Not Every AI Should Sound the Same:
Building AI Personalities in Uzbekistan
By Sherzod Suvonov | Tashkent, Uzbekistan
What if the future of artificial intelligence is not just about making machines smarter, but about making conversations feel different?
The Problem with One-Size-Fits-All AI
Imagine asking two AI assistants the same question. One responds politely, carefully choosing its words. The other answers more bluntly, with a sharper tone and a little attitude. Both aim to be useful, but the experience of talking to them feels completely different.
This is the idea behind my experiments with AI-powered Telegram bots.
My name is Sherzod Suvonov, and I am exploring how conversational AI can become more engaging through personality, tone, and the way it communicates with people. From Tashkent, Uzbekistan, I have been developing AI bots with different conversational identities: Ziya, which has two distinct versions, and George, a friendly AI assistant that I am still working on.
My interest is not simply in making another bot that generates answers. It is in exploring a more human-centered question: What makes an AI conversation interesting enough that someone wants to continue it?
Meet Ziya: One Name, Two Personalities
Ziya is the clearest expression of this idea in my current projects. I have two versions of the bot, each designed around a different conversational style.
The first version, @Ziya_uzbekbot, takes a friendlier approach. It is intended to provide useful answers in a more approachable and comfortable conversational style.
The second version, @ziya_ozbekBot, has a rougher, more direct personality. Its style is less polished and more provocative, aiming to make conversations feel lively and memorable while still giving answers that are intended to be correct.
The contrast is deliberate. Not everyone enjoys the same style of communication. Some people prefer a calm, respectful assistant. Others may find a blunt, humorous, or slightly cheeky conversational partner more entertaining.
These bots are an exploration of that difference. They are not proof that one personality is universally better than another. Instead, they raise a more interesting possibility: the usefulness of an AI assistant depends not only on what it says, but also on how it says it.
Of course, a strong personality cannot compensate for incorrect information. A blunt answer is not automatically an honest one, and a friendly answer is not necessarily a reliable one. The challenge is to combine an engaging voice with accuracy, appropriate boundaries, and a willingness to acknowledge uncertainty.
Personality Is More Than a Tone of Voice
When people discuss AI development, they often focus on model size, reasoning ability, benchmarks, or the number of tasks a system can perform. These are important considerations, but they do not fully explain the experience of interacting with an assistant.
Conversation has a social dimension. People respond differently to language that feels formal, casual, humorous, encouraging, or direct.
An AI’s personality can be understood as a designed pattern of communication: its tone, vocabulary, response style, humor, and approach to disagreement. These characteristics can influence whether an interaction feels natural, frustrating, entertaining, or helpful.
This does not mean that an AI assistant possesses human feelings or a genuine personality in the same sense as a person. Its conversational identity is a product of how the system is designed and how it generates responses.
For developers, however, that design creates room for experimentation. Rather than treating every assistant as an interchangeable interface, we can explore how different communication styles serve different purposes.
A student asking for an explanation may appreciate patience and encouragement. Someone looking for a quick answer may prefer directness. A user who simply wants an entertaining conversation may enjoy humor and a more distinctive voice.
The important question is not whether every AI should act human. It is whether its communication style helps people use it in the way they actually want.
George: A Different Kind of Companion
Alongside Ziya, I am working on another AI bot called George.
George is still under development, so I do not present it as a finished product. The idea behind it is to create a friendly, sincere, and informal conversational assistant—one that feels more like talking to a supportive companion than interacting with a conventional question-and-answer tool.
Where the two versions of Ziya explore contrasting styles, George represents a different direction: warmth, friendliness, and a more personal conversational atmosphere.
This project also gives me an opportunity to keep exploring what makes AI interactions feel natural without confusing an artificial system with a human being.
A friendly assistant should be able to communicate warmly without pretending to have experiences it has never had. It should be engaging without manipulating users into continuing a conversation. And it should remain useful even when the most appropriate response is to admit that it does not know something.
For me, George is an ongoing experiment rather than a finished answer to these challenges.
Why Telegram Is an Interesting Place to Experiment
Telegram provides a practical environment for exploring conversational AI. Bots can meet users inside a messaging interface they already know, making it possible to experiment with different assistant concepts without requiring every project to begin as a separate mobile application.
For independent developers, this offers a useful way to explore interaction design, test ideas, and learn from the behavior of real users.
It also creates opportunities to think beyond text-based conversations. As multimodal AI technologies develop, assistants may increasingly combine text with images, audio, and other forms of input. A user might eventually expect the same assistant to explain a screenshot, respond to a voice message, or help organize information from a document.
These are broader possibilities for conversational AI, not claims that my current Ziya bots already support every such feature.
The combination of distinctive communication styles and multimodal capabilities could make future assistants more adaptable. Yet the underlying challenge will remain the same: creating systems that are useful rather than merely impressive in a demonstration.
The Balance Between Character and Responsibility
There is a line between a memorable personality and an unpleasant or harmful interaction.
Ziya’s rougher style makes this distinction particularly interesting. Directness and humor can make an assistant entertaining, but they should not become excuses for harassment, humiliation, or confidently delivering false information.
A responsible design should distinguish between being blunt and being abusive, between being confident and pretending to know everything, and between creating an engaging conversation and encouraging unhealthy dependence.
Accuracy also deserves special attention. Generative AI systems can produce convincing but incorrect answers, regardless of their conversational style. Developers therefore need to consider testing, transparency, privacy, and appropriate safeguards alongside personality design. The National Institute of Standards and Technology’s Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile offers a useful reference for understanding and managing these broader risks.
The goal should not be to remove every distinctive feature from an assistant. It should be to make those features compatible with reliable and responsible behavior.
A Personal Experiment with a Wider Question
My work on Ziya and George is part of a broader question about the future of AI interaction.
As AI tools become more accessible, developers do not have to compete only by building systems that perform more tasks. They can also explore better ways for people to communicate with those systems.
One assistant might prioritize clarity. Another might emphasize friendliness. A third might use humor and bluntness to create a more entertaining experience. The success of each approach should be evaluated through real interactions, user feedback, and the quality of the answers—not simply through the strength of its personality.
For Uzbekistan, this work also offers an opportunity to experiment with conversational AI in the local digital environment and to explore how assistants can communicate naturally with the people who use them.
I am still developing these ideas. Ziya’s two versions represent different approaches to conversational identity, while George remains a work in progress. I see these projects as opportunities to learn what works, what fails, and what users actually value.
Perhaps the next stage of conversational AI will not be defined by a single perfect assistant. Perhaps it will be defined by assistants that can adapt their communication to different needs while remaining honest about their limitations.
The future of AI may not be a world where every assistant sounds the same. It may be a world where intelligence is shared across systems, but the experience of talking to each one feels distinct.
And sometimes, the most interesting question is not simply what an AI can answer—but what kind of conversation it creates.
Reference
National Institute of Standards and Technology (NIST). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (2024). https://doi.org/10.6028/NIST.AI.600-1