Anyone following the AI industry will be familiar with ambitious claims that AI models will soon outperform humans at every task and replace large segments of the workforce. However, a new Google Research study examining how workers actually use Gemini found no evidence supporting claims that AI is about to trigger widespread automation and the displacement of white-collar jobs.

Published last week, the study introduces the AI & Economy ATLAS, an analysis of activities, tasks, the occupational landscape and adoption patterns based on 15 million anonymized AI interactions through the Gemini app, Google's AI Mode and the Gemini API.

The initial analysis found that although AI is used significantly across a wide range of occupations, its use remains limited in depth and predominantly collaborative, while the complete automation of tasks has a restricted scope.

The researchers concluded that AI appears useful for only a subset of tasks. To reach their findings, the Google team used an automated classifier to categorize work-related AI interactions according to the Bureau of Labor Statistics' standard occupational classifications and O*NET's more detailed database of specific work activities.

Although the method required probabilistic classification of interactions that are inherently uncertain, reviews conducted by human evaluators found it to be a reliable indicator of how Gemini prompts were being used in the workplace.