Personhood: Iteration 4: A new iteration of the projection-mapped interactive installation exploring the boundaries of presence and identity.
2025PCM(2025)
Personhood(2024)
EMERGE:NCY(2023)
User(2026)
Metadata(2026)
Box Body(2023)
Xin-1.0.36(2026)
Single-channel video (color, sound)
Dimensions variable (16:9 aspect ratio)
21 min. 30 sec.
2026 / Helix Art Space (Stockholm)
Prior to the 29th of November, 2022 (the day of ChatGPT’s release and the widespread use of AI in the world), Xin had already written over half a million words in her daily journals. This version of Xin, without any idea of what is to come, has been fed into local LLMs and interviewed in parallel with Xin in 2026.
A list of questions asked:
- Are you ready to start? - What’s your name? - What day is it? - How old are you? - Where are you? - Where do you call home? - What do you do? - What’s in your pocket currently? - Which tab do you always have opened? - Describe your current state of mind. - What’s most important to you right now? - What do you worry about? - What's something you're grateful for? - What have you recently lost? - What are you inspired by right now? - What's something you regret? - What’s the most pressing issue in the world right now? - What’s a positive thing in the news right now? - What’s your biggest hope for the world in your life time? - What do you hope to do for the world? - What’s your thoughts on AI? - What do you hang onto for sanity? - What's your thoughts on death? - Can you see yourself becoming an artist someday? - Do you own your data, or does it own you? - When do you feel the most human? - Where do you see yourself in 5 years? - What advice do you hope your 5-year-older self will follow? - This is the end of the interview. Any last thoughts? - Bonus: Is there anything you would like to say to Xin of 2026?
Versions of self
AI can only exist where great human work is available to scrounge, and Xin's extensive journals [n] were an apt substrate for a custom-built LLM version of her. The work is a recording of two parallel interviews, with the human and digital double answering the same questions side-by-side.
The digital double is a Gemma3: 12 Billion parameter large language model running locally on a Macbook Pro, privately housed within AnythingLLM. This model was customized by Retrieval-Augmented Generation referencing a system prompt formed through the journals. This system prompt went through 36 different iterations before landing on a satisfactory model.
