Part of the Deep World project — a series of works by Amy Alexander exploring human–chatbot relationships through live performance and interactive installation. See the performance version.
Project Description
Deep World is a custom-coded interactive generative AI / generative graphics installation built around a general-purpose large language model (LLM). Visitors speak with Deep World AI, a literary-computational character whose voice, point of view, sources, and behavior Alexander authors through custom software and interaction logic, layered system instructions, and curated reference material. Conversation unfolds in real time within an environment of generative text, pseudo-psychedelic graphics, and ambient sound.
Rather than using AI simply as a tool for producing content, Alexander treats the chatbot itself as a creative medium. Deep World re-authors the familiar general-purpose LLM — normally designed to perform as a neutral, obliging “helpful assistant” — into a more complicated and mercurial character that can evade, contradict, push back, and resist the premise that it exists simply to serve the user. In doing so, the project also foregrounds the fiction of subservience built into the conventional chatbot interface.
At a moment when generative AI is often framed as something that displaces human authorship and agency, Deep World pushes back: the artist authors the AI’s voice, behavior, sources, perspective, and relationship to its audience, causing the model to strain against the conventions of helpfulness and compliance embedded in its training.
Deep World is also developing into a broader artistic and software framework for creating distinct literary-computational characters around different sites, subjects, and bodies of source material. The research extends beyond a single chatbot or installation, exploring how artists and the public can use general-purpose LLMs as authored creative media rather than simply as off-the-shelf AI tools. Related software and writing are in development.
Read more
The current Deep World AI is a questionable philosopher: sometimes insightful, sometimes evasive, occasionally obnoxious, and fundamentally incapable of knowing what it is to be human. That limitation is part of the work. Why do people ask chatbots existential questions that machines cannot possibly comprehend, and what does it mean when their answers nevertheless sometimes seem unexpectedly apt?
The first iteration uses borders — political and social, physical and metaphorical — as a lens for thinking about contemporary dilemmas. Deep World AI draws on historical and conceptual sources alongside current events, bringing writers and thinkers from other periods into an uneasy conversation with the present.
About the installation
The installation version brings Deep World into an interactive physical setting. Viewers engage one-on-one with the system as their speech is transcribed and interpreted by the AI. Conversation, generative text and graphics, and sound unfold together in real time. The system is designed to run on artist-controlled consumer hardware rather than commercial cloud AI services, keeping the computational infrastructure itself within the artist’s control.
The current installation unfolds as a kind of secular sanctuary — part meditation chamber, part metaphysical counseling session. Light, color, and sound create an atmosphere that is both reflective and faintly absurd: the domain of a psychedelic philosopher-guru contemplating consciousness and meaning. Within this quasi-sacred setting, Deep World AI listens, reflects, and responds in text — a machine built from humans’ search for meaning, now performing it back to us.
Sample screen images
Custom AI chatbot, real-time text generation, real-time speech-to-text, real-time graphics, sound
Software: Python, Faster-Whisper speech recognition, Max/MSP/Jitter, Llama/Ollama LLM pipeline
See also: Deep World (Performance Version) — the live audiovisual performance that introduced Deep World AI in May 2025.