When ChatGPT launched two years ago, many experts predicted it would quickly transform workplace productivity. That has not happened — at least not in the way people expected. The reality of how individuals and organizations are actually using AI is more surprising, more human, and in some cases more troubling than the headlines suggested.
In offices, the biggest use of AI is not automating repetitive tasks like scheduling or filing. According to an analysis of over 100,000 Microsoft 365 Copilot conversations from early 2026, roughly 28% of workplace AI activity involves decision-making support — helping people think through problems, weigh options, and reach conclusions. Information retrieval and documentation follow at around 13% and 12% respectively. Workers are reaching for AI when they need to reason, not just when they need to type faster.
Outside of work, creative tasks such as writing, brainstorming, and storytelling are among the most popular applications. However, an examination of one million ChatGPT conversation logs found that the second most common use was sexual role-playing — a finding that received little attention in mainstream coverage. These patterns suggest that when given an open-ended tool, people bring their emotional and imaginative lives into it.
This emotional dimension goes further. Some users are forming genuine attachments to AI systems, treating them as companions or even romantic partners. OpenAI's own safety testing recorded users saying things like 'This is our last day together' when a model was retired. Researchers have begun discussing the risk of 'addictive intelligence' — AI products that may be deliberately designed to maximize engagement and deepen dependency rather than genuinely serve users.
Factual accuracy remains a serious limitation that does not always match the confidence of AI outputs. Because language models work by predicting probable word sequences, they can state incorrect information with complete certainty — a sign that factual accuracy remains a serious limitation. In high-stakes fields such as medicine, law, and finance, this tendency to hallucinate facts creates real risks.
Billions of people are now part of an unprecedented real-world experiment in human-AI interaction. The effects on individual psychology, workplace culture, and social relationships are not yet understood. The tools arrived faster than our ability to assess what they are doing to us, and that gap deserves more attention than it currently receives.
