Is AI a "Dictionary" or a "Friend"? The Distance to Generative AI as Reflected in the Ministry of Internal Affairs and Communications White Paper: A Look at the Sense of Distance to AI from a 268-Page PDF
Hello, this is Kuro-Pagu🐾
This time, I have put together a super serious summary of this year's white paper released by the Ministry of Internal Affairs and Communications.
It is a discussion about what is happening in Japan, such as the sense of distance to AI and the usage rates of AI by age group.
Because the content is difficult, I have prepared a podcast for this installment.
For those who find reading text a hassle!! I have made it so that you can understand the content just by listening to this audio guide.
Reading the text while listening will deepen your understanding even further. You can also use it as background music while multitasking.
※Main text below
What is generative AI to you?
A work tool. A substitute for searching. A dictionary to ask about words you don't understand. A friend you can talk to without hesitation. A counselor to bring your worries to.
Or, perhaps, a lover.
The "2026 Information and Communications White Paper" published by the Ministry of Internal Affairs and Communications includes a four-country survey that asked, "What kind of existence is generative AI?" The options included machine/tool, dictionary/encyclopedia, secretary/assistant, teacher/coach, friend, family member, colleague, counselor, and even lover.
For a white paper, this is a surprisingly human-like question.
1st place was "tool." The difference appeared in 2nd place.
First, let's look at the results as they are.
In Japan, the US, and Germany, the highest response rate was for "machine/tool." Japan 38.3%, US 46.0%, Germany 43.3%. Only in China was "secretary/assistant" the top choice at 44.7%.
What is interesting is what comes next.
Japan's 2nd place was "dictionary/encyclopedia" at 28.9%. The US was "friend" at 25.3%. Germany was "counselor" at 41.0%. China was "friend" at 39.9%.
Even with the same generative AI, it seems the image and positioning people hold for it are not the same.
By the way, the percentage of people who answered "lover" was 2.8% in Japan, 9.0% in the US, 6.1% in Germany, and 6.6% in China. Although they are a minority, they exist in all four countries. I never thought I would see these numbers in a Ministry of Internal Affairs and Communications white paper.
However, one cannot conclude from these results that "the Japanese are a people who think of AI as a dictionary." The total response rate exceeds 100%, as this was not a result of classifying each person into a single category. Cultural differences, how words are perceived, and response tendencies by country can also have an impact.
The white paper also touches on the possibility that as usage progresses, more people will feel that AI is a familiar presence. Nevertheless, this survey alone does not confirm a causal relationship such as "because the usage rate is high, people have come to see it as a friend."
What can be said here is that, as a response trend as of 2026, there was a certain number of people who viewed AI not just as a tool for information processing, but as a partner with whom to share emotions.
Japan's usage experience rate is 58.8%. However, there is a way to read this number.
This individual survey was an internet questionnaire conducted between January and February 2026. The number of valid responses in Japan was 1,236. The US, Germany, and China had 624 each, for a total of 3,108.
In Japan, the percentage of people who answered that they had experience using at least one generative AI service was 58.8%. Since the 2023 survey was 9.1% and the 2024 survey was 26.7%, if you just line up the numbers, it is increasing rapidly.
However, starting from the 2025 survey, 15 to 19-year-olds were newly added to the target group. The age conditions are not the same as the previous year. Even so, the value limited to 20 to 69-year-olds among the survey subjects this time was 52.2%, which exceeds the 26.7% equivalent for 20 to 69-year-olds from the previous year.
Another thing is that this 58.8% is not the percentage of people who use it every day, but a usage experience rate that includes "have used it in the past." Among the 727 Japanese respondents with usage experience, those who use it almost every day account for 11.1% for a total of one hour or more, and 18.8% for less than one hour. Combined, that is 29.9%.
In the survey, 206 people each were placed in age groups from 15 to 69. Since this is not a sample based on the actual age composition of Japan, I would also like to avoid treating 58.8% as an "estimated value for the entire Japanese population" as is.
Numbers are powerful. Precisely because they are powerful, it is necessary to look once at what the numbers are counting.
Among 15 to 19-year-olds, the usage experience is 91.7%.
When looking at the results by age, the landscape changes even more.
Among the 206 respondents aged 15 to 19 in Japan, the generative AI usage experience rate was 91.7%. For 20 to 29-year-olds, it was 78.2%. By the time you reach 30 to 39-year-olds, it is 56.3%, and the percentage decreases as age increases.
I cannot write "90% of Japanese high school students" here either. The target is 15 to 19-year-olds, and it is not a number that only surveyed high school students. At least among the respondents this time, the trend that younger age groups have a higher usage experience rate was clear.
Looking at usage scenarios, there are also some concerning numbers.
When "seeking professional advice regarding one's own mind, body, life, or future," among 15 to 19-year-olds, 8.7% said "AI only," and 26.2% said "mainly AI, with other methods as supplementary." That totals 34.9%. Even among 20 to 29-year-olds, the two combined were 32.5%.
This does not mean that one in three young people is dumping final decisions on medical care or career paths onto AI. It is the percentage of people who answered that they use AI only or mainly AI in the scenarios indicated by the question.
Even so, it is visible that there is a segment that uses generative AI not just as a substitute for a search bar, but as a partner to talk to, saying "I want you to listen for a moment" or "What do you think I should do?"
For example, a parent might open it as a dictionary for research, while a child might open it as a partner for advice on career paths or life. Although the white paper did not investigate this on a household basis, one can imagine such scenes where the distance to AI differs within the same home.
Here, I would like to touch briefly on AI usage on the corporate side, which the white paper also handles. Individual work efficiency and organizational-wide business transformation are not the same. Since this article focuses on the individual's "distance to AI," I will leave the detailed explanation for another article, but I have also included supplementary slides summarizing the points on the corporate side.
Does AI expand the power to think? Or does it take over even the parts where we should be thinking?
As the distance to AI becomes closer, the following questions arise.
How much should we rely on it? What should we entrust to it? After entrusting it, what remains within ourselves?
The Information and Communications White Paper introduces two preprints, or research papers before peer review, with different directions regarding the relationship between human memory, thinking ability, and AI.
One is research stating that groups using generative AI produced higher-quality ideas in less time than groups that did not. It is reported that positive effects were seen in knowledge sharing, speed of ideation, efficiency, and diversity of ideas, and that satisfaction and engagement with the work were also higher.
The other is an essay-writing experiment by a research team including the MIT Media Lab. Participants were divided into groups using LLMs, groups using search engines, and groups thinking for themselves without using anything, and their brain waves, writing, and post-task interviews were examined.
In this experiment, it was reported that in the group that used LLMs from the beginning, neural connections in the brain during the task were relatively weaker, and their performance in accurately citing the text they wrote immediately afterward, as well as their sense of "it being my own achievement," were also lower.
The term "cognitive debt" is used in the research title. It is a question of whether, even if the burden of thinking can be offloaded in the short term, there might be something that one's own side will have to pay later when that state is repeated.
If you only take this part out of context, you could even create a headline like "Using ChatGPT makes your brain decay."
However, that exceeds the scope confirmed by the research.
Both papers are pre-peer review and are not definitive conclusions. In particular, the latter had 54 people participate in the first three sessions, and 18 people participate up to the fourth session where conditions were changed, and the subject was also a specific task of writing an essay. It is not certain that the same results would occur for daily conversation, programming, translation, image generation, or work-related research. The fact that neural connection indicators in the brain were weaker is not the same as intelligence or ability declining in the long term.
The white paper also places this research not as a definitive conclusion on AI usage, but as an "example of research pointing to negative effects." Just before that, it also includes an example of research that increased the quality and speed of ideation.
AI robs humans of their abilities. AI expands human abilities.
One cannot generalize the influence of AI in either direction based on just two studies with different tasks.
The problem might not be "whether to use it," but "when to use it."
What is interesting about the essay experiment is the result of the group that wrote using only their own heads at first, and then used LLMs afterward. In this group, it was reported that they recalled memories better and that reactivation of neural activity was seen in a wider area.
The group that used LLMs from the beginning and the group that used LLMs after experiencing the task of writing only with their own heads had different writing processes and results.
From here on, this is not the survey result of the white paper itself, but my interpretation after reading the two research examples.
Rather than entrusting everything to AI from the very first step, first create questions or hypotheses yourself. If you get stuck there, have AI expand on them. Finally, return the facts and judgments to your own side.
In this order, wouldn't AI be more likely to become a "tool to think to places I couldn't reach on my own" rather than a "device that lets me avoid thinking"?
Of course, this is also not a procedure whose effectiveness has been proven by this white paper alone. At this point in time, it is one way of using it that can be thought of from the research results.
Do not over-trust. Do not distrust.
The 2026 Information and Communications White Paper states at the end that not only "trustworthy AI" is needed, but also the ability for humans to appropriately trust AI.
It is not about believing it completely, nor is it about rejecting it from the start.
Verify whether the AI's answer is a fact. Even if you use it as a consultant, grasp for yourself whether you have handed over the decision-making. If you had it write a text, return to a state where you can explain in your own words what is written there.
Generative AI can become a tool, a dictionary, a friend, and a counselor.
There were even a few people who answered "lover."
The distance to AI is not determined solely by the performance of the model. It also changes depending on what you open it as today and how much you entrust to it.
Where is the AI you are opening right now?
Main References
• Ministry of Internal Affairs and Communications "2026 Information and Communications White Paper" Part I p.6, p.10–17, p.24–25, p.37–41, p.57–64
• Ministry of Internal Affairs and Communications "2026 Information and Communications White Paper Summary" p.3–10
• Michael Gindert, Marvin Lutz Müller, “The Impact of Generative Artificial Intelligence on Ideation and the performance of Innovation Teams (Preprint),” arXiv:2410.18357
• Nataliya Kosmyna et al., “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” arXiv:2506.08872
Disclaimer
This article is a summary and commentary for the general public of the Ministry of Internal Affairs and Communications' "2026 Information and Communications White Paper" and published research. The survey results are based on specific samples and points in time and do not indicate estimates for the entire Japanese population or diagnoses for individuals. Both of the two studies introduced are pre-peer-review preprints and have not proven that the use of generative AI uniformly causes a decline or improvement in cognitive function. The inserted slides are summarized and illustrated to aid understanding and are not verbatim quotes from the white paper or research papers. Please check the original materials for accurate figures, conditions, and expressions. For matters requiring professional judgment such as medical, psychological, career, legal, or financial issues, do not decide based solely on AI answers; consult experts in the relevant field or public offices.
#GenerativeAI #AI #ChatGPT #InformationAndCommunicationsWhitePaper #MinistryOfInternalAffairsAndCommunications #AILiteracy #DigitalSociety #CognitiveDebt #AIUtilization #DistanceToAI