Generative AI has the power to super charge productivity.
But for whom?
New research suggests that the benefits are uneven.
The least skilled workers get the biggest boost. Enabling them to accomplish in hours what used to take weeks and months.
While top performers receive little to no benefit. For some it slowed them down.
Surprised? I know I was.
Lets look at a case study.
Brynjolfsson, Li, and Raymond (2023) gave AI tools to customer support staff, and here’s what they found:
AI assistance disproportionately increases the performance of less skilled and less experienced workers. In their experiment, newer agents with 2 months tenure perform just as well as more experienced agents with over 6 months tenure.
Less skilled agents saw the largest gains in productivity. Generative AI helped the lowest skills quintile agents to increase their ticket resolution by 35%.
AI did not increase productivity for the most skilled workers. Research showed that AI may distract the highest skilled workers who are already effective without it.
I had to read the last bullet twice.
And it's not just this one study. Other researchers have reached the same conclusion.
When Noy and Zhang (2023) gave generative AI to professionals for writing, it reduced inequality between high and low performers. Those scoring lower benefited more.
Peng et al. (2023) found GitHub Copilot was more beneficial for less experienced programmers.
When Choi and Schwarcz (2023) used GPT-4 to help students on law exams, they found that bottom students saw huge performance gains with GPT-4 while top students saw declines. GPT-4 had an equalizing effect.
And Doshi and Hauser (2023) found that less creative writers benefited more from GPT-4 for improving creativity of stories. But the technology had little effect for most creative writers.
Stepping back, this makes intuitive sense.
Previous technologies, like programming languages, required humans to "think" like a computer. This gave an advantage to people who were already good at logical, detail-oriented thinking.
But new Generative AI like ChatGPT works more like a helpful assistant. It handles the computational "heavy lifting" so people don't have to think like computers.
“It's like switching from manual spreadsheets to using Excel," explains Doshi. "Expert number crunchers don't get better. But new analysts can do in minutes what took them hours."
This levels the playing field. The AI substitutes for skills that some people previously had to work hard to develop.
So what’s the point?
For me it comes down to three things,
Generative AI will benefit less skilled people more than highly skilled experts. It acts as an "equalizer" by boosting the performance of novices to closer to that of experts.
Which means instead of always looking for rock stars, you can pair a mediocre star with AI and get similar results. This will have a grave impact on how we think about human capital. From whom we hire, promote, reward, pay etc.
When you democratize the computational work, cognitive skills are no longer the competitive advantage. Neither are AI tools, since they will be ubiquitous. So what skills will matter? (drum roll please) Human Creativity.
The more things change, the more they remain the same.
If you enjoyed this, please turn that ♡ into ❤️. This will help others find this newsletter. Thank you so much 😁