An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models
Researchers tested whether noise-like sounds, such as whale vocalizations and white noise, can trigger structured responses in a language model (GPT-2 small). They used a score called Semantic Induction Potential (SIP) to measure this reactivity. Results showed that natural sounds had higher SIP than pure white noise, suggesting models detect underlying structure even without conventional meaning. This points toward a new SETI strategy focusing on data patterns rather than just decipherable messages.