AI adoption far outpaces that of the early Internet — report sheds light on worldwide AI penetration and usage patterns

Saying that AI is everywhere is an axiom by now, but a question that begets answering is “how much, exactly?” That’s what the Financial Times (FT) set out to discover, by collecting data and interviewing execs from popular consumer-facing AI services. While most findings are expected, there are a handful of quite surprising pieces of information, including but not limited to the growth of adoption and shifts in usage patterns. As a minor spoiler, the rate of AI adoption is far outpacing that of even the Internet, as fast as that seemed in the early 2000s.

To illustrate that exact point, OpenAI notes that about one in ten people worldwide have already used ChatGPT in some capacity. Over the span of only three years, the service went from zero to close to 800 million users, a feat that took the Internet some 13-odd years. While there’s an easy argument to be made that chatbots have it easy by relying on existing infrastructure (the Internet itself), a comparative ratio of over 4:1 is nevertheless shocking and potentially unseen in the modern world.

The use cases for AI bots are shifting rapidly, too. ChatGPT data shows the split between personal- and work-related messages was already a 53/47% split in mid-2024, and it’s now shifted to 73/27%, clearly highlighting that users are integrating the bot into their daily routines. For its part, Meta says that its AI assistant, recently embedded into WhatsApp, Instagram, and even Ray-Ban glasses, has reached one billion monthly users.

You may like

As for work-focused tools, Google remarks that its AI overviews on search results pages have 2 billion monthly users worldwide, and that 70% of people use the writing helpers in Google Docs and Mail when suggested to do so. As expected, most conversations with Anthropic’s science-focused Claude revolve around computers and programming, while Microsoft’s Copilot has hit 100 million monthly active users.

The geographical patterns of AI bot usage are mostly predictable, but with some unexpected bits. The countries with the highest percentage of AI tool usage among internet-enabled populations are actually Singapore and the United Arab Emirates, at close to 60% each. The latter country has put significant effort into becoming an AI powerhouse and even has AI classes for children from the age of four; meanwhile, the UAE gives ChatGPT Plus to its citizens for free. Most of Europe is in the next tier, from 45% down to 27%, approximately, followed by the United States. FT quotes research from Microsoft noting “a strong correlation between AI User Share and GDP”. The worldwide average rests at about 28%.

Interestingly enough, FT’s report notes that ‘shadow AI’ — employees using AI tools without explicit company knowledge or approval — has grown substantially. MIT research posits the main reason is that corporate AI deployment initiatives often remain stalled in pilot phases, with companies having organizational trouble integrating AI tools into their workflows, or a general distrust in the potential productivity gains. The FT research does note that the media and technology industries have indeed “clear signs of structural disruption from AI”, as anyone witnessing the gigantic layoffs in FAANG-type companies can attest.

When it comes to the usage pattern for task completion, most of the data points towards AI tools being used as assistants rather than full-blown replacements. Anthropic’s data does show that Claude is increasingly being used for automating entire tasks, although Peter McCroy, the company’s head of economics, states the underlying reason for the shift is still hard to pin down, and lists both the model’s increased capabilities and users’ improved ability to describe tasks as possibilities.

While on that particular topic, FT lists several research efforts showing that over-reliance on AI tools can result in diminished reasoning skills, as many suspected. Data scientist Austin Wright aptly compares the phenomenon to the lower overall sense of direction following the widespread usage of GPS. An MIT experiment focused on essay split three ways between no AI, AI-assisted writing, and full AI writing concluded the latter group could suffer “cognitive debt” and a “likely decrease in learning skills”. Reportedly, LLM users had a hard time quoting their own “work” and didn’t identify themselves much with it.

Google Workspace VP of product notes that Gemini and similar AI assistants are meant to “amplify users’ work”, while OpenAI announced a ‘study mode’ in July that offers step-by-step instructions rather than immediate problem solving, in a bid to help students with reasoning. The aforementioned MIT research does point out that users should rely on themselves for initial drafts and ideas, and use the bots as an assistant, rather than having the bot do it for them, as that can become a habit that’s hard to break.

AI adoption far outpaces that of the early Internet — report sheds light on worldwide AI penetration and usage patterns

Tags: