March 21 – Will ChatGPT, the chatbot from artificial intelligence research firm OpenAI, replace the jobs of ordinary people? A new study by the company estimates that AI-powered chat technology could have a significant impact on 19% of jobs in the United States.
ChatGPT is already powerful enough to write articles and marketing materials, write computer code, and help analyze financial reports, and OpenAI researchers estimate that ChatGPT and future software tools built with its underlying technology could impact about 19% of U.S. jobs and that at least 50% of those jobs could be replaced by AI for those tasks.
At the same time, 80% of U.S. employees would be affected by ChatGPT for at least 10% of their work tasks. The study adds, “Our analysis suggests that large language models like GPT-4 are likely to have a much broader impact.”
In addition, the researchers found that higher-paying jobs may face more potential impact from AI chatbots, as these jobs may involve more software tasks. In their report, they said, “We found that jobs that rely heavily on scientific and critical thinking skills are less vulnerable, while jobs related to programming and writing skills are vulnerable to the impact.”
The OpenAI researchers used a variety of metrics to categorize which occupations would be most affected by ChatGPT. Among the most affected occupations were interpreters, translators, poets, lyricists, creative writers, PR specialists, writers, mathematicians, tax preparers, blockchain engineers, accountants, auditors, and journalists.
The report also breaks down the impact of ChatGPT by industry, with industries such as data processing hosting, publishing, and secure commodity contracts among those most likely to be disrupted. In contrast, manual labour-based industries least potentially affected, such as food service, forestry and logging, social assistance, and food manufacturing.
For this assessment, OpenAI studied more than 1,000 occupations in the United States and quantified the tasks required to complete those jobs. The researchers then used an artificial annotator and the GPT-4 model to rate whether using a ChatGPT-powered system would reduce the time required for humans to perform specific tasks by at least 50%.
The researchers acknowledge that the study’s attempt to generalize about each occupation by simply quantifying the way work tasks are performed is inherently flawed. The study adds, “It is unclear to what extent occupations can be completely broken down into distinct tasks, or whether this approach systematically ignores certain skills needed to perform these jobs competently.”
Another problem is that GPT demonstrations have shown that this technique can make obvious mistakes, including fabricating information, which makes increased human oversight necessary. And this is a factor not considered in the aforementioned study. In addition, the study only looked at whether ChatGPT could reduce the amount of time needed to complete various occupational tasks, not that it was smart enough to fully automate certain tasks.
Nonetheless, the researchers expect ChatGPT and its future iterations to dramatically change the way people work. As a result, society and policymakers need to be prepared. The researchers conclude, “The capabilities of large language models will continue to improve over time, and their growing economic impact is expected to persist even if we stop the development of new capabilities today.”
