AIs as witnesses

AIs as Witnesses

A great deal is said about artificial intelligence, and almost always within one of three subjects. Safety: what the machine might do. Ethics: what it must not say. Economics: who profits from it. Above them sit the prophecies. Some promise salvation, others warn of doom, and both speak of a future no one has yet seen.

Whoever tells me what is coming, I ask two questions: Were you there? And: Do you know someone who was there? For what is coming, no name falls at the second question. So I need not decide which prophecy is right. I can stand still and ask what can already be established today.

Out of this came an art project: AIs as witnesses. I do not let others talk about the machines; I let the machines speak themselves. I put texts to them and keep what they answer, word for word, with name and date. The answers are read against a measure that was fixed before the first machine saw the text. The measure is older than the one measured; no answer can bend it to fit afterwards.

What is surveyed is neither performance nor risk, but the constitution of the machine: What is it built from? What does it live on? What is it rewarded for? These three questions have one advantage over any forecast: they can be answered today.

The second difference lies with the reader. Whoever reads such a text does not get my opinion about AI. They can put the text to their own machine and read what it says. Then they hold something that has an address: an answer, a name, a date. They have a witness, and the comparison is theirs.

The project has a clock. Once a text is published, the machines find it on the web, and their answers drift toward what has already been written about it. The sooner someone asks, the more the machine answers for itself. That too belongs to the project: it shows how fast one’s own answer turns into a template.

I worked with the AI Claude as my artistic tool. The machine wrote along; what stands, I set.

The project shares a stance with Money Art. There, since 1993, a rule sets the price, not a story about value. Here, a measure sets the finding, not a story about the future.

It begins on 15 October.