers to study the broader productivity implications across the entire economy. Their conclusion was striking: large language models (LLMs) have the potential to raise total factor productivity by 1.1 percentage points.
If an increase of 0.5% is enough to stabilize US government finances for decades, the implications of a 1.1% boost are profound. Such an increase would likely go a long way toward solving a multitude of looming fiscal and economic challenges.
Naturally, there are plenty of reasons to approach such optimistic forecasts with a healthy dose of skepticism. The Anthropic study found, for instance, that Claude can save teachers four hours of labor by generating complete lesson plans and curricula in just 11 minutes. However, translating those localized time savings into broader economic output requires complex economic modeling, an exercise notoriously fraught with best-guess assumptions and false precision.
Even if Anthropic’s calculations regarding time savings are entirely accurate, the company’s model could still be wrong about overall productivity growth. It is entirely possible that society will simply use the time saved by AI to engage in activities that are economically unproductive, such as watching more short-form videos or reading newsletters. In that scenario, AI would successfully elevate human welfare by granting people more free time, but it would fail to expand aggregate wealth or economic output. While that outcome would be a net positive for individuals seeking more leisure, it would offer no direct relief to governments desperately hoping for a silver bullet solution to their mounting debt crises.
Conversely, there are strong arguments suggesting that Anthropic’s model might actually be overly pessimistic. As the researchers themselves note, their methodology does not take into account the future rate of technology adoption, nor does it account for the vastly larger productivity effects that will inevitably stem from much more capable AI systems in the pipeline.
In other words, the study’s baseline assumes that society will continue to use AI exactly as it does today, relying on today’s language models—unimproved and static—for the next ten years. Given that language models are noticeably improving every few months and that humanity has only just begun to learn how to effectively deploy them, Anthropic’s researchers acknowledge that their estimate may well represent an approximate lower bound on the true productivity effects of artificial intelligence.
If 1.1% truly serves as the lower bound for AI-induced productivity, society might successfully pay down national debt while simultaneously enjoying more leisure time. Furthermore, that projection accounts only for AI’s impact on non-physical, digital work. The calculus will shift even more dramatically once advanced robotics enter the commercial workforce.
To completely dismiss this optimism is to assume that the trillions of dollars corporations plan to spend on AI capital expenditures, research, and development will be entirely wasted. While technology revolutions do not always arrive precisely on schedule, and capital misallocation is always a risk, the most compelling reason for optimism is that Anthropic’s 1.1% estimate is based entirely on AI making existing tasks faster to complete. The model does not factor in AI’s transformative potential to completely reinvent the way those tasks are executed.
Historically, transformative productivity improvements—such as the widespread introduction of electrification, computing, or the internet—did not stem merely from speeding up legacy tasks, but rather from fundamentally reorganizing the entire structure of production. While it is impossible to accurately model these entirely new ways of operating prior to their widespread adoption, the ultimate economic impact will likely eclipse anything captured in current measurements.
Anthropic’s researchers have been careful to contextualize their hopeful findings by explicitly detailing the limitations of their methodology and outlining the numerous assumptions underpinning their models. Skeptics might rightly point out that even if all those assumptions prove correct and an AI-driven productivity boom temporarily solves the US government’s debt problem, lawmakers could easily spend their way right back into fiscal jeopardy.
Nevertheless, given the near-universal pessimism surrounding the trajectory of public finances, even a small probability that Anthropic’s estimates are accurate offers a compelling reason to update our economic priors. The financial trajectory of the US government may not be as intractable as widely believed, and the long-term outlook for the US dollar may not be as perilous as the doom-sayers suggest. In the long run, productivity remains the ultimate driver of human progress, and artificial intelligence may be on the verge of making the global economy significantly more productive.
Comments (0)