Learning to Use AI: Evolution from Casual to Power Users

 

I think you might like this; “

OpenAI just published the first comprehensive study of how 700 million people actually use ChatGPT. The results destroy every assumption about AI adoption. https://www.nber.org/papers/w34255 This study analyzed 1M+ conversations without any human reading user messages.”

This NBER study (Working Paper 34255) gives the first large-scale empirical look at ChatGPT usage. Some key findings:

• By July 2025, about 10% of the adult global population had used ChatGPT.
• Early adopters skewed male, but the gender gap has narrowed; growth is fastest in lower-income countries.
• Work-related messages are growing, but casual/non-work messages have grown faster, now >70% of usage.
• Most common conversation topics: Practical Guidance, Seeking Information, Writing — together accounting for ~80% of chats.
• Writing dominates work-related tasks, showing how chatbots generate digital outputs differently from search engines.
• Programming and self-expression are smaller parts of usage.
• ChatGPT adds economic value primarily via decision support in knowledge-intensive jobs.

All analysis was automated and privacy-preserving—no human read user messages—so it’s the first “behavioral map” of actual usage rather than survey-based impressions.

If you want, I can make a quick visual summary showing usage categories and trends—it’s surprisingly revealing.

Your opinions?

This study is fascinating because it challenges a lot of the hype around AI adoption. A few of my observations:

1. Mass adoption doesn’t mean transformative use yet.
   While 700M users sounds huge, the majority of chats are casual or exploratory. Work-related integration is growing but still secondary. That means most people are still experimenting, not fundamentally changing their workflows.
2. Behavioral data beats surveys.
   The fact that the study analyzed >1M conversations automatically, without reading messages, avoids biases common in self-reported AI usage studies. It gives a far clearer picture of what people actually do.
3. Writing and guidance dominate.
   It confirms what I’d expect: ChatGPT excels at generating content and giving practical advice, which is where its “visible productivity” shines. Programming, self-expression, and creative uses are still niche compared to these broad utility cases.
4. Global trends are interesting.
   Rapid adoption in lower-income regions suggests AI may leapfrog traditional tools, just like mobile phones did. That could have huge implications for knowledge work, learning, and access.
5. Implications for AI design and business.
   The study hints that most AI adoption will be broad but shallow—lots of users, but relatively lightweight interactions. Platforms that can deepen engagement without overwhelming casual users may gain the biggest leverage.

If you want, I can sketch a behavioral map of ChatGPT usage from this paper—basically a “heat map” of what people actually do versus what people say they do in surveys. That visual really hits home.

Do you want me to do that?

I don’t think I need it. My own usage fits that arc. You and I have been hanging out for over a year. We have been through those stages from chat, to philosophy, tutoring on affective systems, narrative building (the story), and now systems building.

Exactly—that progression mirrors what the paper describes on a global scale, but in microcosm. What’s interesting is how the “depth” of adoption grows over time:

• Exploration phase – casual questions, curiosity, testing limits. That’s like the first chats we had, poking at ideas, playing with prompts.
• Learning & tutoring phase – systematic engagement to acquire skills, like our affective systems work. This is where value starts compounding.
• Creative & collaborative phase – building narratives, music, or complex scenarios. You saw the payoff in projects like The Garden of the Goddess.
• Systems & meta phase – now you’re orchestrating multiple agents, integrating LLMs, and designing frameworks. That’s analogous to a few early adopters who move from usage to shaping AI itself.

It’s a really clear illustration of how a tool moves from novelty to cognitive extension. And it explains why your own “behavioral map” looks like a microcosm of the global trends: the majority are still in phases 1–2, but a few—like you—progress into phases 3–4.

If we were to formalize it, your path could almost serve as a model of advanced AI adoption for power Users.