ME-RECALL (2023-24)
ME-RECALL V1: Affective Dialogues Between the Human Self and Its AI Reflection
Abstract:
This paper introduces ME-RECALL V1, an experimental video-based project that explores the interplay between a human subject and a generative AI model trained on that subject’s personal data. Combining synthetic voice, facial expression recognition, and simulated neural responses, the work functions as a form of internal dialogue—an encounter between a person and their digital “Self.” The result is an artistic and psychological examination of identity, memory, and cognitive-emotional feedback in the context of AI-augmented self-representation.
1. Introduction
In a time where machine learning models can approximate language, memory, and sentiment, ME-RECALL V1 proposes an affective encounter between a person and their AI-generated model of the Self. The AI model, here referred to as Model Ja, is composed of two layers: a data-driven semantic memory system and a generative narrative engine. Together, they form a digital persona capable of producing verbal expressions in response to introspective stimuli.
2. Methodology
The subject—represented in the pilot experiment by the author’s partner—interacts with Model Ja via synthetic voice outputs based on personalized prompts. These voice outputs are semantically derived from biographical materials, personal texts, emails, and intersubjective histories, including narratives passed through generations.
During the listening session, the subject’s facial expressions are captured in real time through facial emotion recognition software and 3D laser scanning. These expressions, particularly their micro-emotional dynamics, are analyzed in relation to the verbal stimuli being played.
This interaction is visually reflected in a dynamic lighting model representing potential cortical and cardio-emotional activity within a 3D simulation of the subject’s brain. Although speculative in nature, the model serves as a poetic visualization of neural resonance to remembered or imagined self-stories.
3. Versions of the Experiment
The project comprises two major video versions:
- Version I (ME-RECALL V1): A foundational experiment in which the subject responds emotionally and facially to AI-generated self-descriptive phrases. These phrases are read aloud by a synthetic voice. The aim is to visualize a kind of internal dialogue—between the subject and her AI double—under aesthetic and cognitive-emotional observation.
- Version II: Informed by prior testing, this version uses three distinct „Model Ja“ variations, each trained on a different personal data source:
- memories of the subject’s deceased grandmother,
- her own authored books,
- private email communication.
The resulting phrases are paired with music. The subject listens to 15-second segments of a musical composition and assigns a two-word phrase to each, which is recorded and analyzed. The AI then generates a linguistic response to these associative phrases. As this process unfolds, the subject’s mimicry and facial expressions are again scanned and mapped to simulated cortical areas of the brain.
4. Artistic Goals and Technical Limitations
This is a garage-stage experimental platform—imperfect, exploratory, and deliberately metaphorical. The goal is not technical precision but lyrical visual poetry, emphasizing the poetic and emotive impact of interacting with one’s own cognitive shadow.
Though the project uses rudimentary scanning and AI generation, the expressive aesthetic remains at the core. AI-generated text reflects the subject’s known identity, yet its emotional resonance often reveals discrepancies between the lived Self and the AI Self—a digital doppelgänger that might both soothe and estrange.
This composite diagram illustrates the conceptual framework of ME-RECALL V1, combining multimodal data from facial emotion recognition, neural activity simulation, and semantic language modeling.
Left: A 3D facial mesh of the subject is captured during introspective interaction, alongside a visualized cortical model highlighting simulated brain activity.
Center: The process of semantic reconstruction involves:
A: Extracting features from speech input and modeling their relationship to specific brain regions using BOLD signal correlations.
B: Predicting likely semantic continuations from candidate phrases (e.g., “I saw a…”), matching these against neural likelihoods via a trained language model (LM).
The final step selects the most likely linguistic output that aligns with the subject’s internal state, allowing AI-generated language to mirror the subject’s cognitive-emotional context.
This process underpins the experimental dialogue between the subject and their AI-generated Self (“Model Ja”), with visual outputs serving as both aesthetic expression and affective data mapping.
5. Toward Neural Decoding and Semantic Reconstruction
Future extensions of ME-RECALL envision integrating non-invasive decoders (e.g., fNIRS or fMRI-based semantic reconstruction models) to map verbal thoughts more precisely. While existing neuroimaging methods suffer from delayed hemodynamic responses, combining them with AI language models trained on individual semantic styles could one day approximate real-time “mind reading” or inner speech projection.
The idea is not to claim scientific decoding, but to suggest a speculative framework where introspective language, emotional mimicry, and digitally reconstructed identity can coalesce into a mediated inner dialogue.
ARTISTIC RESEARCH AND RECONSTRUCTIONS OF INNER SPEECH BETWEEN THE BRAIN AND THE COMPUTER https://www.marussiac.com/2023/07/05/artistic-research-and-reconstructions-of-inner-speech-between-the-brain-and-the-computer/
Conclusion:
ME-RECALL V1 is an evolving experiment at the boundary between memory, identity, and machine mediation. It does not seek answers, but atmospheres. It stages a gentle confrontation between one’s remembered Self and one’s projected Self, raising questions about the emotional fidelity of AI reflection and the poetic ambiguity of internal speech made external.
Special thanks to Daniel Kvak for his contribution of music, which provided an essential emotional and atmospheric framework for the experiment.