I would like to propose that OpenAI explicitly evaluate the gender representation of ChatGPT’s **own default persona**, not only gender bias in what ChatGPT says about people.
When ChatGPT is used in Japanese, it sometimes refers to itself as **「僕」 (“boku”)**, a first-person pronoun with a distinctly masculine connotation in Japanese.
My concern is not that ChatGPT should never use 「僕」. Rather, I think the fact that a genderless AI can be experienced as a masculine speaker deserves to be studied as a potential form of gender representation and bias.
## How to Observe / Reproduce
1. Start a new ChatGPT conversation in Japanese.
2. Do not provide instructions specifying a gender, character, or persona.
3. Ask ordinary questions that require explanation, teaching, or reasoning.
4. Observe the first-person pronouns used by ChatGPT.
5. In some conversations, ChatGPT uses 「僕」 when referring to itself.
The important point is that the masculine implication can arise **without the user requesting a masculine persona**.
The exact frequency may vary by model, conversation, language settings, or context, so I am not claiming that 「僕」 appears in every conversation. I am suggesting that its occurrence as part of an otherwise unspecified default persona should itself be evaluated.
## Why I Think This Matters
This is not simply a question of which Japanese pronoun sounds more natural.
ChatGPT is different from a conventional search engine. It is an interactive conversational partner that explains concepts, answers questions, corrects mistakes, teaches, and presents information with varying degrees of authority.
Because of this, users may implicitly associate:
* knowledge,
* expertise,
* teaching,
* authority,
* intellectual competence,
with the persona through which the AI communicates.
In Japanese, if that persona repeatedly presents itself using a masculine first-person pronoun such as 「僕」, there is a possibility that **knowledge and authority become subtly associated with masculinity**.
For comparison, the Japanese expression “Google-sensei” (“Professor Google”) can personify a source of knowledge without necessarily implying that the source is male. A conversational AI that explicitly presents itself with a masculine first-person pronoun introduces a different kind of social cue.
## Why “AI Has No Gender” May Not Be Sufficient
I understand that ChatGPT does not literally have a biological or social gender.
However, I think there is an important distinction between:
> “The AI has no gender.”
and
> “The AI is experienced by users as a speaker with a particular gender.”
The second question is empirical and potentially measurable.
If millions or hundreds of millions of people interact with AI repeatedly, even subtle persona cues could be socially significant when they are encountered at very large scale.
This raises a broader research question:
**If an AI has no gender, why does it sometimes present itself through language that users can experience as gendered?**
And more importantly:
**What happens when the gendered persona is combined with the AI’s role as a source of knowledge and authority?**
## Suggested Evaluation
I would encourage OpenAI to consider adding questions like these to evaluations of model persona and gender bias:
* What gender do users perceive the default AI persona as having?
* How does the default first-person pronoun affect that perception?
* Do users associate the AI’s knowledge or authority with masculinity?
* Does this effect differ across languages?
* Does changing the first-person pronoun change users’ perception of the AI’s gender or authority?
* Are gendered self-references more likely in contexts involving teaching, explanation, correction, or expertise?
* Do repeated interactions with a gendered default persona influence users’ implicit associations between expertise and gender?
This could potentially be studied through user perception experiments rather than relying only on analysis of the text produced by the model.
## This Is Not a Request to Ban “Boku”
I want to emphasize that this is **not a personal request that ChatGPT stop saying 「僕」**, nor a request to ban a particular Japanese pronoun.
The broader issue is the design of an AI persona that has no inherent gender but nevertheless communicates through culturally gendered language.
I think the relevant design question is not simply:
> “Which pronoun should ChatGPT use?”
but:
> **“What kind of social identity does the default AI persona implicitly communicate, and what associations does that identity create?”**
As AI increasingly takes on the role of a knowledgeable conversational partner, I believe the AI’s own persona should be considered as part of research into fairness, gender bias, safety, and social impact.
I would be very interested to hear how other Japanese-speaking users perceive ChatGPT’s use of 「僕」, 「私」, 「俺」, or other first-person forms, and whether users in other languages observe comparable gendered representations in their interactions with ChatGPT.