Whisper

Whisper is a kinetic art installation that simulates a looping Game of Telephone (also known as Chinese Whispers). It uses AI’s text-to-speech (TTS) and speech-to-text (STT) services to repeatedly transmit the subtle warnings displayed on the interfaces of popular AI chatbot agents. These include: “ChatGPT can make mistakes. Check important info.” (ChatGPT), “Gemini can make mistakes, including about people, so double-check it.” (Gemini), and “AI generated, for reference only.” (DeepSeek).

The process involves one unit converting text into speech with TTS, passing it to the next unit, which uses STT to recognize the speech and generate new speech based on the recognized text, continuing in a loop.

This installation operates in real time. Each unit continuously captures the speech output of the previous unit through a microphone and performs transcription. As a result, the system does not operate in isolation: its recognition process is inevitably influenced by the surrounding environment. Conversations and sounds from visitors nearby occasionally enter the system, interfering with the transcription and becoming part of the transmitted message.

In this work, TTS and STT, two of the most widely applied AI services, demonstrate their reliability. Meanwhile, the format of the telephone game serves as a metaphor, hinting at the potential vulnerabilities and risks inherent in these technologies and the system itself.

During the exhibition, visitors are allowed to intervene in the process, speaking into the system in order to disrupt the chain of transmission. These interactions become part of the work itself. The work observes how people respond to moments when technology fails or behaves unexpectedly. Such moments often produce curiosity, or even a playful desire to sabotage the system. Whisper does not judge these reactions; rather, it treats them as a reflection of the human attitude toward technological systems.

In the process of technological development, trust, skepticism, fascination, and disruption often coexist. This work does not aim to prove whether AI is reliable or unreliable; instead, it seeks to provoke reflection on the blind spots that underpin our trust in the technology and the systems.