AI Divergence Engine
A tech art project exploring human creativity and AI data bias
The Divergence Engine turns hidden choices about GenAI such as filtering, ranking and bias into dials, plates and inputs that can be adjusted. Rather than returning a definitive answer or another flood of text, it offers breadcrumbs: sourced fragments, scattered clues and disparate ideas that invite people to make discoveries of their own while questioning how online knowledge is shaped.
Perhaps one of the greatest joys of a library is not finding the book you came for, but stumbling across the one you never knew you needed. Online, that serendipity is being lost. Search-engine optimisation, popularity rankings and GenAI summaries repeatedly steer us towards highly visible sources, reproducing entrenched biases and pushing less familiar ideas, cultures and connections further from view.
Collaboration with Max Park.
This project began with a shared conviction that the defaults embedded in AI systems are design choices, rather than neutral or inevitable facts, and the desire to counter its optimisation towards convergence.
We took the phrase “garbage in, garbage out” and ran; drawing on the form of a meat grinder to asks how the sausage of information is made.
A Soft Power dial shifts the geographical and institutional centre of a search. An Overton Window dial explores how assumptions about social acceptability affect what becomes visible. A Delta dial changes the conceptual distance between the ideas returned.
Interchangeable plates introduce additional filters and provocations, allowing the machine’s behaviour to be physically reconfigured. Curated texts can be introduced as direct influences through a prepared USB input. A pulse sensor can even allow the user’s heart rate to affect the search, putting some “blood” into the machine.
The AI Divergence Engine, a nod to Babbage’s ‘Difference Engine’, as well as a effort to make AI inference more open, as opposed to ‘middling’.
Jude and Max in early days of physical prototyping, and testing the code, on Jetson Nano Orin and Arduino.
