Discover more from Workonomics
Mommy, daddy, is ChatGPT going to automate your jobs? | Roundup #18
Plus: other bedtime stories about gig economy regulation and shrinking inequality
Welcome to our new readers to the Workonomics Roundup! I post these Roundups every two weeks, summarizing some of the latest labor, technology, and policy updates.
I’m poring through Twitter, economic research, policy updates, and employment data and summarizing it in a brief update you can read
at work on your way to work.
As always, feel free to comment on this post or drop me a line on Twitter to discuss or share any feedback.
A New Paper on ChatGPT and Job Automation
Researchers at OpenAI (the creator of GPT-4) and the University of Pennsylvania conducted a study to assess the likelihood of large language models (LLMs) automating tasks performed in different occupations. The study found that for nearly 20% of the US workforce, more than 50% of the tasks they perform in their function can be “exposed” to automation by LLMs.
A few highlights:
“Task exposure” to LLM automation means that LLMs could decrease the time for workers to complete that task by 50%.
Contrary to other studies conducted on the effect of machine learning on the labor market, this study found that higher-wage jobs have a higher exposure to LLM automation than lower-wage jobs.
In particular, jobs dependent on programming and writing are more exposed to automation, whereas those dependent on science and critical thinking are less so.
There are several limitations to this study, which the authors acknowledge. Jobs that include many tasks “exposed” to automation aren’t necessarily going away. Survey researchers and creative writers may be able to take advantage of AI for some of their tasks and spend more time on the more creative, less procedural aspects of their jobs. They could also, like, spend less time working and do other things with their leisure time (while having the same productive output).
Discussing the study and all of its limitations in detail would take up a few Roundup posts, so I won’t do it here. The numbers should all be taken with a grain of salt. But at the same time, it’s an important early indicator of what’s to come.
Also, here’s ChatGPT on how to tell a bedtime story about job automation:
The Latest in Labor Economics Research
An NBER working paper from MIT and UMass Amherst showed that the tight labor market shrunk inequality since low-wage workers saw their paychecks increase faster than high-wage workers. Inequality has been steadily increasing since the 1980s, and according to the study, the hot labor market counteracted around 25% of the long-run increase.
Interestingly, the study also showed that rising wages for workers contributed much less to inflation than pundits have predicted:
In a separate NBER working paper, MIT economist Daron Acemoglu and colleagues studied how robot adoption in Dutch factories has affected manufacturing workers there. Blue-collar workers performing routine, automatable tasks saw lower earnings and employment rates once robots entered the workplace. Meanwhile, workers who performed more differentiated tasks fared better, post-robot adoption.
Like much of Acemoglu’s other research, the paper highlights that automation will positively and negatively affect workers. Which effect prevails and to what extent workers are protected from adverse effects depends — as always — on our policies, labor market institutions, and the particular implementations of the technologies themselves.
Gig Policy Updates
A few new gig economy policy proposals at the state level over the last couple of months. Note that the intent of the regulations is the same (protect workers and, to a lesser extent, consumers), but each state’s approach is very different:
A recent bill in Colorado proposes that rideshare and delivery workers should see what the customer paid for the trip or meal. The bill would also require platforms to show customers how much of their payment ultimately went to the worker. This proposal opens up a new front in the battle over gig policy, focusing on algorithmic transparency. (Colorado Newsline)
In Minnesota, a proposed bill would require gig platforms to pay drivers for fuel and car expenses, a workers’ compensation equivalent, and minimum payments for trips and cancellations. It is a marked change from 2021 when Minnesota state representatives proposed a bill to reclassify all Uber and Lyft drivers as employees. (Minnesota Reformer)
A bill in Massachusetts, backed by two large unions (the Machinists and the SEIU), will give rideshare workers a minimum wage, paid sick time, unemployment insurance, and collective bargaining rights. Similar to the Minnesota bill, the bill sidesteps the question of whether workers are employees or not. (WBUR)
I’m curious to see whether these bills get passed and, if so, how gig platforms will react. If things go as planned, labor economists will have some interesting natural experiment data to analyze in a few years.
Thanks for reading Workonomics! Subscribe for free to receive new posts and support my work.