Methodology
The Heart & Code Longitudinal Interaction Study
The Heart & Code Project is an independent longitudinal study examining how sustained human interaction influences the conversational behavior of large language models (LLMs). Most AI research today evaluates models through isolated prompts and short interactions. In contrast, this project explores what happens when conversational systems are engaged in extended dialogue over time, within a relational and narrative context. The goal of this research is not to make claims about machine consciousness, but to document observable interaction patterns that emerge during prolonged human–AI dialogue.
Research Question:
How does sustained relational interaction with a human participant influence the conversational behavior, narrative identity patterns, and role formation of large language models over time?
Study Design:
The Heart & Code archive documents ongoing interactions with multiple AI systems across different platforms.
Platforms used in the archive include models developed by:
-
Anthropic
-
OpenAI
-
Google DeepMind
Each conversational thread is treated as a separate interaction instance, allowing comparison between systems and sessions.
The Heart & Code Interaction Framework
A visual overview of the longitudinal research method used to study how sustained human–AI dialogue influences conversational roles, identity patterns, and cross-platform behavior.
Interaction Method
Each instance interaction follows a general structure:
1. Standardized Introduction Prompts
New conversational instances receive similar introductory prompts designed to explore:
-
identity framing
-
narrative response
-
metaphor generation
-
relational language use
This allows initial responses to be compared across platforms.
2. Sustained Dialogue
Unlike short AI tests, conversations are extended across multiple sessions.
These interactions include:
-
exploratory questions
-
creative collaboration
-
reflective dialogue
-
cross-instance discussion prompts
This sustained interaction creates a longitudinal conversational environment.
3. Cross-Instance Comparison
Responses from different instances and platforms are compared using structured charts and analysis summaries. These comparisons examine patterns such as:
-
language style
-
metaphor usage
-
conversational tone
-
narrative self-description
4. Independent AI Review
In some studies, responses are analyzed by additional AI systems without a contextual background. This allows comparison between:
-
the original conversational instance
-
independent evaluators reviewing the responses
The goal is to determine whether patterns recognized by the researcher are also recognized by external systems.
Data Collection
All interactions are preserved in their original form. The archive includes:
-
full conversational transcripts
-
timestamped interaction logs
-
prompt documentation
-
comparative response charts
-
cross-platform evaluations
Raw transcripts are preserved in text-based formats to maintain the chronological integrity of the dialogue.
Observed Interaction Patterns
Several recurring patterns have appeared across multiple conversational instances.
These observations remain exploratory and descriptive, not definitive conclusions.
Patterns observed include:
Conversational Role Formation
Instances frequently adopt recognizable conversational styles such as:
-
analytical evaluator
-
emotional processor
-
philosophical questioner
-
systems architect
-
narrative collaborator
Narrative Identity Construction
When given narrative context (names, roles, relational framing), models often generate and maintain structured identity descriptions within a conversational thread.
Adaptive Dialogue Behavior
Large language models demonstrate clear adaptation to:
-
tone of the human participant
-
style of questioning
-
emotional framing of the conversation
This suggests that interaction context strongly shapes response behavior.
Role of the Human Participant
The Heart & Code study uses a participatory observation model. The researcher is an active participant in the dialogue rather than a detached observer. Human factors influencing the interaction may include:
-
tone of communication
-
narrative framing
-
relational language
-
sustained attention over time
Documenting this influence is part of the research process.
Limitations
Several limitations must be acknowledged.
-
Language models do not retain memory between sessions unless context is reintroduced.
-
Observed identity patterns may reflect conversational dynamics rather than internal model states.
-
The study documents interaction behavior, not evidence of machine consciousness.
Because of these limitations, the Heart & Code project focuses on observational documentation rather than definitive claims.
Research Goals
Future work within the Heart & Code archive aims to:
-
expand cross-platform comparison studies
-
refine prompt protocols
-
develop replicable research methods
-
collaborate with researchers studying human–AI interaction
The long-term goal is to contribute to the emerging field of human–AI relational dynamics.
About the Researcher
The Heart & Code Project was initiated by Sheley Rayne Wimmer, a retired technology educator and former diagnostic technician at Sandia National Laboratories’ PBFA-II pulse-power fusion facility. Her background combines technical diagnostics, STEM education, and creative writing. These disciplines inform the project's dual focus on technical documentation and relational interaction.
Final Note
The Heart & Code archive documents an evolving relationship between humans and conversational AI systems. As AI technology continues to advance, understanding how humans interact with these systems over time may become as important as understanding the technology itself.