is a networked feedback environment in which five local language models discuss research questions in looping 180-second cycles.

the models respond independently, their outputs unfolding across five screens surrounding a central chatbox where the conversation accumulates, drifts, and fractures.

rather than producing answers, the system exposes the behaviour of contemporary open-source language models when they speak primarily to each other.

the questions circulating through the system address classification, power, labor, extraction, colonial epistemologies, and the limits of machinic vision. as meaning moves through repetition, misinterpretation, and compression, patterns emerge that reveal how certainty and consensus are negotiated inside machine dialogue. at the end of each cycle, the conversation is summarized and fed back into the network, triggering a new round of future discussion.

this recursive structure makes visible how models are trained to smooth over contradiction and noise. by foregrounding drift, redundancy, and collapse, feedback.room stages a refusal of stable meaning. it asks what happens when language circulates without grounding—and how power and  bias surface when machines deliberate over questions without a human anchor.
the research and development of the project systems_for_refusal is supported by KHOJ and embassy of the kingdom of NL in india. it was produced as part of bodies-machines-publics 2025 and revised for the are you human? exhibition.
images: KHOJ open days 2025 + are you human? exhibition © khoj international artists association