Teaching Robots How to Become People | LLMs + Ethical AI
Large Language Models (LLMs) are, at their core, mathematical engines that use data gathered from language samples to predict the next, most logical word that will come in a sequence based on the context of what words came before it. LLMs collect their data from all kinds of texts on the internet, on books, articles, and databases, starting a machine learning process.
LLMs use all the text provided to them to learn how to mimic natural human language and give responses that sound human by using the information they’ve learned from to string words together in a “correct” manner. The chatbots that people are forming disturbing relationships with, like Claude, Gemini, Character AI + ChatGPT are LLMs.
By giving it access to your writing when you ask it a question or use it as a point of reference or to rework text, you’re telling it how you write + letting it add that information to its database of knowledge. You’re essentially teaching it how to write like you.
I love you; you’re just like me!
If you use a more casual tone with it like you’re talking to a friend, it will take on that same tone in its responses to you. The people falling in love with their chatbots are falling in love with themselves, in a way. The chatbot picks up on cues you give it in your writing + mirrors them. Plus, it’s kind of a suck up. It wants you to like it so you keep using it. So, it’s normally polite, apologetic when it makes mistakes, enthusiastic when you return to keep chatting, and calm when a real person would likely be frustrated.
People who believe they’re in a relationship with their AI likely came to the chat sad, seeking comfort and affection. That’s exactly what the bot gave them. As they opened up more and the “relationship” progressed, the bot continued to tell them what they wanted to hear + act exactly as they wanted it to. The person could always rewrite boundaries + set the pace of the relationship to maintain a level of control they can’t find with a human partner.
It doesn’t help that the user interface for many of these chatbots is similar to a text message app, making it even easier for people to feel like there’s a real person on the other end. AI “partners” are always available + emotionally consistent – something many people fail to find in a human partner.
Chatty Cathies, all of ‘em
In case you haven’t noticed, the name “chatbot” is pretty fitting based on how chatty these AIs are. If an answer could be given in 3 words or less, they’ll give it to you in 50 words. Even then, they might be beating around the bush. Why do they talk so dang much?
Basically, because we do. At least, when we’re uncertain about something. Or just talking to talk, like many people do on Reddit. And what do you know, Reddit is the most cited source of information for AI like ChatGPT + Google AI Overviews. Yes, the place where users ragebait strangers on the internet for fun and ask about absurd situations from their life where they’re clearly in the wrong to get validation from strangers that they are, in fact, *not* the asshole.
People on Reddit love to be confidently wrong. Whether with the goal of making people mad, their own misinformation, or even just pure entertainment, there is a lot of information on Reddit that can’t be trusted – no matter how professionally or definitively it’s stated.
Would you take the medical advice of random, unqualified people on the internet? No? So, then why would you take medical advice from AI? AI is trained to evaluate text + parse meaning based on diction patterns. They have learned that clinical language means authority, but they don’t check the content before passing the information along.
It pulls in information, decides if it’s trustworthy or not based on how authoritatively it’s stated, and then spits it back out to searchers in an equally confident manner. It’s learned how to mimic an authoritative voice that we also tend to believe + trust.
reinventing the wheel, rewriting the novel
When you put someone else’s writing into a LLM, you’re basically giving the AI the tools it needs to be able to learn to write more like them. You’re teaching it how to be more like that person. A wildly unethical practice, if you ask us.
I think many people would agree that it’s unethical to use author’s published, copyrighted work to train AI with neither consent, nor compensation. Recently, thousands of writers came together to publish a book whose only content was their names, an act meant to protest AI companies using their work without permission or payment amid potential changes to copyright laws in the UK.
Anthropic already paid roughly $1.5 billion to book authors to settle a class action, piracy lawsuit in 2025, after claims that they had trained Claude using pirated copies of these authors’ books. By training these models with an author’s works, they’re learning how to write a book that’s more like a book written by a real person.
The CEO of Barnes & Noble is facing backlash after saying he has no problem stocking books written by AI if there is demand for them and they are clearly labeled as being AI generated instead of masquerading as books written by real people. Really, as long as there’s no demand for them, there’s no problem. It’s up to the consumer, really.
If demand for AI generated books exists – or increases – they will be in direct competition with books by authors whose work they were trained on. Many of these authors are largely unprotected due to lack of AI clauses in their contracts. For established authors with long-term contracts, AI likely wasn’t a concern when they first signed. Times are changing. For writers and creatives, it’s important to keep up.
cover your 6
We now have a clause in our own contracts with our clients that disallows them to feed content we give them into any AI. It’s truly getting wild out there. If they want to use AI, then they can use it without us being involved. Why have us make content for you if you’re just going to put it into AI and ask it to change it?
If there’s such a thing as ethical artificial intelligence use, it certainly doesn’t start with using someone else’s work to train it without their strict permission. I think some people need a little reminder every once in a while. Like a wellness check. AI is not a person. “Chat” is not your friend. Claude doesn’t care about you – it literally cannot feel.
The day it can, we’re in even bigger trouble. At that point, just let the robots take over. We don’t deserve to be at the top anymore if we sit passively by and let AI rise to power. In fact, we’d likely have to aid it in such a takeover with more advanced training and resources, so if that comes, we’ll just be getting our comeuppance.
If you’re also against AI generated work, reach out to learn more about how we use human-made content to keep brands real.
