While using ChatGPT, workers wonder: Will this take my job?
In December, the staff of the American Writers and Artists Institute — a 26-year-old membership organization for copywriters — realized that something big was happening.
The newest edition of ChatGPT, a “large language model” that mines the internet to answer questions and perform tasks on command, had just been released. Its abilities were astonishing — and squarely in the bailiwick of people who generate content, such as advertising copy and blog posts, for a living.
“They’re horrified,” said Rebecca Matter, the institute’s president. Over the holidays, she scrambled to organize a webinar on the pitfalls and potential of the new artificial intelligence technology. More than 3,000 people signed up, she said, and the overall message was cautionary but reassuring: Writers could use ChatGPT to complete assignments more quickly, and move into higher-level roles in content planning and search-engine optimization.
“I do think it’s going to minimize short-form copy projects,” Matter said. “But on the flip side of that, I think there will be more opportunities for things like strategy.”
OpenAI’s ChatGPT is the latest advance in a steady march of innovations that have offered the potential to transform many occupations and wipe out others, sometimes in tandem. It is too early to tally the enabled and the endangered, or to gauge the overall impact on labor demand and productivity. But it seems clear that artificial intelligence will impinge on work in different ways than previous waves of technology.
The positive view of tools like ChatGPT is that they could be complements to human labor, rather than replacements. Not all workers are sanguine, however, about the prospective impact.
Katie Brown is a grant writer in the Chicago suburbs for a small nonprofit group focused on addressing domestic violence. She was shocked to learn in early February that a professional association for grant writers was promoting the use of artificial intelligence software that would automatically complete parts of an application, requiring the human simply to polish it before submitting.
The platform, called Grantable, is based on the same technology as ChatGPT, and it markets itself to freelancers who charge by the application. That, she thought, clearly threatens opportunities in the industry.
“For me, it’s common sense: Which do you think a small nonprofit will pick?” Brown said. “A full-time-salary-plus-benefits person, or someone equipped with AI that you don’t have to pay benefits for?”
Artificial intelligence and machine learning have been operating in the background of many businesses for years, helping to evaluate large numbers of possible decisions and better align supply with demand, for example. And plenty of technological advancements over centuries have decreased the need for certain workers — although each time, the jobs created have more than offset the number lost.
ChatGPT, however, is the first to confront such a broad range of white-collar workers so directly, and to be so accessible that people could use it in their own jobs. And it is improving rapidly, with a new edition released last month. According to a survey conducted by the job search website ZipRecruiter after ChatGPT’s release, 62% of job seekers said they were concerned that artificial intelligence could derail their careers.
“ChatGPT is the one that made it more visible,” said Michael Chui, a partner at the McKinsey Global Institute who studies automation’s effects. “So I think it did start to raise questions about where timelines might start to be accelerated.”
That’s also the conclusion of a White House report on the implications of AI technology, including ChatGPT. “The primary risk of AI to the workforce is in the general disruption it is likely to cause to workers, whether they find that their jobs are newly automated or that their job design has fundamentally changed,” the authors wrote.
After decades of study, researchers understand a lot about automation’s impact on the workforce. Economists including Daron Acemoglu at the Massachusetts Institute of Technology have found that since 1980, technology has played a primary role in amplifying income inequality. As labor unions atrophied, hollowing out systems for training and retraining, workers without college educations saw their bargaining power reduced in the face of machines capable of rudimentary tasks.
The advent of ChatGPT three months ago, however, has prompted a flurry of studies predicated on the idea that this isn’t your average robot.
One team of researchers ran an analysis showing the industries and occupations that are most exposed to artificial intelligence, based on a model adjusted for generative language tools. Topping the list were college humanities professors, legal services providers, insurance agents and telemarketers. Mere exposure, however, doesn’t determine whether the technology is likely to replace workers or merely augment their skills.
Shakked Noy and Whitney Zhang, doctoral students at MIT, conducted a randomized, controlled trial on experienced professionals in such fields as human relations and marketing. The participants were given tasks that typically take 20 to 30 minutes, like writing news releases and brief reports. Those who used ChatGPT completed the assignments 37% faster on average than those who didn’t — a substantial productivity increase. They also reported a 20% increase in job satisfaction.
A third study — using a program developed by GitHub, which is owned by Microsoft — evaluated the impact of generative AI specifically on software developers. In a trial run by GitHub’s researchers, developers given an entry-level task and encouraged to use the program, called Copilot, completed their task 55% faster than those who did the assignment manually.
Those productivity gains are unlike almost any observed since the widespread adoption of the personal computer.
“It does seem to be doing something fundamentally different,” said David Autor, another MIT economist, who advises Zhang and Noy. “Before, computers were powerful, but they simply and robotically did what people programmed them to do.” Generative artificial intelligence, on the other hand, is “adaptive, it learns and is capable of flexible problem solving.”
There’s another takeaway from the initial research: ChatGPT and Copilot elevated the least experienced workers the most. If true, more generally, that could mitigate the inequality-widening effects of artificial intelligence.
On the other hand, as each worker becomes more productive, fewer workers are required to complete a set of tasks. Whether that results in fewer jobs in particular industries depends on the demand for the service provided, and the jobs that might be created in helping to manage and direct the AI. “Prompt engineering,” for example, is already a skill that those who play around with ChatGPT long enough can add to their résumés.
Since demand for software code seems insatiable, and developers’ salaries are extremely high, increasing productivity seems unlikely to foreclose opportunities for people to enter the field.