🔍 Read the full analysis: The Basics Of Measuring AI And Its Capabilities on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
ThorstenMeyerAI.com’s Nov. 11 article uses the history of national income accounting to explain a challenge in measuring AI: market revenue can miss benefits people receive from free or low-cost tools. It cites a Stanford working paper estimating $172 billion a year in consumer value from U.S. AI chatbots, while stressing that GDP and similar measures do not capture welfare on their own.
ThorstenMeyerAI.com published an explainer on Nov. 11 about measuring artificial intelligence, using the history of gross domestic product to show why sales and market prices may not capture the full value of AI tools. The article cites a Stanford working paper estimating $172 billion a year in value to U.S. consumers from AI chatbots, a figure it contrasts with the revenue those tools earn.
The article begins with the creation of national income estimates during the Great Depression. At the Senate’s request, economist Simon Kuznets supervised a 1934 report that found U.S. income produced fell from $83 billion in 1929 to $39.4 billion in 1932, a decline of 53%. The estimates gave lawmakers a way to quantify the economic collapse, but Kuznets cautioned that national income was not a direct measure of public welfare.
That distinction frames the article’s discussion of AI. GDP records market-valued production, so a tool that is free or priced below the value users place on it can contribute benefits that are not apparent in its sales revenue. The article points to a 2017 experiment in which the median participant would have required $17,530 to give up search engines for a year. The figure is an estimate of willingness to accept compensation, not a measure of search companies’ revenue or a direct measure of economic output.
For AI chatbots, the source cites a Stanford working paper’s annual estimate of $172 billion in U.S. consumer value. It does not provide the study’s methods or a comparison period in the supplied material. The article’s broader point is that capability assessments and economic impact measures answer different questions: performance tests can indicate what systems do under specified conditions, while revenue and GDP capture market activity rather than every benefit or cost experienced by users.
Infographic 01 · The sector clock, by output
China: how GDP changed shape
Share of GDP (value added at current prices) by sector, 1978 to 2025. In 2023 farming still held 22.8% of all jobs but produced 7.1% of GDP.
- Farming27.7%
- Industry47.7%
- Services24.6%
Reform starts
- Farming14.7%
- Industry45.5%
- Services39.8%
- Farming9.1%
- Industry45.4%
- Services45.5%
Services overtake industry
- Farming6.7%
- Industry35.6%
- Services57.7%
Communiqué 2025
Measured by output, not by jobs: the sector clocks of earlier episodes count workers. Jobs vs. output for farming, 2023: 22.8% of employment (year-end) vs. 7.1% of GDP.
Sources: NBS China, China Statistical Yearbook 2024, tables 3-2 and 4-2; NBS, Statistical Communiqué 2025 (28 Feb 2026).
Powered by Thorsten Meyer AI
AI Value Beyond Sales
How AI is measured can affect decisions by businesses, governments and the public. Revenue is observable but incomplete: a paid subscription or enterprise contract can be counted, while time saved, access to information or assistance delivered at no charge may not appear as equivalent market output. A consumer-value estimate can help describe that gap, but it should not be treated as interchangeable with GDP, company income or a verified productivity gain.
The distinction also matters when claims about AI capabilities are discussed. A benchmark score, a user survey, an estimate of willingness to pay and national economic accounts measure different things. No single number establishes overall capability or social benefit. Readers need to know what was measured, who was included, over what period and against what baseline before drawing conclusions about AI’s effects.
Products Worth Considering
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
Bose TV Speaker (2nd Gen) Compact Soundbar
Anker SOLIX C300 DC Portable Power Station with 60W Solar Panel, 288Wh, 300W
As an affiliate, we earn on qualifying purchases.
From Kuznets to AI Accounting
The source traces national accounting from the U.S. Senate’s 1932 request for income estimates to the 1934 Kuznets report. In the 1940s, the United States developed gross national product estimates for wartime planning. The first international System of National Accounts followed in 1953, and the U.S. adopted GDP as its headline measure in 1991, according to the article’s account of the Bureau of Economic Analysis history.
Kuznets explicitly identified omissions, including unpaid household labor, how income was distributed and the unpleasantness of work. The article illustrates the broader problem with unpaid work and free services: valuable activity may not carry a market price. It says that a 2017 search-engine experiment found a median willingness-to-accept figure of $17,530 for giving up search for one year. These examples are relevant to AI because many tools are free to some users, bundled into other products or offered at prices that may not reflect the benefits users perceive.
“The estimates cover primarily only efforts whose results appear on the market place of our economy.”
— ThorstenMeyerAI.com, citing the 1934 report
Limits of the Chatbot Estimate
The supplied source does not identify the Stanford working paper’s authors, publication date, methodology or assumptions behind its $172 billion annual estimate. It also does not state the chatbot revenue figure used for comparison, so the size of the gap cannot be calculated from the provided material. The estimate should be described as a working-paper finding, not a settled measure of AI’s contribution to the economy.
It is also unclear from the source how the estimate distinguishes benefits from different chatbot uses, or how it accounts for costs, harms and differences among users. Consumer value is not the same as demonstrated productivity growth, and the article supplies no benchmark results or independent evaluation of specific AI systems. Those questions require separate evidence.
What Better AI Measures Require
The next step for readers and policymakers is to examine the underlying working paper and compare its assumptions with other evidence, including user studies, business data and evaluations of specific systems. Any future estimate should state its population, time window, baseline and method clearly, and distinguish consumer benefit from revenue and measured output.
The article presents measurement as an ongoing problem rather than a final verdict on AI’s economic effects. Future assessments will need multiple indicators: what systems can do under defined tests, how people use them, what users gain or lose, and whether organizations achieve measurable changes in output or costs. The supplied source does not announce a new benchmark or policy decision.
Key Questions
What is the main development in this article?
ThorstenMeyerAI.com published an explainer connecting the history and limits of GDP with the challenge of measuring AI’s capabilities and value.
What does the $172 billion figure refer to?
The article attributes the estimate to a Stanford working paper and describes it as annual value to U.S. consumers from AI chatbots. The supplied material does not give the paper’s methods or publication details.
Does GDP measure the full value of AI?
No. GDP measures market-valued production, not every benefit or cost. Free services and time saved may be difficult to capture, while consumer-value estimates measure something different from GDP.
Did Kuznets say national income measures welfare?
No. In the 1934 report, Simon Kuznets warned that national welfare could scarcely be inferred from national income alone and described important omissions in the estimates.
What remains unknown about the chatbot estimate?
The source does not specify the Stanford paper’s authors, method, assumptions, comparison period or chatbot revenue figure. The estimate therefore cannot be independently evaluated from the supplied information alone.
Source: ThorstenMeyerAI.com
