Microsoft Research summarizes a recent paper, Working with AI: Measuring the Occupational Implications of Generative AI, which mapped real user interactions with generative Artificial Intelligence chatbots to occupational tasks. The team analyzed anonymized Bing Copilot conversations from January to September 2024 and compared the activities users sought help with to task descriptions in the O*NET database. The stated goal was to identify where Artificial Intelligence may be useful as a tool across occupations rather than to predict job loss.
The study found that Artificial Intelligence chatbots showed strongest applicability to knowledge work and communication tasks, including writing, gathering information, and learning. Occupations that rely heavily on those tasks may benefit from treating Artificial Intelligence as a workflow aid. Conversely, the paper observed lower direct chatbot applicability for physical tasks such as performing surgeries or moving objects. The authors emphasize that applicability scores indicate potential to assist with subtasks and not the ability to perform whole occupations.
The post reiterates the paper’s methodological limitations. O*NET task lists do not capture interpersonal judgment, domain expertise, ethical considerations, or full contextual nuance of work. The dataset reflects only Bing Copilot usage and may be shaped by user awareness, access, or comfort with tools, and the analysis cannot reliably distinguish conversations conducted for work versus leisure. The study also focused solely on chatbot interactions and did not evaluate other forms of Artificial Intelligence. The authors explicitly cautioned against using their findings to infer job elimination and call for continued, nuanced research into how Artificial Intelligence can complement human strengths and inform skilling and policy responses.
