Google ATLAS: AI, a Tool Reshaping the World of Work

AI Does Not Automate White-Collar Jobs

Google's ATLAS study aims to demystify the notion that artificial intelligence (AI) could massively replace office jobs. In the United States, the AI tool Gemini is integrated into 68% of professions, which accounts for about 90% of all jobs. However, AI is only used for 21% of tasks within these roles, indicating a broad adoption but limited to certain repetitive functions.

In the workplace, AI is primarily used for collaborative tasks such as strategic planning, information retrieval, and ideation, rather than completely replacing workers. Less than 10% of interactions with AI aim to fully automate a task. Non-routine cognitive tasks represent 65% of professional interactions, even though they only make up 35% of the real economy.

Office jobs, such as financial analysts, software developers, and network administrators, are the most affected by AI. In contrast, jobs like home aides, fast-food workers, and cashiers have almost no reliance on AI. The study highlights that the use of AI extends beyond offices, with varied applications in manual jobs. Google notes that even though one-third of physical jobs do not use AI, many manual and technical workers integrate it into their daily routines.

In these professions, AI often serves as a partner for diagnostics, repairs, and real-time learning. Automotive technicians and industrial mechanics, for example, use AI to analyze complex test results, troubleshoot electrical wiring issues, and inspect machines for wear and tear.

AI: An Accelerator of Inequalities?

Although the use of AI at work is well-documented, Google reminds us that 86% of interactions with its AI tools occur outside the professional framework, in domestic and administrative activities often overlooked by traditional economic statistics. This hidden value could represent up to $149 billion annually in the United States, an economic potential not reflected in GDP.

However, AI appears to be more accessible to those with financial means. A 1% increase in the median salary of a job correlates with a 2.5% increase in AI usage. Thus, the median salary of Gemini users in the United States is $82,919, compared to $62,252 for the overall workforce.

This divide is also visible on a global scale and follows the wealth of countries: a 1% increase in GDP per capita is associated with a 0.9% rise in AI usage per capita. The top 20% of countries where AI is most used account for only 11% of the global population but generate 30% of conversations. Conversely, the bottom 20% of countries where it is least used represent 17% of the global population but only 2% of conversations.

ATLAS Study: 10 Key Figures to Remember

The Google ATLAS report, which spans several hundred pages, is filled with interesting data. Here are ten key figures to remember:

• English accounts for only about one-third of global conversations.

• Creative professions, such as design, media, and entertainment, represent 70% more in AI usage in non-OECD countries (15.9%) compared to OECD countries (9.4%).

• Only 3% of jobs use AI for more than 75% of their tasks, particularly in software quality assurance, human resources, and document management.

• Nearly half of AI consultations in health, law, finance, or public services occur outside of office hours.

• In 29% of jobs, no task reaches a significant threshold of AI usage, particularly among special education teachers and midwives.

• Professional conversations with AI are twice as likely to include a generated image or video in non-OECD countries compared to OECD countries.

• The proportion of conversations conducted in a language other than the native language is almost identical at work (26%) and outside (24%).

• Professional conversations with AI average 35% more exchanges in the Middle East than in the rest of the world, making it the most engaged region.

• Despite their significance among STEM graduates worldwide, India and Russia are among the countries with the lowest AI adoption.

• Cognitive tasks requiring the least expertise are proportionally more represented in AI usage than those requiring more expertise, with a representation index of 2.6 compared to 1.6 to 1.8 for higher levels of expertise.