Study shows generative AI helps low-skill workers but not high-skill workers
Using productivity as a success metric hides the risk to high-skill workers
There are a growing number of studies on the economic impact of AI on jobs and I just finished reading “Generative AI at Work” study recently published by the National Bureau of Economic Research. Bloomberg claimed, “These findings run counter to the prevailing notion that automation tends to hurt low-skilled workers most.”
This is the wrong conclusion to draw from this study as it misidentifies the type of AI (model) tool in this study and glosses over the disturbing implications for high-skill workers. Here is a quick summary of the key takeaways.
✔ The research study is based on *one* AI assistance tool — built on a recent version of the Generative Pre-trained Transformer (GPT) from OpenAI — to assist and augment human contact center agents. It did *not* study the impact on agents (workers) from automation or human replacement AI systems.
✔Agent skill and productivity measures were based on average agent performance along 3 dimensions — handle time, call resolution, and customer satisfaction. The study found that AI tool increased worker productivity and helped newer agents move more quickly down the experience curve.
✔Productivity gains occurred disproportionately because ML systems operate by capturing and disseminating behavior patterns of the most productive (and high skill) agents.
✔ High-skill workers have less to gain from AI assistance because AI recommendations are based on their own knowledge and behaviors. High-skill workers in the setting for this study were not compensated for their contributions and are likely to see smaller direct benefits from AI adoption.
✔ The researchers predict generative AI can help retention, but also point out that increases in worker-level productivity do not always lead workers to be happier with their jobs.
✔This paper does *not* cover employment or wage effects of generative AI tools. However, the researchers acknowledge the need for companies to think about how workers should be compensated for the data that they provide to AI systems.
Note: Generative AI models are seen to have the potential to distribute human knowledge and replicate (productive) behaviors, which were previously thought to be difficult to codify. The agents in this study had discretion over whether to incorporate AI suggestions, addressing the commonly cited issue of generative AI model output being off-topic or incorrect.
TL;DR: A research study shows that AI assistance tools can increase new worker productivity but fails to mention that faster onboarding time and lower costs may increase the likelihood of employers treating workers as easily replaceable. It raises the question of how high-skill workers should be compensated for the data that they provide to AI systems. It reiterates the urgent need for organizations to fundamentally rethink their incentive and compensation structures while also investing in up/reskilling of their more experienced workers.
#generativeAI #workers #training #reskilling #upskilling #compensation #wages #productivity #AI #jobs
