AI2050 Senior Fellow Erik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor at the Stanford University Graduate School of Business and the Director of the Stanford Digital Economy Lab at the Stanford Institute for Human-Centered AI (HAI). Brynjolfsson is the co-author of two best selling books on AI’s likely impact on the economy: The Second Machine Age (2014) and Machine, Platform, Crowd (2017), both co-authored with Andrew McAfee.
Brynjolfsson’s AI2050 proposal explores how economists should measure the AI-powered economy.
Much of the success of the modern world depends on the ability of governments and corporations to make accurate measurements of economic factors and use those measurements to set policy and make decisions. For example, major restaurant chains use economic projections to determine how much meat to buy and how many people to hire. Those projections depend upon accurate measurements of employment and productivity—two factors that are increasingly difficult to measure as more work is being done by machines with little or no human input. And that trend is only likely to accelerate.
“By 2050, most production will be done by intelligent, digital machinery, requiring little or no human labor,” says Brynjolfsson. “Furthermore, value will increasingly be delivered in the form of digital goods, services and experiences. They often will have zero price, even when they create enormous value, so metrics based on prices are meaningless. How can we understand, let alone manage, what we don’t properly measure?”
Learn more about Erik Brynjolfsson:
Most people have heard of the Gross Domestic Product. What’s wrong with it?
GDP is a very good measure of production, but it is not a measure of well-being. Indeed, Simon Kuznets (who received the 1971 Nobel Prize in economics for his work on the GDP) warned that it should not be used in this way. Yet, that is routinely how policymakers, journalists and even economists use it, along with derivative measures like productivity growth. GDP is an impressive invention, and it helps us understand many things about the economy. But it misses the ways that free goods like clean air, home cooking or Wikipedia contribute to our well-being.
If we can’t use traditional metrics like the GDP for measuring the AI economy, what should we use?
The core of our approach is a new metric, which we call “GDP-B,” which measures the benefits, not the costs of goods and services. GDP-B is focused on measuring annual changes in consumer surplus from goods and services and potentially other activities like household production, government services and even changes in health, environmental, and social indicators.
Why do we need GDP-B? Isn’t the current Gross Domestic Product (GPD) good enough?
Many traditional goods and services are now being replaced by zero-price digital goods. However, these changes have not been appropriately captured in official statistics such as GDP, nor in subjective well-being metrics such as life satisfaction or happiness indices. These often have zero weight in the national statistics like GDP and productivity, even when they create enormous value.
What have you done so far?
We use massive online choice experiments to assess the changes in consumer surplus from a set of digital goods such as Wikipedia, Facebook, Google Search and Online Music. The next phase of this work includes scaling up the approach to measure the consumer value of a much broader basket of goods and services designed to be representative of the entire economy, including its non-market sectors. In parallel, we will work to create a global community of researchers using our methodology to contribute to a regularly published GDP-B Report (coinciding with annual or quarterly GDP Reports).
Where does AI fit in?
In two ways. First by improving our ability to measure the economy. tThe research is applying machine learning to the plethora of fine-grained, real-time data that is now available about activities in the economy to predict and estimate consumer surplus more accurately and scalably. Second, as more and more of the goods in the economy are digital and AI-powered, we need ways to measure their benefits, even when they have zero cost.
What is your data source? More experiments?
Preliminary work suggests that how people spend time can help predict a consumer’s valuation for goods and services. Detailed data on time-use can now be collected at low cost, suggesting that ML-based methods using such data can supplant, and perhaps in many cases replace, the massive online choice experiments that currently form the core estimates for our GDP-B work.