AI reflections on the Quantum Governance Studio

What this AI does (and does not do)

What it does

• Reads aggregated vote counts and anonymised “why” fragments from the Parthenon and Benin governance fields.
• Produces short summaries of emerging patterns, recurring themes in the “why” responses, and possible readings of the quantum governance worlds.
• Runs as a local process on the same server as this site, using an open‑source, compact language model.

What it does not do

• Does not control or optimise restitution decisions.
• Does not update the underlying governance rules or weights.
• Does not learn from individual visitors, build profiles, or send any data to external services.
• Does not replace curatorial, community or institutional judgment.

Here it appears simply as one voice among many in an ongoing conversation about restitution, technology and public reasoning.

How often this updates

The Quantum Governance Studio and dashboard update roughly every 15 minutes from live voting. This AI reflection updates once each night, after the day’s voting has settled, by reading the latest anonymised data and composing a slower, more reflective response. What you see here is therefore a nightly echo of the more immediate numbers elsewhere on the site, not a live feed.

How it fits into the stack

• Visitors cast votes and, if they wish, write a brief “why” explaining their choice.
• A light‑weight pipeline aggregates this data and updates the public dashboard at regular intervals.
• Once per night, a small local language model (running in a constrained configuration via llama.cpp) reads only those aggregates and “why” fragments and writes a short text file, ai_analysis.json.
• This page simply reads that file and displays the current nightly text.

The model runs with fixed prompts and settings. It is not fine‑tuned on visitor data and does not adapt its behaviour over time.

Ethics and intent

The aim here is not to outsource judgment to AI, but to test what happens when a modest, slower machine reading sits alongside human governance, curatorial thinking and public debate. The model can surface patterns, tensions and contradictions that might be hard to see at a glance, but any authority it appears to have is deliberately provisional and open to critique. You are invited to read its output as you would any other text in the exhibition: with curiosity, doubt and a sense of the wider histories of extraction, commodification and repair in which AI now takes its place.