Is it ethically problematic to present text written by AI as one's own words?

Generative AI like ChatGPT can quickly prepare meeting notes, emails, and speech drafts, but if you don't disclose that you used AI and present the resulting text as your own, it can lead to misunderstandings. Syavosh Sahebi and Thomas Montefiore, who study philosophy at Macquarie University in Australia, discuss the problems that can arise when you don't disclose that you're using AI.

People are using AI to communicate without disclosing it. Is this morally wrong?
https://theconversation.com/people-are-using-ai-to-communicate-without-disclosing-it-is-this-morally-wrong-283773

Sahebi and his colleagues first cite a scenario where a generative AI like ChatGPT is used to summarize notes taken during a meeting. For example, if a colleague praises the summarized notes as 'easy to understand,' but the notes were actually compiled by an AI rather than the person who took them, the colleague might assume that the person who took the notes compiled them clearly if the colleague is not informed that an AI was used.

Another example given is a funeral eulogy. Imagine a long-time friend reading a heartfelt eulogy at the funeral of the deceased, expressing wishes for peace in the afterlife. However, if it is later discovered that the friend did not write the eulogy at all, but that an AI wrote the text, the recipient's impression would change dramatically.

In both examples, the act of not disclosing that AI is being used leads to misunderstandings, but Sahebi et al. believe that 'the severity of the problem with such actions varies depending on the situation,' and they explain the background behind why people might want to hide the fact that they are using generative AI.

According to Sahebi et al., the reason people want to hide the fact that they are using generative AI is the fear that 'revealing that they used AI may affect their relationships.' Previous research has shown that 'people who do not reveal that they are using generative AI are motivated by a desire to avoid affecting their relationships,' that ' disclosure that a text was written with AI assistance lowers the evaluation of the text,' and that 'AI-mediated apologies affect the evaluation of the sincerity of the person apologizing and whether the recipient will forgive the apology.'

Furthermore, research has reported that 'people who use AI tend to be perceived as having lower abilities.'

Research suggests that using AI makes you more susceptible to harsh criticism from others - GIGAZINE

Thus, whether or not to reveal that you are using generative AI involves interpersonal relationships and reputation, and is not merely a matter of explaining work procedures. To clarify what kind of deception it constitutes to not tell someone that the contents of a memo were organized by AI, or to read a eulogy written by AI as if it were your own words, Sahebi et al. refer to philosopher John Danaher's ' Classification of Deception by AI and Robots .' Here, deception is the act of trying to make someone believe that you are lying about something you know to be false.

Danaher categorizes AI and robot deception into three types. The first is 'external state deception,' which misleads the other party about the outside world, such as saying 'I saw a horse on the street' when in fact they didn't see one. The second is 'superficial state deception,' which misleads the other party about themselves, such as showing an AI-generated picture as their own work and making them think 'this person is a great artist.' The third is 'hidden state deception,' which makes it appear as if they don't possess the thoughts, feelings, or abilities they actually do, such as pretending not to understand a language they actually understand in order to eavesdrop.

According to Sahebi et al., applying the above classification, the example of meeting notes falls under 'deception of external conditions.' Those who don't disclose the use of AI are leading their colleagues to believe they have the ability to 'organize notes clearly.' While this ability itself may be genuine, the fact that they are leading others to believe they 'actually used that ability to organize the notes' is a misrepresentation. However, Sahebi et al. state regarding the meeting note example that 'minor deceptions happen daily, and people aren't always obligated to reveal everything they did and how they did it. Therefore, while it's an ethically problematic deception, it's not always so serious as to be unacceptable.'

As an example of how the importance of a problem changes depending on the context, Sahebi et al. cite spell-checking functions. In situations where the content of the text is important, whether or not you corrected the spelling yourself is not a major issue, so in most cases, it's not a problem to not tell the other person that you used a spell-checking function. However, it would be a problem if you used a spell-checking function in a test where you have to correctly spell English words. Thus, even with the same tool, the importance of not disclosing that you used it changes depending on the context in which it is used.

Regarding the example of a funeral eulogy, Sahebi et al. explain that it is 'similar to the example of meeting notes, but not something that can be taken lightly.' While it is unclear whether the interpretation that 'a friend wishes peace for the deceased' is incorrect, the interpretation that 'the eulogy was written by the friend himself and expressed the friend's wishes' is clearly not true. Sahebi et al. point out that when someone presents AI-generated text as their own, they are not only acknowledging its content but also indicating that 'they wrote that text themselves.'

According to Sahebi et al., in a eulogy, not only the content of the words but also the fact that 'a long-time friend wrote it in their own words in remembrance of the deceased' is meaningful, so if it is not revealed that AI was used, it becomes difficult for the recipient to correctly determine whose thoughts and feelings the eulogy originated from. However, Sahebi et al. do not believe that 'hiding the use of AI is always unacceptable.' For example, if the friend was too deeply saddened to write the eulogy and had an AI write it to get it done in time for the funeral, it could be argued that they accepted the AI-generated text as their own and read it aloud. In such cases, Sahebi et al. say that 'it may be acceptable depending on the circumstances.'

Sahebi and his colleagues argue that disclosing the use of AI allows the other party to more accurately judge the content of the words and the speaker's thoughts, feelings, and abilities, stating that 'it is important to disclose the use of AI in situations where there is a risk of serious misunderstanding.' However, they note that what constitutes a 'situation where there is a risk of serious misunderstanding' will vary depending on the social norms and communication context of that situation.