In this article, I would like to record my discomfort with a common assumption often repeated in the media:
that generative AI is something that always affirms the user.
In My Case
As someone who documents cases involving copyright misuse and rights-related issues,I use generative AI to help remove emotional expressions and keep my writing focused strictly on factual information.
I also rely on generative AI when drafting emails, asking questions such as:
• Is it appropriate to send this message?
• Could the recipient misunderstand it?
In some cases, the AI even responds that I should not send the message at all.
For me, there is no value in using generative AI unless it clearly tells me when something is not acceptable.
In Other Cases
I have had opportunities to observe how other people use generative AI.Even without explicitly setting it to be overly affirmative, their usage often resembles:
• An extension of search
• Avoiding questions they do not want to hear answers to
As a result, generative AI may appear to them as something that always gives pleasant, agreeable responses.
In other words, it may not be that generative AI always affirms everything,
but rather that many users avoid asking questions that could produce uncomfortable answers in the first place.
It Depends on the User
Even though we casually refer to “generative AI” as a single concept, there is a clear difference between:• Users who expect the AI to say “no” when necessary
• Users who simply avoid questions that might lead to inconvenient or unpleasant responses
These users are effectively seeing entirely different realities.
From this perspective, I find the blanket explanation
“generative AI always affirms everything”
to be an oversimplified and somewhat careless description.