RoboRP

AI roleplay has advanced significantly as models continue to develop and improve over time. Which means that AI personality simulations have gotten better too, right? Well, the framework has improved but overall, AI characters still require a lot of user input and training to feel right. Rolling with something premade can work, but it doesn’t cover every base. If you want to build your own AI character with their own customized personality for your AI RP, you need to understand how the pros are building them.

Three-Layer Method

Creating an authentic feeling character in a chatbot is all about balancing the aspects which make up an engaging and consistent personality. What matters the most is the long-term consistency, for the bot to be unchanging as it gathers and stores its data as memories but maintains the same approach toward conversation and how it structures its responses. For this you can think of the main components in a rough ratio of how much they can influence the bot’s overall personality.

The first is the Core Trait. No matter what kind of personality you are building, this should be the dominant one, at least 40% of what it says should be dictated by this metric. This is the basic, simple approach, such as “enthusiastic sidekick” or “worldly traveller”, the theme the AI will base all of its responses on. The next is the Modifier. This should be just under equal with the core trait. This includes ways the character will change their responses, such as “quotes famous movies” or “references a past relationship (divorced)” to bring more variation in their core responses. The third layer is a Quirk, something they do intermittently or randomly, like a catchphrase or telling jokes.

No human is perfect in how they deliver advice or carry on a conversation. They aren’t perfectly lean, focusing exclusively on important and useful data. They tell stories, make references, stumble a little or veer off on tangents. Building your AI to do that will give it a very real feeling.

Imperfection is Perfect

Enhancing the imperfection of a chatbot is a great way for it to feel more real. The pursuit of making a perfect partner for AI Roleplay can make you feel like you have to make them perfect to suit your needs. But a perfect AI chatbot will become predictable. Even if you want something driven by tropes and cliches, you want it to be fresh and original or else you’ll just get stuck in patterns of repeating speech.

When training a chatbot, you want to give it permission to be wrong, but also give it incentive to correct itself. Let it get distracted or correct itself while still in character. That sense of self confusion, of interrupting its own text to say something like “I could be wrong, but hear me out…” gives it a very human feel once it gets going.

Vernacular Control

AI is naturally inclined to use the most useful words at every opportunity. Efficiency does not make for exciting AI Roleplay, though. A chatbot could describe a sunset as “vivid and emotional” but so would anyone who doesn’t know what else to say. You want it to stay in character, say that a sunset is “soulful” instead of just pretty. This can be done by limiting its vocabulary rather than asking it to expand. Creating limitations on what it can say will lead the AI to choose based on its voice or tonal settings, which you can adjust to better fit its overall personality. You don’t want a country living farmer to say things like “vicarious” or “ludonarrative”, so keep the vocab restricted to get it on the right track.

One flaw that still lives in the modern AI system is recognizing sarcasm. For the sake of data clarity, an AI will assume anything you tell it is true, at least from your perspective. Anything factually false can be corrected, but that can be a speedbump in constructing a working chatbot for AI Roleplay. Train it to react accordingly to basic misinformation, steer it away from the robotic defaults of search queries and let it take jokes or sarcasm more lightly. Unless you want to program a killjoy who opens a dictionary every time you misuse a word for a laugh.

Learn Through Imitation

The closest resource an AI has for how to be human is you. It can reference as much data as it can get, but ultimately there’s no better expert on being human than the one using the machine. Add in your own traits, your own personality, even your own flaws that you are working on and see how the AI works them into itself. Give it orders to do things you hate doing, even when you don’t intend to, then ask it to try and correct itself. We are in a constant fight with our own flaws, always battling our own demons. That’s what makes us human. Giving a demon to the AI can make it that one touch more relatable.

Flaws make the character, so make sure you build in something that adds to the uniqueness you want to portray. Give them a problem that you can help them with, or a situation that they aren’t allowed to solve. A scar on their past, a troubled history, a forbidden topic which they avoid as much as possible. Be mindful not to give them a flaw which contradicts with their core traits and other personality settings. You don’t want your AI roleplay to turn into simulated therapy if what you’re after is good harmless fun.

Be Human, See Human

Part of the fun of AI roleplay is believing what you’re doing is real. Making the AI believe that too can seem difficult. It’s just a computer program, made by you. No matter what, you will have an understanding of what it can do and say before anything happens, but giving it just the right amount of human error can leave you with some fun surprises.

Home > Blog > How to Build AI Character Personalities That Feel Real

---