An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models

The gist: This exploratory framework tests whether noise-like input can induce structured responses in language models, suggesting that generative reactivity may offer a new way to identify data worth closer attention for SETI.

Hypothesis and Motivation

The authors propose that highly advanced extraterrestrial civilizations may favor low-power or indirect signaling strategies over beacon-like transmissions, which are vulnerable to distance-related attenuation and location revealing (Page 3). This approach is motivated by the limitations of language itself, as communicating based on syntax assumes common foundations for processing information that may not hold in interstellar communication (Page 3). The framework suggests that input data does not need to be decoded as a message but can be evaluated for its capacity to trigger structured behavior in a highly advanced system (Page 3).

Method: Probing Structured Output from Noise-Like Data

The researchers used GPT-2 small, an 117M-parameter model trained on English text, to test whether noise-like input can trigger structured generative output (Page 4). To assess reactivity, they defined a composite score called Semantic Induction Potential (SIP), which combines entropy, syntax coherence, compression gain, and repetition penalty (Page 4). The Semantic Triggering Detection Pipeline (STDP) involves several steps:

1. Data projection: Audio inputs were resampled to a common sampling rate and converted into log-mel spectrograms [24, 43] (Page 6).

2. Tokenization: The flattened spectral vectors were used for unsupervised clustering to produce sequences of discrete symbolic tokens, preserving sequential structure in the input prompt (Page 6).

3. LLM provocation: These token sequences are passed as prompts to a pretrained language model without conditioning or instruction (Page 6).

4. Semantic response measurement: The SIP metric integrates four subcomponents: Token-level entropy (Htoken), Syntax coherence score, Compression gain, and Repetition penalty (Page 8).

Semantic Induction Potential (SIP) Metric

The SIP is computed as a weighted sum of four components: SIP = α · (1 − Htoken) + β · Syntaxscore + γ · Compressiongain − δ · Repetitionpenalty (Page 9). The coefficients used were set as α = 2.0, β = 1.5, γ = 1.0, and δ = 0.5 to give higher priority to entropy and syntactic coherence (Page 9).

** Token-level entropy (Htoken) reflects uncertainty in the output [37] (Page 8). **

** Syntax score estimates syntactic fluency using the inverse token-level crossentropy loss [36] (Page 8). **

** Compression gain measures structural regularity via the relative reduction in UTF-8 byte length after zlib compression [12] (Page 8). **

** Repetition penalty penalizes excessive token-level redundancy, measuring the proportion of tokens that occur more than once [8] (Page 8). **

Key Results and Interpretation

The results showed that whale and bird vocalizations had higher SIP scores than white noise, while human speech triggered only moderate responses (Page 10). This suggests that language models may detect latent structure even in data without conventional semantics (Page 10). The findings indicate that the model responds to structural patterns rather than semantic content, reinforcing the idea that SIP works more like a detector of pattern density than a classifier (Page 10). The framework points toward a shift in SETI detection strategy, suggesting focus on data rich in internal patterns even if they do not resemble language (Page 12). This leads to the concept of cosmic linguistic seeding, where communication is based on triggering symbolic behavior rather than content delivery (Page 15).

Future Directions

The work suggests several avenues for future investigation, including:

1. Application to public SETI archives, such as baseband and filterbank datasets, to evaluate whether discarded segments trigger structured model responses (Page 14).

2. Establishing SIP baselines across different stellar classes or target types to identify outlier segments with unexpectedly high SIP scores (Page 14).

3. Developing lightweight structure-sensitive language models that can identify data with high potential to trigger generative behavior (Page 12).

This approach aims to uncover structure that escapes conventional energy-based or periodicity-based filters, potentially supporting a shift from real-time monitoring to long-term data mining in SETI efforts (Page 12). The framework offers a new way to think about active SETI by suggesting advanced civilizations might send data meant to activate symbolic behavior in whoever receives them, which is called cosmic linguistic seeding (Page 15). This work demonstrates that generative linguistic reactivity is both measurable and sensitive to structural patterns, asking whether detectability must rely on embedded meaning or if structure alone can provoke linguistic behavior (Page 16).

