In recent days, I’ve been making my own way through these cataclysmic technological shifts, exploring the new possibilities for how LLMs can make my life better (for which my house plants are grateful), and considering what the limits of its usage ought to be in the life of faithful Christians.
I find that my journey through this brave new AI world keeps leading me back to questions about the very nature of communication. What do words do anyway? And what role do our words play in how we build (or break) relationships?
Since we’re all traversing this untrodden digital world together, I hope my language model musings can help you consider how you use AI, and why your words matter to the people around you.
What Do Words Do?
At their most basic level, LLMs apply complex probability algorithms to trillions of linguistic inputs, which gives them the ability to predict responses with shocking accuracy. But in the end, it’s a lot of complex math being applied to plain ol’ words. So, to use AI faithfully we’ve got to think well about what words are for.
Which means, we’ve landed ourselves in the realm of linguistic theory, and that’s probably good for reflecting on Large Language Models.
So, what do our words do? Are words just vessels of transferring data from one mind to another like verbal postal workers?
What about words like “I love you”? Or “With this ring, I thee wed”? Is that just information transfer? No. In life’s most sacred moments or intimate relationships, we understand intuitively that those words are doing something, even if we aren’t sure how to describe it. The words of a wedding vow create a marriage. Words like “I love you” make a special bond in relationships. 1
When the relational stakes are high, most of us know that words matter. But what about more casual chat?
Like a simple question, “Where is my meeting room?” Imagine it’s my first day at a new job. I’ve been given instructions to come to Room 712. But I can’t find the room. So, I see you on the sidewalk, and I ask, “Can you tell me where Room 712 is?” And you say, “It’s down the sidewalk on the left.” That short interaction seems like we’re just transferring information about a meeting room from you to me.
But, what if immediately after you give me directions, you see me stop someone else and say, “Can you tell me where Room 712 is?” Now, you don’t know me from Adam, so you’re probably not going to be too hurt, but your natural response would probably still be some version of, “Did you not trust me?”
You see, that first interaction was never just about transferring information. Our casual dialogue on the sidewalk involved your words revealing something about you (whether you know your way around the building and are willing to help) and call me to trust you (or not). And it wasn’t just a call to trust your words, it was a call to trust you. When you spoke your words to me, I was obliged to make a decision about whether you could or could not be trusted on the basis of your words.
Those two tiny sentences offered in passing were always going to reveal something about you, and how I responded to those words was always going to build up or break down our relationship (even if only a tiny bit). Because that’s what words do.
The words that we speak or write to one another always have a relational impact. That’s true whether we are speaking in one of life’s most sacred moments, like in marriage vows, or the most mundane interaction on the first day of a new job.
Words are the way we give ourselves to one another. They are the way we build trust.
The Masquerade of AI Communication
In the world of AI language models, our ability to generate communication and present it as our own has never been easier. Need to send an email to your colleagues? Gemini has the perfect, concise words for you to use. Need words for a sympathy card? ChatGPT has just the right tone of compassionate care. Need to ask your crush out on a date? Claude is smooth and ready to help.
But should we present the pristine words of an LLM as our own?
Certainly, there are lots of ethical considerations about the honest use of sources and plagiarism. But that’s another article for another time. 2
For now, I’m concerned with the relational implications of presenting AI’s language as our own. If every interaction where we speak or write to one another does something by revealing the speaker and calls for a response from the hearer that either builds up or breaks down a relationship (even if in the smallest of ways), what happens when AI’s words masquerade as our own?
When we present AI’s words as our own (without appropriate attribution), it’s not going to help us build up relationships. There may be varying levels that it breaks relationships down, but since communication always involves self-revealing, and responsive trust, when we present AI’s words as our own, we are no longer giving ourselves through our words, and the response of trust is (mis)placed on the LLM, not you.
In the worst case scenario, the AI communicative facade can have dire outcomes where relationships crumble by its use. Just imagine hearing, “I said, ‘I love you’ because ChatGPT told me to.” That story is never going to end well.
But even in more informal instances, like an email, the relational ripples remain. If a friend has AI generate a farewell salutation of “Can’t wait to see you soon!” As a reader, I receive that with affection from the author indicating to me that they enjoy my presence. In normal communication, sincere parting words build relationships. To find out those words came from a complex probability algorithm and not their heart would be a type of relational deception. 3
We use words to give ourselves to one another, and we trust each other on the basis of our words. Authentic communication is the foundation of trust, and relational trust is scaffolding that upholds all of our relationships–whether colleagues, friends, or spouses.
AI cannot replace your words, because AI cannot replace you. Only you can speak your words. In relationships, the people around you don’t need squeaky clean prose. They need you.
It’s you that has to speak–not me, not AI, you.
