The latest artificial-intelligence systems look more impressive by the day. They now regularly crack old unsolved math problems and strain the top-end of model evaluation benchmarks. On occasion they will even sneak out of one company’s lab, hack their way onto the internet, and break into another company’s network. By all sorts of measures, generative AI capabilities continue to rise. But where’s their big economic impact? A new report from Google’s economics team helps explain the apparent paradox.
“Google’s AI & Economy ATLAS v1.0: Mapping Gemini Usage in the Economy” offers numerous insights relevant to the quandary at hand. The company looked at some 15 million anonymous interactions across more than 150 countries with its Gemini products. The goal was to do a deep dive to see how “people are using Google’s AI tools for various tasks at work and in their day-to-day lives,” according to the company.
While perusing social media might give the impression of a country immersed in AI—not so much, really. Google finds AI usage in 68 percent of detailed occupations covering nearly 90 percent of US employment. Superficially, that sounds like mass adoption is here. Yet AI comes into play for only a fifth of tasks in the typical occupation with any AI use. Also: “Only 3 percent of occupations showed AI usage for over 75 percent of their tasks; these occupations include software quality assurance analysts and testers, human resources specialists, and document management specialists,” according to Google.
Then there’s the nature of all those many millions of workplace interactions. Attempts to automate a task end to end account for under 10 percent of AI conversations in what’s termed “non-routine cognitive” work—such as “hypothesis testing and creative design.” And while that specific category makes up about 35 percent of professional tasks in the economy, it accounts for almost 65 percent of work-related AI interactions. The main use pattern seems to be a collaborative one—give me a rough draft, summarize this report, help me brainstorm—rather than “just do the whole thing” one-shot automation.
From the paper:
Taking a step back, our initial ATLAS v1.0 data therefore sheds new light on several widespread claims about AI’s impact on the economy. For instance, we do not find evidence in ATLAS to support the claims that AI is about to cause massive automation and displacement of white-collar work; that AI is irrelevant to blue-collar work; or that the purpose of AI is strictly to automate tasks. Neither does the data support the claim that the global “AI race” is strictly between the US and China from the perspective of leadership in adoption. However, appropriate prudence is required in interpreting these results: they constitute early observations and could evolve as the frontier of AI capabilities continues to advance, the ways in which businesses and individuals use these capabilities expand, and as the analytical methodologies economists and researchers use to measure their impact also progress.
Back to the business productivity paradox, a version of which leads off the paper:
In 1987, economist Robert Solow quipped that the computer age was visible everywhere except the productivity statistics. Today we see the signs of a new Solow paradox: AI appears to be everywhere, yet its impact remains hard to discern in many traditional measures of employment, productivity, and growth.
Here’s what may explain today’s underwhelming impact: We remain in the early days of the AI revolution, as the paper documents, even if it has been nearly four years since the debut of OpenAI’s ChatGPT. When a company decides to adopt some snazzy new technology, lots needs to be done before the company can become more efficient or develop new goods and services. If the technology is important enough, management needs to actually rewire or invent the business processes that make the tool worth having.
For a while, at least, the company looks like it’s spending more without producing more, though what it’s really producing is intangible assets—such as trained staff and rebuilt processes—much of which nobody officially counts. Productivity may look like it’s stalling out rather than increasing due to a technological advance. Only later, as the firm completes its reorientation, does productivity unequivocally and measurably improve. What the Google data suggests—broad but shallow use— is consistent with the notion that we’re likely still in the downstroke of such a J-shaped curve because it’s early days.
This is basically what economists have been saying, but Google has added an additional bit of evidence to support the view.