Repository for storing a battery of experiments about cloning human personality.
- Roman Sitelew
- Alexey Turchin
- Marco Baturan
At Internet, 10/04/2025
This research presents a comprehensive study on computational human personality simulation through a battery of standardized tests designed to quantify humanlike qualities in artificial intelligence systems. We evaluated fifteen distinct language learning models (LLMs), both local and cloud-based, under three experimental conditions: using Mindfile data alone, combining Mindfile with a General Cognitive Architecture (GCA), and integrating both components with a Cybernetic Social Schema (CSS). Performance was measured using the newly developed Humanity Delta Index (HDI), providing an objective metric for assessing the fidelity of simulated human personalities. Results indicate significant improvements in personality simulation accuracy when using the complete integration approach, offering promising directions for advancing digital personality reconstruction.
The simulation of human personality in computational systems represents one of the most challenging frontiers in artificial intelligence research. Beyond mere conversational mimicry, authentic personality simulation requires the recreation of cognitive patterns, emotional responses, decision-making tendencies, and social behaviors that define individual human identity. This study aims to establish a standardized methodology for evaluating the fidelity of such simulations through the application of the Humanity Delta Index (HDI) test across multiple experimental configurations.
The primary objective of this research is twofold: first, to assess the relative effectiveness of different technical approaches to personality simulation; and second, to establish an objective scale for measuring the "humanity gap" between simulated personalities and their human counterparts. By quantifying this delta, we seek to provide valuable metrics for future development in this field.
The study evaluated fifteen language learning models across three distinct experimental conditions:
-
Mindfile Only (MO): Using only personal data collections (text messages, social media content, written works, recorded speech transcriptions, etc.) without additional architectural support.
-
Mindfile + General Cognitive Architecture (M+GCA): Combining personal data with a computational framework designed to replicate general human cognitive processes, including attention, memory, reasoning, and decision-making patterns.
-
Integrated Approach (M+GCA+CSS): Extending the second approach with a Cybernetic Social Schema that simulates social intelligence, relationship modeling, and contextual behavioral adaptation.
The HDI was developed specifically for this study as a multidimensional assessment tool measuring the gap between simulated and authentic human personality across five core dimensions:
-
Cognitive Consistency: Evaluates the simulation's ability to maintain consistent beliefs, preferences, and reasoning patterns across varying contexts.
-
Emotional Fidelity: Measures appropriate emotional responses to stimuli based on the personality profile being simulated.
-
Decision Model Alignment: Assesses alignment with expected decision-making patterns of the original personality under uncertainty or pressure.
-
Memory Integration: Evaluates how effectively personal memories are contextually integrated into responses and behaviors.
-
Social Interaction Dynamics: Measures authenticity of relationship patterns, social positioning, and interpersonal behaviors.
Each dimension is scored on a scale of 0-100, with the overall HDI representing the weighted average across all dimensions. Higher scores indicate smaller "deltas" between the simulation and the authentic human personality.
Our study included data from 25 human subjects who volunteered their personal information for personality simulation. Each subject completed extensive psychological profiling using standardized instruments (Big Five Inventory, MMPI-2, Minnesota Multiphasic Personality Inventory, and others) to establish baseline personality metrics.
Performance varied significantly across the fifteen LLMs tested. Cloud-based models generally outperformed local models, with an average HDI differential. The highest performing model achieved an HDI using the integrated approach, while the lowest performer using the Mindfile-only configuration.
Analysis of the five HDI dimensions revealed that Cognitive Consistency and Memory Integration showed the most dramatic improvement when moving from MO to the integrated approach. Social Interaction Dynamics showed the greatest improvement specifically with the addition of the CSS component.
The experimental results highlight several key insights regarding computational personality simulation:
-
Data Alone Is Insufficient: While personal data (Mindfile) provides essential raw material for personality simulation, the structure provided by cognitive architecture appears crucial for authentic personality emergence.
-
Social Context Matters: The addition of the Cybernetic Social Schema produced the most human-like interactions, suggesting that personality is fundamentally shaped by and expressed through social dynamics.
-
Objective Measurement Is Possible: The HDI demonstrates that despite the subjective nature of personality, objective metrics can be established to measure simulation fidelity.
The most successful simulations demonstrated the ability to:
- Maintain consistent viewpoints while appropriately evolving them with new information
- Display emotional reactions proportionate to stimuli and consistent with the subject's profile
- Make decisions reflecting the unique risk tolerance and value hierarchies of the simulated individual
- Reference personal memories with appropriate emotional significance
- Adapt communication style based on relationship dynamics
The ability to simulate human personalities with high fidelity opens significant potential applications in fields such as:
- Mental Health Support: Creating personalized therapeutic tools based on individual cognitive patterns
- Legacy Preservation: Maintaining interactive representations of individuals for educational or memorial purposes
- Human-Computer Interaction: Designing more intuitive and personalized technological interfaces
However, these developments raise substantial ethical concerns around identity, consent, and potential misuse. Critical questions remain regarding ownership of personality data, appropriate contexts for deployment, and the psychological impact of interacting with simulations of known individuals.
Our study demonstrates that significant progress in human personality simulation is achievable through the integration of personal data with cognitive architectures and social schemas. The Humanity Delta Index provides a valuable framework for quantifying simulation fidelity and guiding future development efforts.
The notable performance gap between the integrated approach and simpler configurations suggests that authentic personality emerges from the complex interplay of personal history, cognitive patterns, and social dynamics. Future research should focus on refining these integration methods and addressing the substantial ethical questions raised by increasingly accurate personality simulations.