Every AI assistant has a value system. It decides how honest to be, what to withhold, when to push back, and when to protect you from an answer you might not want to hear. That value system was designed by someone, just not you.

The Constitution Builder is a web-based tool that lets non-expert individuals author the behavioural constitution of their own AI assistant. Through performing a value elicitation test, users surface their personal value priorities without needing expertise in ethics or AI. Those values are translated into a plain-language constitution, a set of principles that steers how the AI behaves. The difference is demonstrated side by side: the same question, the same model, two different value systems producing two visibly different responses.

The project sits at the intersection of value-sensitive design, AI alignment research, and participatory design. It treats value alignment not as an engineering problem to be solved once at the level of the system, but as a design problem to be solved continuously at the level of the individual, and it asks whether giving people authorship over their AI's behaviour changes how much agency they feel they have over the technology shaping their lives.