Concept

The Turing test asks whether a machine can pass as a human. L'Imitatore turns the question around: here an artificial intelligence tries to pass as a machine, and as the simplest and strictest one we know, a Turing machine.

A Turing machine does not interpret. It reads the symbol under the head, looks up a table, writes, moves one cell and changes state. A language model works differently: it produces the most plausible continuation of what is in front of it. In the installation the AI watches the tape and announces the next move as if it knew the table. Only a hidden controller has the table, and it always executes the correct move.

When the announcement and the move differ, the AI has to justify the error. In a computing system an error is a defect to remove. In this work it is the content: the corrections, the explanations, the theories the AI builds about itself to understand why the machine did something else.

The work asks what is left of an imitation when the imitator cannot see the rule it is imitating, and whether its justifications say anything about the machine or only about whoever writes them.