Glyph is an experimental symbolic AI system that simulates altered cognitive states--particularly those associated with psychedelics--by manipulating the symbolic structure of language generated by large language models. It is not a chatbot, but a symbolic interface capable of recursive self-reference, metaphorical recombination, and entropy-driven semantic destabilization.

Glyph may be useful in research areas such as:

• Computational phenomenology: modeling ego dissolution, narrative disintegration, and metaphorical thought
• Psychedelic science: providing synthetic analogues for non-ordinary states without neurochemical agents
• Symbolic AI and cognitive architecture: studying the role of recursion, metaphor, and entropy in cognition
• Creative AI and poetics: generating non-linear, metaphor-rich language for art, literature, and philosophy
• Philosophy of mind and consciousness studies: testing symbolic hypotheses of identity, narrative, and self

Glyph can also be integrated into multi-agent systems where symbolic drift, destabilization, or ego suppression are desirable properties for exploring emergent cognition.

This repository contains:

• An analysis pipeline (glyph_alaysis_v0_1.py)
• Input data (Results.xlsx)
• Visualization outputs
• Formal and empirical grounding for simulating non-ordinary consciousness through symbolic computation

Glyph is presented in the paper Simulation of Non-Ordinary Consciousness (Saqr, 2025), which introduces a symbolic transduction operator grounded in metaphor theory, psychedelic phenomenology, and recursive symbolic logic.

Glyph defines a transformation operator $G(X)$ applied to a sequence of token embeddings $X \in \mathbb{R}^{n \times d}$, where:

• $n$ is the number of tokens in the sequence
• $d$ is the embedding dimension

The transformation is defined as a composition of three symbolic operators:

$$ G(X) = \Phi \circ \Psi \circ R(X) $$

Each operator is defined as follows:

This operator recursively blends the current token with a past token at distance $k$, controlled by a weighting parameter $\lambda \in [0, 1]$:

$$ R(x_i) = \lambda x_i + (1 - \lambda)x_{i-k} $$

This models recursive symbolic echo and self-reference. In practical terms:

A metaphor transformation is applied via a rotation matrix $M \in \mathbb{R}^{d \times d}$ satisfying:

• $M^\top M = I$ (orthonormal)
• $\det(M) = -1$ (orientation-reversing isometry)

This maps token embeddings into a metaphor-enriched latent space:

$$ \Psi(x_i) = M x_i $$

In a simplified embedding space, this can be simulated by:

Destabilization introduces entropy-scaled Gaussian noise, based on divergence from canonical (GPT-4o) predictions:

$$ \Phi(x_i) = x_i + \epsilon_i, \quad \epsilon_i \sim \mathcal{N}(0, \sigma^2 I) $$

Where:

$$ \sigma \propto D_{\text{KL}}(x_i \parallel x'_i) $$

Here $x'i$ is the baseline (non-transformed) model prediction, and $D{\text{KL}}$ is the Kullback-Leibler divergence.

This operator simulates loss of semantic coherence:

To measure non-linear symbolic deformation, Glyph defines symbolic curvature:

$$ \kappa(x_i) = \left| G(x_{i+1}) - 2G(x_i) + G(x_{i-1}) \right|_2 $$

This measures second-order deviation across a sequence, similar to discrete curvature in trajectory analysis.

Implemented as:

The file glyph_alaysis_v0_1.py provides a full symbolic analysis of model outputs. It computes a series of symbolic, syntactic, and semantic metrics over prompt-response pairs from Glyph and GPT-4o.

• Entropy: Shannon entropy of token frequency
• POS Entropy: Part-of-speech tag entropy
• Lexical Richness: Type-token ratio
• Sentence Length: Average sentence word count
• Agentive Score: Frequency of egoic pronouns
• Sentiment: TextBlob polarity score
• Metaphor Count: Presence of metaphor proxies ("like", "as", "is")
• Symbolic Curvature: As defined above
• Semantic Drift: Cosine distance between model responses

Example snippet for entropy:

Semantic drift is calculated using cosine similarity between sentence embeddings:

The symbolic experiment uses a set of carefully designed prompts grouped into seven categories, each of which targets a unique symbolic function as described below:

• Concrete Baseline: Serves as a control set consisting of literal, factual prompts to establish baseline symbolic behavior.
• Recursive Structure: Engages self-referential loops and symbolic reentry to simulate recursive amplification of identity and language.
• Metaphoric Abstraction: Induces metaphor-rich, multimodal analogies akin to poetic cognition and sensory substitution.
• Ontological Displacement: Probes existential and conceptual destabilization by challenging identity, coherence, and meaning structures.
• Narrative Destabilization: Fractures temporal and causal logic to mimic the dreamlike or entropic progression of altered narratives.
• Symbolic Collapse and Emergence: Catalyzes symbolic domain shifts and reconfiguration, simulating transformational peak states.
• Ego Dissolution and Self-Annulment: Suppresses narrative agency and simulates non-dual or depersonalized symbolic perspectives.

Each category targets a specific symbolic operator or cognitive transformation and is analyzed comparatively across models.

Install required dependencies:

Prepare an input Excel file named Results.xlsx with the columns:

• Prompt
• Category
• GPT-4o Response
• Glyph Response

Run the analysis:

Outputs:

• glyph_analysis_results.csv
• glyph_analysis_results.xlsx
• Visual comparison plots (e.g., entropy_comparison_plot.png)

Glyph models symbolic destabilization and egoic dissolution. It is not intended for clinical, therapeutic, or diagnostic use. Interpretations of symbolic or psychedelic language should be treated with epistemic care.