Advances often launch near universities, new research finds, but ripple out geographically and to lower-skills positions.
Linda A. Cicero
Rising inequality has focused attention on the benefits of new technologies. Do these primarily benefit inventors, early investors, and highly skilled users, or does society gain as their adoption generates employment growth? And are a few firms monopolizing jobs in these new technologies, or are they spreading out across society?
In a new paper, my colleagues and I use novel, text-based methods to trace out the impact of 20 new technologies on employment of low- and high-skilled workers across the United States. We find that while most new fields originate in high-skilled, highly educated regions around universities, many jobs spread out over time throughout the country and into lower-skilled positions.
We analyzed the full text of more than 3 million patents, 200 million online job postings, and 300,000 conference calls held by listed firms between 2002 and 2020 to measure the spread of new technologies across geographies, jobs, and firms. We first identified two-word phrases, or “bigrams,” associated with a new technology, by examining several decades of U.S. patent filings. We focus on bigrams because they are more precise than single words: While words like “autonomous” or “cloud” could have a variety of colloquial meanings, “autonomous vehicle” and “cloud computing” are much less ambiguous. We then determined the 500 most common bigrams used in business discussion by analyzing 321,373 earnings calls by publicly listed firms.
From this list, we use manual discretion to select those bigrams that appeared to clearly and unambiguously reflect specific technological advances that we believe changed the way businesses operate. We then group those bigrams into 20 technologies, which form the basis of our analysis. This includes, for example, “artificial intelligence,” “cloud computing,” “machine vision,” “hydraulic fracturing,” “solar power,” “driverless cars,” and “3-D printing.”
In the final step, we match these bigrams for our 20 key technologies to the Burning Glass online U.S. job-posting database. This provides almost all online job postings for 2007 and 2010-2020, with industry, geography, and skill coding. This has been extensively used by economists to evaluate U.S. labor markets and appears to match well to overall U.S. hiring (e.g., Hershbein and Kahn 2018). With these data, we can then map out the geography of hiring over time and across firms in some of the major new technologies of the last 20 years.
New Jobs in New Technologies
We tracked two measures of activity on an annual basis for each technology: the share of earnings calls that mention the given technology and the number of new jobs in Burning Glass mentioning those technologies. In some cases, such as touch-screen and RFID, the number of mentions climbs and then fades, presumably reflecting the increasing ubiquity (and hence the declining competitive relevance) of the technologies. In others, such as 3-D printing and artificial intelligence, there is a steady climb over time.
While in some cases a technology continues to be important even after its mentions in earning calls drop (e.g., GPS technologies), in general, we found the mention of a technology to closely correlate to the number of jobs. To take one example, artificial intelligence — we see that by 2019 over 15 percent of earnings calls mention one of the key words around artificial intelligence, as does nearly 1 percent of the job announcements.
New Jobs Spread into Lower Skilled Jobs over Time
Our research found a sharp decline in the skill level required for the positions associated with new technologies over time. Even in cases where demand for positions is sharply accelerating, such as AI and virtual reality, the share of skilled positions subsides over time. Indeed, on average across all 20 technologies, the share of college-educated workers declines by 1.2 percent a year. Technologies emerge on average with two-thirds of new workers being college educated, but 12 years later, this share has fallen below half.
Why is this happening? Early in the life of a new technology many of the jobs focus on research and development. For example, in the early 2000s when the original smartphones were being developed, much of the employment was of engineers and scientists, many of these in Silicon Valley. As the product became widely adopted, employment spread out across the country and skill distribution, as jobs were created in selling, repairing, and customizing smartphones.
New Jobs Spread Out Geographically over Time
We also find that hiring in new technologies became less concentrated over time. We see very clearly that while technologies start very concentrated (on average in the first year 45 percent of jobs are in the 5 largest local labor markets), they spread out across the country, so that within 10 years only 38 percent are in the 5 largest markets. Our research also showed that the geographic spread of new job postings is strongest for low-skilled jobs and weakest for high-skilled jobs. So, while the low-end jobs rapidly disperse across the country, the higher-end ones are more likely to remain bunched together.
The Birthplaces of New Technology Retain High-Skilled Hiring
One fact driving the slower spread of high-skilled jobs is the presence of technology hubs — these are the areas where the new technology emerged. We define technology hubs as those that collectively account for 50 percent of jobs in the emerging year of the technology (or 2007 if they started before).
What we find is that while low-skilled jobs rapidly spread out over time, so that within 20 years there is no significantly higher share near the initial hubs, this is not true for high-skilled jobs. These spread out more slowly, and the initial advantage around the technology hubs persists, although at a more diminished rate over time.
The Formation of Technology Hubs
So, what causes these initial technology hubs to form? We plotted the job postings in our technology’s emergence year against local measures of university activity and skills. We examine the size of universities in the area — a measure of university presence — and find this is strongly related to having a large number of new technology jobs. Similarly, we found new technology jobs are also more likely in areas with a high density of university students or college-educated population. In summary, regions like the San Francisco Bay Area or Boston with a greater academic presence — whether manifested by greater research university presence or a more educated workforce — were more likely to enjoy a surge of initial hiring as a new industry emerges.
We can also evaluate this looking at specific technologies and find every technology except fiber optics is more likely to have been founded in areas with a strong university presence, suggesting that this pattern is not driven by any single technology.
In conclusion, we argue our evidence suggests employment in new technologies does slowly spread out over skills and space. While there is a clear concentration in high-skilled jobs in a few areas at the emergence of new technologies, over time this dissipates. Interestingly, these birthplaces of new technologies do, however, retain some long-run advantage, particularly in high-skilled jobs. We find that proximity to universities, particularly research universities, and a highly skilled labor market are key drivers of the initial location of new technologies. This suggests that a local focus on universities, skills, and research can generate a persistent local advantage while also generating longer-run national employment benefits.
Nicholas Bloom is the William D. Eberle Professor of Economics, a Stanford Institute for Economic Policy Research senior fellow, and the co-director of the Productivity, Innovation and Entrepreneurship program at the National Bureau of Economic Research. His research focuses on management practices and uncertainty.
Stanford HAI's mission is to advance AI research, education, policy and practice to improve the human condition. Learn more.