Question Time

An artificial intelligence model drawing on the content of the CCCB’s archive answers your questions if you ask them out loud. But who is really answering? Does what they say make sense? Does it add anything new? And what does this experiment tell us about the possibilities and limitations of AI? «Question Time» is  The project also maps out the contents of the archive and the issues that have been raised at the CCCB over the years. 

The installation consists of a mosaic of screens to which you can pose a question. An AI model receives it and uses the knowledge accumulated in the CCCB archive to generate the answer, a kind of palimpsest that remixes voices and moments from the Centre’s history, since it was set up. The screens present not just the answer but also similar questions, and the talks on which the answer draws. «Question Time» can be seen and experimented with in the Foyer of the CCCB for a month.

This experiment by Estampa collective showcases the voices and knowledge generated by the CCCB in its 30-year history and examines how we approach a statistical tool like AI. Does it give us interesting analyses of reality? Can we take the generative variations of AI to be true? The installation is also a challenge and a game for the people who take the microphone to formulate a question. What is the point of the questions we ask? How can we ask a question to get interesting answers?

How does «Question Time» work?

For «Question Time», the CCCB’s archive of talks has become an artificial source of questions and answers. Transcripts of the events have been broken down into paragraphs, and a question associated with each has been generated using AI tools. This has created a body of more than 200,000 questions and answers, which can be seen as a description of the activity and concerns of the Centre over the course of these years.

This content has been mapped by grouping it by similarity, using AI tools that allow the task to be automated and performed in a complex way. The textual list is thereby transformed into a series of thematic and stylistic zones, where each question and answer is surrounded by other, similar ones.

When we ask a question, the installation uses this virtual map to answer it. Our question is projected onto this map, and it is the nearby answers that articulate the reply. The final wording is produced by a text generation model that receives these nearby answers and uses them to write the sentence that appears on the central screen.

Some aspects of this cartography of the Centre’s archive are shown on the secondary screens. First, the original fragments of the talks on the basis of which the answer is written. Then a progression of questions that link the last one asked with the current one.