--- Page 1 ---

An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models Po-Chieh Yu1,2 1Taiwan Astronomical Research Alliance (TARA), Taiwan 2 Institute of Astronomy and Astrophysics, Academia Sinica, Taipei, 10617, Taiwan Abstract We present an exploratory framework to test whether noise-like input can induce structured responses in language models. Instead of assuming that extraterrestrial signals must be decoded, we evaluate whether inputs can trigger linguistic behavior in generative systems. This shifts the focus from decoding to viewing structured output as a sign of underlying regularity in the input. We tested GPT-2 small, a 117M-parameter model trained on English text, using four types of acoustic input: human speech, humpback whale vocalizations, Phylloscopus trochilus birdsong, and algorithmically generated white noise. All inputs were treated as noise-like, without any assumed symbolic encoding. To assess reactivity, we defined a composite score called Semantic Induction Potential (SIP), combining entropy, syntax coherence, compression gain, and repetition penalty. Results showed that whale and bird vocalizations had higher SIP scores than white noise, while human speech triggered only moderate responses. This suggests that language models may detect latent structure even in data without conventional semantics. We propose that this approach could complement traditional SETI methods, especially in cases where communicative intent is unknown. 1 Introduction Over the past 40 years, the Search for Extraterrestrial Intelligence [SETI; 9, 13, 39, 42] has reported candidate signals, including the “Wow!” signal [25] and more recent anomalies near Proxima Centauri [41]. However, none have been independently confirmed [17, 18], and the latest detections have been attributed to terrestrial interference [38]. Conventional SETI approaches typically assume that extraterrestrial signals carry specific and decodable content, such as mathematical sequences and narrowband transmissions that are designed to be received by another civilization. These assumptions imply that extraterrestrial civilizations have advanced technological capability and communicative intent. However, alternative perspectives like the Zoo Hypothesis [3, 11, 15] and Dysonian SETI [4] suggest that extraterrestrial intelligence may either avoid direct contact or express itself through indirect, observable phenomena. These views support using a wider range of methods to interpret anomalous signals. Wright & Oman-Reagan [46] indicated that SETI often reveals more about our conceptual limitations than about extraterrestrial intentions, highlighting the role of self-reflection in signal interpretation. Cabrol [7] further emphasizes the need to develop new frameworks that move beyond assumptions tied to human senses and languages. They proposed that more diverse cognitive and interpretive models should be incorporated into future detection strategies. Based on these arguments, rather than assuming the data must be decoded, we explore whether certain input might trigger structured responses in generative models. Such responses, especially in systems trained on human text, could provide a measurement of detectability in data with latent structure. Most SETI workflows focus on measurable features, such as narrowband peaks or repeating patterns. Segments that appear broadband, aperiodic, or spectrally flat are typically dismissed as noise. Large portions of SETI data are never examined for hidden structure beyond statistical characteristics. As a result, many of these segments are treated as meaningless, even though some may contain forms of complexity that conventional methods fail to detect. Rather than asking whether the data is interpretable in human terms, we ask whether it can trigger linguistic behavior in generative models. While recent studies have applied machine learning to SETI signal classification [8, 10, 19, 30], the rise of large language models (LLMs) [e.g., 6, 32, 34, 36, 44] offers a different approach: generating structured output from inputs that carry no specific meaning and do not follow any known communicative conventions. This kind of model reactivity may reveal latent patterns in data that remain undetectable using conventional methods. Unlike previous work which analyzes potentially meaningful signals for linguistic patterns [2], our method begins with unfiltered, noise-like data and observes whether any internal regularity is sufficient to trigger language-like responses. In this work, we present a framework for testing structural reactivity in LLMs, using input from human language, animal vocalizations, and white noise. These inputs provide a baseline for evaluating model responses to different levels of latent structure. We begin by outlining the conceptual hypothesis and methods (Section 2), followed by a demonstration of model responses and interpretive analysis (Section 3). In Section 4, we discuss the broader implications for SETI and signal detection, and summarize key conclusions in Section 5.

--- Page 2 ---

Proxima Centauri [41]

--- Page 3 ---

human speech, humpback whale vocalizations, Phylloscopus trochilus birdsong, and algorithmically generated white noise. All inputs were treated as noise-like, without any assumed symbolic encoding.

1. Bold header: Semantic Induction Potential (SIP) Metric Integration

This improvement involves implementing a system that uses Semantic Induction Potential (SIP) to quantify linguistic reactivity in generative models, allowing the AI to detect latent structure even in data without conventional semantics.

2. Bold header: Noise-Like Input Screening

The improved system can screen massive datasets for potential signals by treating all inputs as noise-like and measuring their capacity to trigger structured responses, shifting the focus from decoding to viewing structured output as a sign of underlying regularity in the input.

3. Bold header: Structure-Sensitive Communication Synthesis

The AI can be used to design communication strategies based on cosmic linguistic seeding, where structured inputs could be designed to resemble background noise while still containing enough internal organization to trigger responses in generative systems.