I think this is one of those things where the root cause is a social problem rather than a technical one, and trying to use technical solutions is somewhat helpful at best and masking huge issues at worst. If people can't write professionally, then the proper solution is better education, perhaps some education through onboarding in the job, and/or the boss being more flexible when reading. At least the common usage of spell checkers I see don't meaningfully change the tone of the text. The LLM-powered spell checker, akin to a different human writing the email for the employee, is unacceptable. It has perverse incentives and outcomes. Minor touch ups are one thing, but at some point it becomes deception.
I think there is only deception if you believe the original words written in the original tone was the intended message (or I suppose if the intent of the communication is to demonstrate your personal ability to write in a given style). If a coworker does something extremely stupid that harms our project and I sit down at my desk and write an angry email full of invective and spit and fury, save it in my drafts and go for a walk, then come back and rewrite the email to be constructive and professional, have I been deceptive? When I started the email I certainly intended to write the things I wrote in that first draft. But sending that would have been counter productive.
I might even still think my co-worker is a flaming moron who shouldn't be allowed out of the house unsupervised. But if I know that sending that in an email isn't going to solve anything and just make things worse, am I being deceptive if I remove that sentence from the email?
Or consider an alternative scenario. I attended a conference where a speaker made a reference to the Alamo. The speaker was older, and the reference would have been the same sort of "make a stand" reference any number of speakers would have made over and over in the 90's. But after their talk, I was talking with some younger attendees, folks born after the turn of the millennium. Among them, the speaker's metaphor was a hot topic. Specifically the "yikes" factor of referencing the Alamo in any way that in any way implied the defenders should have been looked up to. The speaker's intended message was lost in the specific details of their chosen metaphor because the words which they absolutely intended to speak did not hold the same meaning for the audience to which they were spoken. If some hypothetical AI speech editor existed where you could punch in the age range of your intended audience and it would edit out metaphors and references that would land wrong with the audience, is that being deceptive or is that good editing and "reading the room"?
If you start to write an angry email, pause, and genuinely think of valid logical arguments (knowing that the previous anger may still be biasing the reasoning), that isn't deceptive. If you're masking the anger and not trying to reason calmly, which implies that you're using motivated reasoning, that is deceptive. Similarly with an (AI) speech editor, it depends on whether you're just trying to score points easily or whether you have the genuine intention to connect your experiences to your audiences' and give a thought out speech. Unfortunately, the results might be similar, but we should all aim to encourage the latter and discourage the former where we can.
