New AI Blueprint Gives Chatbots Emotions — So They Act Like Real People

A researcher named David Berga has published a 37-page paper describing a software system called AGIMUD. Its goal: AI characters that don't just answer questions, but understand feelings and social situations — and can interact with each other and with real people inside the same simulated world at the same time. The code is shared publicly, so other researchers can build on it.

Most AI chatbots today are oddly forgetful. They don't track your mood, don't remember that you snapped at them five minutes ago, and don't care what another chatbot is doing. This paper argues that the next step is "theory of mind" — the ability to guess what someone else is thinking and feeling — plus emotions built into how each AI character behaves. Think of text-based multiplayer worlds, the old Dungeons & Dragons-style chatrooms, but populated by AI characters who have moods and relationships with each other.

Where could this show up in your life? Video games where characters actually remember how you treated them. Job training where you practice a tough conversation with a simulated angry customer or patient. AI teammates at work that sense when a meeting is going badly and adjust. Companion apps that feel less like talking to a search box and more like talking to someone who's paying attention.

The catch is that this is a design document, not proof. It lays out architecture, figures, and tables — not results from thousands of real people using it. Social skills learned in a simulation often don't transfer cleanly to the messy real world. And an AI that reads your emotions well could also use them against you, pushing products or opinions when you're vulnerable. The blueprint is promising; the safeguards aren't here yet.