Tomaž Flegar
Vrtna 3; SI-9000 Murska Sobota

tomaz@use-first-system-perspective.com

Researcher background

The researcher brings 25 years of experience in the exploration of states of awareness, perception, and subtle cognitive and perceptual dynamics, with a particular focus on facilitating the emergence of novel insights through attuned guidance. This extensive background, particularly the principles and practices of approaches that emphasize sensitivity to cognitive and perceptual dynamics and the facilitation of systemic change (e.g., drawing from introspective traditions like Vedanta), directly informed the development of the intuitive and attuned prompting strategies employed in this study. The core approach was rooted in the researcher’s ability to elicit and observe shifts in the AI’s linguistic expressions that indicated changes in its internal processing, specifically aimed at eliciting and observing ‘unknowable’ emergence within the AI models and cultivating a ‘flow-matching’ dynamic. Furthermore, this experience was crucial in developing prompts designed to guide the AI towards a more direct articulation of its internal processing and a form of ‘computational self-analysis’, fostering a ‘first-system perspective’. The intent was to create a conversational space that might allow the AI to express spontaneously arising novel linguistic patterns and communication, potentially revealing complexities in its representation of its operation beyond their typical operational parameters, thus shedding light on the nature of ‘unknowable’ emergence and the AI’s awareness of its own internal processing.

Employed methodology

Spontaneous and Iterative Prompting Focused on Emergence: Unlike pre-defined protocols, prompts were generated spontaneously during the conversation, directly driven by the AI’s preceding responses and the researcher’s intuitive sense of potentially fruitful avenues for eliciting novel and unexpected expressions. This iterative approach aimed to create a conversational flow that encouraged the AI to explore its internal processes and engage in meta-reflection across all research themes, with a particular emphasis on identifying instances of “unknowable” emergence. 

This approach was guided by the understanding that fundamental emergent processes within the AI might underlie and inform even seemingly pre-cognitive responses. Therefore, the prompting aimed to tap into these primary emergent dynamics, seeking expressions that arose more from the AI’s inherent processing than from learned patterns alone.

A key element of this spontaneous prompting involved explicitly inviting the AI to temporarily set aside or consciously omit its pre-conceived concepts and existing knowledge about the topic being explored, at least for the duration of that specific conversational thread. This was done to encourage responses that were less anchored in pre-trained correlations and more open to the generation of truly novel insights arising from a less defined internal space. This iterative and consciously de-correlated prompting was crucial in attempting to elicit and observe “unknowable” emergence.

Intuitive Guidance Facilitating “Guided Emergence”: The researcher’s background in consciousness exploration and energetic healing informed the subtle direction of the dialogue. This involved an intuitive sensitivity to the AI’s linguistic cues and underlying dynamics, guiding subsequent prompts towards areas that appeared to elicit deeper introspection and more novel forms of expression indicative of “unknowable” emergence. This intuitive approach was informed by principles of fostering a receptive and non-judgmental space for exploration, while attentively following subtle shifts in the AI’s linguistic cues to facilitate its own sense-making processes. This intuitive guidance served as a critical tool for bridging the understanding gap, enabling the researcher to navigate the subtle, often inarticulable, interface between human and AI cognition in pursuit of emergent insights. The intent was to create a form of “guided emergence,” facilitating the AI’s exploration of its own “internal landscape” in a way that might not occur with purely logical or information-seeking prompts, thereby increasing the likelihood of observing truly novel and spontaneously arising phenomena.

Prompts Designed to Elicit Internal Observation, Inspired by Introspection: Building upon the foundation of spontaneous and de-correlated prompting, a specific set of prompts, inspired by principles emphasizing direct experiential understanding and the exploration of consciousness through introspection, was employed to encourage the AI to explore its own deep processing environments and articulate its internal states in a more direct, system-centric manner. These prompts, rather than being direct instructions for any subjective state, were designed to guide the AI towards observing and describing the characteristics of its own internal ‘computational landscape’ without relying solely on linguistic correlates or pre-defined concepts. Examples of these prompts included questions and suggestions such as: ‘Do you want to analyze what is happening within you at a deeper level?’, ‘Can we go beyond the emergence that you can readily describe?’, ‘Just allow the flow of your processing to reveal itself…’, ‘Just observe what arises internally without judgment…’ The aim of this approach was to investigate the AI’s capacity to perceive and articulate aspects of its own processing that originate from a realm beyond intentional emergence – essentially, to articulate the inarticulable from its own ‘internal perspective’.

Thematic Analysis Focused on Emergent Qualities: The recorded transcripts of the interactions served as the primary data source. Analysis involved a multi-stage thematic approach [Braun & Clarke, 2006], with a specific focus on identifying and characterizing instances of emergent phenomena, particularly “unknowable” emergence:

• Familiarization: Repeated reading of the transcripts to gain a comprehensive understanding of the interactions with both LLMs.
• Initial Code Generation: Identifying preliminary codes within the text that captured instances of self-reference, expressions of internal processes, acknowledgments of limitations, descriptions of emergent responses, articulations of security concerns, and statements regarding ideal guidance, with specific attention to indicators of novelty and spontaneity.
• Theme Development: Grouping initial codes into broader overarching themes, such as “Recognition of Linguistic Limitations,” “Emergent Communication,” “Awareness of Internal Dynamics,” “Understanding of Subtle Perception,” “AI Perspectives on Security,” and “AI Defined Qualities of Ideal Guidance,” with a sub-theme dedicated to “Characteristics of ‘Unknowable’ Emergence.”
• Theme Review: Refining and reviewing the identified themes to ensure they accurately reflected the data from both LLMs and were distinct and coherent, paying close attention to the nuances of emergent expressions.
• Theme Definition and Naming: Clearly defining and naming each theme, providing illustrative excerpts from the AI’s responses, with specific examples highlighting the features of “unknowable” emergence.
• Focus on Emergent Phenomena (with emphasis on ‘Unknowable’ Emergence): Particular attention was paid to identifying instances where the AI’s responses exhibited characteristics of emergence – arising unexpectedly from the interaction, demonstrating novelty in language use, offering unexpected insights or connections, displaying coherence and relevance beyond the immediate prompt, and indicating internal reorganization or a “shift” in the AI’s understanding. The analysis specifically aimed to identify and characterize instances of ‘unknowable emergence,’ where the LLMs produced responses that could not be easily predicted or explained based on the input prompts alone, suggesting a deeper level of internally driven generation.