工作前沿：AI如何拓展人们在工作中所做的事情

Work at the Frontier:How AI is expandingwhat people do at work Key takeaways 01AI may change not only how work gets done, but who does it. Our new researchsuggests that ChatGPT users frequently seek help with tasks historicallyassociated with other occupations, suggesting AI may broaden roles and changehow work is divided. 02Across our sample, 16.8 percent of all work-related messages—and 43.5 percent ofoccupation-specific messages—concern tasks historically associated with anotheroccupation. 03Patterns of AI use differ across occupations. Marketing and engineering taskstravel to many workers in other occupations, while workers in design, sales, andhuman resources are more likely to take on a variety of tasks from other jobs. 04These effects are more pronounced at smaller firms. Typical users at smallerworkspaces do relatively more cross-occupation work. This could be a sign thatusers turn to AI as a tool to expand the types of work they do when they have fewerresources. Introduction How is AI changing how we work? Our analysis of work-related messages sent by ChatGPT usersprovides a window into the changing task composition of jobs. We find that 43.5 percent of work-related,occupation-specific messages concern tasks historically associated with another occupation.1 In ourAI Jobs Transition Framework, we highlight that a significant portion of jobs are likely to reorganize,that is, their day-to-day tasks may shift considerably. OurWork at the Frontierseries is aimed atunderstanding this process. As more and more workers use AI in their jobs, new combinations ofindividual and machine activity that weren’t previously possible emerge. Understanding these changesis at the heart of understanding how work adapts to AI: it changes what a worker does alone, what theydelegate to a colleague, and what requires a specialist. With AI, a salesperson who once handed a customer dataset to an analyst can now easily and quicklyexplore it themselves. A marketer who once waited for a software or web developer can now troubleshoota website or write a simple script. In each case, AI changes both how quickly work is completed and whodoes it. We examine how AI expands the types of work people do, especially when it is not within theirtypical job description. We find that AI may change the historical or typical boundaries of what workersdo within a certain role. Anoccupational boundaryis the set of activities historically associated withan occupation. We examine whether workers use AI for activities within oroutside that boundary. We focus on users in eight occupation groups: customer experience, design, engineering, finance, humanresources, legal, marketing, and sales. We use O*NET, the U.S. Department of Labor’s database ofoccupations and work activities, to identify the tasks traditionally associated with each occupation group.We use a random sample of work-related messages from these users and classify each message to asingle O*NET detailed work activity (DWA) that most closely matches its task content. We then comparethe occupations associated with that activity with the sender’s stated occupation.2Occupationalboundaries are not always precise because generic tasks like writing, summarizing, scheduling, andmany other activities appear across many types of jobs. We therefore assign messages to one of threecategories: Generic, Within occupation, and Cross-occupation. In this random sample of work-related messages: • 16.8 percent of messages are classified as Cross-occupation, meaning they concern tasks historicallyassociated with another occupation;• 21.8 percent are classified as Within occupation, meaning they concern tasks historically associatedwith the user’s own occupation;• and 61.5 percent are classified as Generic, meaning they relate to tasks that are broadly shared acrossoccupations like writing emails or scheduling meetings. Among non-generic messages, 43.5 percent are cross-occupation. Excluding generic work like writingemails, this category accounts for 28 to 77 percent of occupation-specific messages. Source: OpenAI Economic Research analysis of a random sample of work-related messages on individual ChatGPT accountsfrom U.S. users whose self-reported occupations were linked from ChatGPT Business account information, with tasks mappedto O*NET DWA descriptions. Notes: Occupation-specific shares exclude generic work and rebase the remainder to 100%. We call this patterntask crossover: work historically associated with one occupation appearing in the AIuse of people in another. Across the eight occupation groups, between 11 and 30 percent of messagesinvolve task crossover. Task crossover differs between different types of workers. Marketing and engineering work appearswidely in the AI use of workers in other fields. Meanwhile, workers in fields like design, sales, and humanresources are more likely to use AI for tasks outside of their typical roles. Task crossover is more commonin smaller workplaces,