Particle.news
Download on the App Store

Google Releases ATLAS v1.0 Mapping 15 Million Gemini AI Interactions

The de-identified dataset aims to give researchers, policymakers, and businesses a clearer empirical view of how people use AI across tasks, occupations, and countries.

Overview

  • On Thursday Google published ATLAS v1.0, an initial release built from 15 million aggregated, de-identified interactions drawn from the Gemini App, AI Mode, and the Gemini API covering more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks.
  • ATLAS finds the bulk of AI use happens outside formal work, with over 86% of interactions linked to household activities, errands, or personal tasks that standard productivity measures often miss.
  • At work, adoption is wide across sectors but shallow within jobs: the typical occupation uses AI for about 21% of tasks and fewer than 10% of interactions fully automate a task.
  • The dataset shows AI is assisting manual and technical trades as well as white‑collar roles, with workers like auto technicians using multimodal AI for diagnostics, troubleshooting, and on‑the‑job learning.
  • Google says ATLAS v1.0 is de-identified and processed using DeepMind’s OCTO taxonomy with additional automated privacy layers, and the company presents ATLAS as an ongoing research resource for follow-up study and policy use.