🔍 Read the full analysis: What AI Metrics Can Show About Today’s Models on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
ThorstenMeyerAI.com’s November 11 article applies the history of GDP to the measurement of AI, arguing that model benchmarks and commercial revenue cannot by themselves show the full value or effects of AI systems. The source provides no specific current model benchmark results; it points instead to a Stanford working-paper estimate of $172 billion in annual U.S. consumer value from AI chatbots and cautions that measurement choices leave important activity out.
ThorstenMeyerAI.com published an article on November 11, 2026, arguing that the metrics used to assess today’s AI models show only part of their value and effects. The article draws on the history of gross domestic product to caution that measurable output, prices and usage do not by themselves establish how much people benefit or how that value is distributed.
The article does not report new benchmark results or a comparison of particular AI models. Instead, it uses a historical case to frame the measurement problem: in 1934, economist Simon Kuznets presented the U.S. Senate with estimates showing national income produced had fallen from $83.0 billion in 1929 to $39.4 billion in 1932. The source describes this as a 53% decline, giving lawmakers a way to quantify the economic collapse.
The source says Kuznets also warned that national income measures covered chiefly activity appearing in markets and could not serve as a direct measure of national welfare. It applies that distinction to AI: a model’s performance on selected tests or the revenue its provider earns may not capture all consumer value, unpaid effort, or effects on people’s lives. Those are arguments about how to interpret measures, not findings from a new study of model performance.
As one example, the article cites a Stanford working paper that estimates U.S. consumers receive $172 billion a year in value from AI chatbots, an amount it says is far above chatbot revenue. It also cites a 2017 experiment in which the median participant would have needed $17,530 to give up search engines for a year, despite paying no price for access. The source does not provide the working paper’s methods or the underlying chatbot revenue figures, so the estimate should be read as an attributed research estimate, not a settled measure of AI’s total value.
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
What AI Value Measures Leave Out
AI metrics can shape decisions by developers, businesses, investors and public agencies. Benchmarks may help compare performance on particular tasks, while revenue and usage figures track commercial activity. But none of those measures, on its own, answers broader questions: whether a system improves users’ lives, who receives the gains, what work it replaces or adds, and what costs fall outside a market transaction.
The article’s GDP comparison matters because it separates measured output from social welfare. A high score on a selected test is evidence about performance on that test; it is not proof of broad usefulness. Likewise, a consumer-value estimate may suggest that people value a service more than they pay for it, but does not establish how benefits and harms are distributed across users or society.
This framing is relevant as organizations choose which AI systems to adopt and how to judge their results. The practical takeaway is not that metrics are useless, but that readers should ask what each metric includes, what it excludes, and what evidence supports any conclusion drawn from it.
Popular right now
Samsung 43-Inch M70H Series 4K Mini LED Smart TV (2026)
Ninja Crispi Microwave & Air Fryer with FusionCrisp Technology, 14-in-1, Black (MW101BK1)
Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
From Kuznets to AI Measurement
The article traces the U.S. national income estimates to a Senate request during the Great Depression. According to its account, the Senate passed Resolution 220 in 1932, and the Commerce Department asked Kuznets to lead the work. The report, National Income, 1929–32, was submitted on January 4, 1934. It counted market-related income such as wages, salaries, profits, interest and rents.
The source says the accounts later developed into broader national measures: the United States added gross national product estimates in the 1940s, the first international System of National Accounts appeared in 1953, and the U.S. adopted GDP as its headline measure instead of GNP in 1991. GDP remains useful for tracking the value of final goods and services produced within a country, but it is not designed to describe every form of work or well-being.
The article points to unpaid household work as one gap in market-based measures and uses a German valuation of such work to illustrate its scale. It then extends the same concern to digital services and AI, where users may receive benefits that are not reflected in the price they pay. The historical comparison is a framework for thinking about measurement, not evidence that AI and national accounts are identical problems.
What the AI Figures Cannot Settle
The article does not identify which AI models, benchmark suites or deployment periods its broad references to “today’s models” cover. It supplies no model-by-model scores, testing conditions, or independent evaluation of accuracy, reliability, safety or performance in real-world use. Readers should not treat it as a current benchmark report.
The cited $172 billion annual estimate is attributed to a Stanford working paper, but the source material does not explain its methodology, assumptions, uncertainty range or publication status beyond calling it a working paper. It also does not supply a comparable figure for chatbot revenues. The estimate therefore cannot be independently assessed from the provided material, and it should not be presented as a confirmed total of AI’s economic value.
More broadly, the article does not quantify AI’s costs, distributional effects or impact on employment and productivity. It makes a case for broader measurement, but it does not establish which metric would capture those effects or how they should be combined.
Better Evidence for Model Impact
The source article sets out a measurement question rather than announcing a new research program or a scheduled policy decision. The next useful evidence would include clearly documented evaluations of specific models, with the task, testing date, comparison group and limitations stated alongside results.
For economic claims, readers can look for the Stanford working paper and examine how it defines consumer value, which users it studies and how it handles uncertainty. Comparisons with provider revenue should use matching periods and clearly stated methods. Further evidence would also be needed to assess effects beyond consumer value, including costs, access and distribution of benefits.
Until such details are available, the article’s central distinction remains a guide to interpretation: metrics can answer defined questions, but a score or dollar estimate should not be treated as a complete account of what AI does for society.
Key Questions
Is this article reporting new AI benchmark results?
No. It reports no new model scores or comparisons. It discusses how to interpret AI metrics using the history and limits of GDP.
What does the Stanford estimate say?
The source attributes to a Stanford working paper an estimate of $172 billion a year in value to U.S. consumers from AI chatbots. The provided material does not include the methods or uncertainty range, so the figure remains an attributed estimate rather than a confirmed total.
Why compare AI metrics with GDP?
The comparison highlights that a measure can be useful while still leaving important activity or welfare out. GDP tracks market production; the article argues that AI benchmarks, usage and revenue figures also capture only specific parts of a broader impact.
What remains unknown about the claims?
The source does not specify which models or benchmarks it means, nor does it provide detailed methods for the consumer-value estimate. It also does not quantify AI’s costs, effects on work or distribution of benefits.
Source: ThorstenMeyerAI.com
