Study: Supplemental GenAI May Soften the Blow to Workers
When AI complements human work, companies benefit by productivity gains. There may even be a silver lining for workers.
By Jennifer Zaino
Since its rapid rise in the past year, Chat GPT and generative AI (GenAI) in general have dominated discussions about labor and jobs. One 2023 forecast predicted that in the next five years 74% of CEOs expected AI to lower their need for labor. Additionally, the overall impact of GenAI on the economy is a topic of debate.
One thing is clear: The pace of GenAI job displacement is likely to accelerate, and it will differ from previous automation waves.
But which jobs and occupations will be hardest hit — and which may even benefit from AI assistance?
At a recent MIT IDE seminar, Andrea Eisfeldt, Laurence D. and Lori W. Fink Endowed Professor of Finance at the UCLA Anderson School of Management, offered some insights. Drawing from research that she and her colleagues published here, occupations that tend to have higher wages — such as those in mathematics and computation — could feel the greatest impact from GenAI.
The authors say the research “is the first to link occupational exposure to Generative AI and firms’ workforce exposures to labor market outcomes including wages and job postings.” And the study, which looked at U.S. firm profits pre- and post-adoption of GenAI, seem to support the idea that the labor-substitution effect will increase firm value.
“Firms whose workforces are highly exposed to Generative AI increased in market value by almost 5% relative to firms with a low exposure in the two weeks following the release of ChatGPT on November 30, 2022,” Eisfeldt said.
“To improve productivity, generative AI can be either a substitute [for] or complementary to labor,” Eisfeldt said at the seminar. When core tasks of an occupation — those that are key to performing the function — have a high potential to be replaced by GenAI, firm value can increase through lower hiring and wages.
GenAI could, for example, substitute for web developers’ core tasks of writing supporting code for web apps or web sites because code generation is based on specifications provided in text input.
Finance Jobs at Risk
Eisfeldt pointed out that the types of occupations using GenAI automation will be different from prior automation waves. The impact will be most significant on non-routine, cognitive, analytical fields of work such as computers and mathematics, management, and other high-wage fields where LLMs can accomplish many tasks.
The study sees telemarketing, bookkeeping, accounting and auditing clerks among the most at-risk occupation titles.
This dovetails with research by Accenture’s Technology Research and Thought Leadership group. In November, H. James Wilson, Global Managing Director, stated that 70% to 80% of the work performed for some job functions, including marketing or finance, is going to be impacted in a significant way by GenAI.
After the release of ChatGPT, demand and wages for certain occupations fell. For occupations for which core tasks can be substituted by GenAI, job posting and hourly wage rates fell significantly more than for occupations with mainly supplemental task exposure. “If the generative AI exposure affects your core tasks, the hiring impact is more negative and your wages decline more,” Eisfeldt said. (See chart 1)
Who will be in the best position to maintain their jobs and wages? Workers are in less jeopardy when humans are still needed to accomplish the critical functions of the occupation, or when GenAI is complementary to the core tasks performed by workers like mechanics or physician assistants, according to the research study.
Supplementing human tasks may even be a bright spot.
If GenAI takes on a supporting role, workers increase firm value by becoming more productive and they may command higher wages amid demand for their specialized labor.
When some supplemental tasks are automated by GenAI instead of full job replacement, “there is a smaller decline, or potentially, even an increase in those job postings,” said Eisfeldt.
Generative AI Exposure and Supplemental Exposure Share by Major Occupation Group
As noted in Chart 2, safer occupations, at least for now, are those that involve primarily physical tasks — think leather workers, stonemasons or hair salon shampooers. A PWC analysis finds that some of the skills with rising demand are those not easily performed by AI such as personal care and services (+155% growth) and environmental services (+48%).
The Data Dividend
Another facet of Eisfeldt’s research looks at the critical value of data to firms that want to reap gains from GenAI. And that’s where established enterprises have the edge over startups and disruptors. “It’s something that incumbent firms have… that would challenge new entrants.”
Among the data requirements she cited are:
• Fine-tuning and customization of Large Language Models (LLMs). This requires large, firm-specific data training assets
• Process automation requires a data pipeline to and from LLMs
• Task-specific prompting is improved by data trained on past customer interactions
- LLM analytical capabilities enable extraction of new data from existing data, decision support, and output data.
