Source: The Conversation (Au and NZ)
Imagine you have used a generative artificial intelligence (AI) tool such as ChatGPT to tidy up notes you took while in a meeting. Your colleague comments on how clear they are. You don’t disclose it was the AI that made the notes clear and not you.
Now consider a different scenario. You are at your mother’s funeral. Her best friend of many years delivers a heartfelt eulogy, wishing her well in the afterlife. But later you discover her friend did not actually write the eulogy in any way – AI did.
The undisclosed use of generative AI in these two scenarios is deceptive. But is it morally wrong?
It’s worth considering this philosophical question in detail, given the rapid uptake of generative AI and the fact research has found people may be strongly incentivised to not disclose their use of generative AI because it may impact their relationships.
This is because people take generative AI outputs, generally, to be less valuable, and regard those who use the technology as less competent and authentic. Distinguishing different kinds of deception Roughly speaking, if you’re engaging in a “deceptive act”, you’re trying to get someone to believe something you know is false.
However, deception can come in different varieties. Philosopher John Danaher provides a useful framework to distinguish between three forms of deception by AI or robots. This framework is useful because, just as robots might convincingly mimic human behaviours, generative AI technologies are allowing humans to do the same.
Some deceptions involve lying or misrepresenting the external world, such as telling someone you saw a horse in the street, even though you didn’t. Danaher calls these “external state” deceptions. Other deceptions involve lying or misrepresenting facts about ourselves, such as making someone believe you’re an accomplished artist, even though the artwork you showed them was generated by AI.
Danaher calls these “superficial state” deceptions. We can also deceive others into believing we lack thoughts, feelings or competencies we actually possess, such as pretending to not understand a language we speak in order to eavesdrop.
Danaher calls these “hidden state” deceptions. So, when is deception immoral?
To go back to the first example, by not disclosing you used AI to tidy up meeting notes, you’ve allowed your colleague to believe you have the capacity to do the work, which might be true.
However, you also allowed your colleague to believe that you demonstrated that capacity by doing the work, which is false. This would fall into the category of an “external state deception”. While this deception is morally objectionable, it’s arguably ethically permissible.
Trivial deceptions like this happen all the time – and surely we aren’t obligated to always disclose everything about what we do and why we do it. For example, we don’t blink an eye at undisclosed use of spell-checking software because, in most situations, being a good speller isn’t important.
But triviality depends on context. If someone used spell-checking software at a spelling bee, we might be less forgiving. The scenario about the funeral speech is similar to the first scenario, but less trivial. Your belief over your mother’s friend’s superficial state – that they wish her well – may or may not be false.
However, your belief over the external state of affairs – that the eulogy is a demonstration of those well-wishes – is certainly false. This is more morally problematic.
When we claim an AI-generated output as our own – and particularly one that aims to reflect something about ourselves – we signal to others that we not only endorse the output, but that we have authored it.
In this case it matters to us that the friend is the source of the eulogy – and not AI. The external state deception is deceptive in a non-trival way. We imply the output was directly caused by our thoughts, feelings or competencies.
When we don’t disclose our use of AI, we deprive others of the kinds of information needed to form true beliefs about how the world is and how others are. This is wrong if people deserve to have such information.
However, the eulogy example can still show how not disclosing our use of AI can – all things considered – sometimes be permissible.
Perhaps, for example, the friend was so wracked with grief, that having AI write the eulogy on their behalf was the only way for them to get the speech written in time for the funeral.
So they endorsed the generated output as their own. In such cases, not disclosing the use of AI may be permissible, even though still a morally objectionable kind of deception. This is because, like triviality in the first scenario, other reasons might sufficiently outweigh the wrongness of non-disclosure.
Using AI more ethically So, how can people use generative AI more ethically? If we are to avoid immorally deceiving others, we ought to disclose our use of it in non-trivial cases. What makes undisclosed AI use non-trivial is malleable and context-dependent.
It changes in response to social norms couched within particular social practices.
Being open about AI use gives others the opportunity to form more accurate beliefs about what we are communicating – and the internal states our communication demonstrates.
Siavosh Sahebi is supported by an Australian Government Research Training Program scholarship.
Thomas Montefiore receives funding from the Australian Government through the Australian Research Council.
Original source: https://analysis1.mil-osi.com/2026/06/04/people-are-using-ai-to-communicate-without-disclosing-it-is-this-morally-wrong/