Tags

NeuroPersona Neural Networks Neural Plasticity Neuronale Netzwerke Reinforcement Learning Simulation Emotional

Share

Title

NeuroPersona: Simulation of Dynamic Cognitive Perspectives Version: 1.3

Ralf KrüMmel

Introduction

NeuroPersona is not a static analysis tool, but a bio-inspired simulation platform designed to replicate the dynamic and often variable processes of human cognition and emotion when engaging with a topic or question. Instead of producing a single, deterministic answer, NeuroPersona explores different plausible "thought paths" or "perspectives" that may emerge in response to a given input.

The system models interacting cognitive modules (creativity, criticism, simulation, etc.), an adaptive value system, a dynamic emotional state (based on the PAD model), and neural plasticity. Each simulation run represents a unique snapshot of a possible cognitive/affective state.

Core Philosophy:

Simulation Over Prediction NeuroPersona’s approach differs fundamentally from traditional AI models:

Variability as a Feature:

The system is inherently non-deterministic. Repeated runs with the same input will produce different yet internally plausible end states due to random weight initialization, stochastic activation noise, emotional dynamics, and path-dependent learning and plasticity processes. This mirrors the natural variability of human thinking. Emergent Perspectives: Each simulation run can be seen as a unique "thought process" prioritizing different aspects of a topic (sometimes innovation, sometimes safety, sometimes fundamentals). The result is not "right" or "wrong" — it is a simulated, plausible perspective. State Interpretation: The goal is to understand the final cognitive and emotional state within a single run: Which internal values dominate? Which cognitive modules are most active? What is the overall emotional tone? Is the state internally coherent (e.g., high creativity paired with high innovation value)? Apparent "inconsistencies" (e.g., high criticism activity but low safety value) are valid results, representing certain cognitive "stances" (such as decoupling analysis from prioritization). Exploration of Possibility Space: By simulating multiple runs (optionally with slightly varied parameters), you can explore the space of possible cognitive reactions to a topic, rather than focusing on a single definitive answer. Key Features Dynamic Input Processing: Utilizes a (simulated) "Perception Unit" to transform user prompts into structured data. Modular Cognitive Architecture: Simulates interacting modules: CortexCreativus: Idea generation and associative thinking. CortexCriticus: Analysis, evaluation, and risk assessment. SimulatrixNeuralis: Scenario thinking and mental simulation. LimbusAffektus: Dynamic emotional state modeling (Pleasure, Arousal, Dominance). MetaCognitio: Monitoring of network states and adaptive strategic adjustments (e.g., learning rate tuning). CortexSocialis: Modeling of social influence factors. Adaptive Value System: Internal values (e.g., innovation, safety, ethics) influence behavior and dynamically adjust through simulation. Neural Plasticity: Simulates structural changes (connection pruning and sprouting) and activity-dependent learning (Hebbian learning, reinforcement). Stochasticity: Purposeful use of randomness to emulate biological variability. Persistent Memory: Long-term storage and retrieval of relevant information via SQLite database. Reporting and Visualization: Generates detailed HTML reports and plots analyzing network dynamics and end states. Orchestration: The orchestrator.py script controls the complete workflow from prompt to final enriched response (optionally integrating an external LLM API like Gemini). Workflow Overview (orchestrator.py) Perception: A user prompt is converted into structured data (simulated CSV/DataFrame) via gemini_perception_unit.py. Cognition/Simulation: This data is fed into neuropersona_core.py. The network is initialized and simulated over a number of epochs, where learning, emotions, values, and plasticity interact. Synthesis (Optional): The results (report, structured data) are used to generate a final, contextually enriched answer, potentially involving an external LLM API (generate_final_response in orchestrator.py). Technical Components (neuropersona_core.py) Classes: Node, MemoryNode, ValueNode, Connection, specialized module classes (as listed above), PersistentMemoryManager. Core Functions: simulate_learning_cycle, calculate_value_adjustment, update_emotion_state, hebbian_learning, apply_reinforcement, prune_connections, sprout_connections, generate_final_report, create_html_report, plotting utilities. Parameters: Numerous constants control learning rates, decay rates, thresholds, emotional dynamics, and allow fine-tuning of system behavior.

Installation

Clone the Repository: git clone <repository-url> cd <repository-folder> Create a Virtual Environment (recommended): python -m venv venv

Windows

venv\Scripts\activate

MacOS/Linux

source venv/bin/activate Install Dependencies: (Make sure a requirements.txt exists) pip install -r requirements.txt

Required: pandas, numpy, matplotlib

Optional: networkx, tqdm, google-generativeai

Set API Key (Optional): If you want to use full orchestration with external LLM (e.g., Gemini):

Windows (PowerShell)

$env:GEMINI_API_KEY="YOUR_API_KEY"

Windows (CMD)

set GEMINI_API_KEY=YOUR_API_KEY

MacOS/Linux

export GEMINI_API_KEY='YOUR_API_KEY' Usage You can run a full simulation either through the GUI or directly through the orchestrator:

Start GUI:

python neuropersona_core.py (The GUI allows you to enter prompts, adjust key simulation parameters, and start the full workflow.) Run Orchestrator Directly: python orchestrator.py (The script will prompt you for input if run directly.) Interpreting Results Remember the core philosophy:

Focus on Single Run Interpretation:

Analyze the generated HTML report and plots for this specific simulation run. Look at the State: How do dominant categories, module activities, values, and emotions interact? Is the resulting "profile" internally coherent? Avoid Rigid Comparisons: Do not expect identical results between runs. Observe the range of plausible states. Value Saturation (Values at 1.0): Often a sign of rapid learning given limited data. Interpret this as "maximum relevance in this run," while recognizing that differentiation at the top end is lost. "Inconsistencies" are Valid: If, for example, Cortex Criticus is highly active while the Safety value remains low, it still represents a valid cognitive stance — not an error. Key Parameters (neuropersona_core.py Constants) DEFAULT_EPOCHS: Number of simulation cycles. DEFAULT_LEARNING_RATE: Base learning rate. DEFAULT_DECAY_RATE: Rate of activation/weight decay without input (important against saturation). VALUE_UPDATE_RATE: Speed of internal value adjustments. EMOTION_UPDATE_RATE, EMOTION_DECAY_TO_NEUTRAL: Control emotional dynamics. PRUNING_THRESHOLD, SPROUTING_THRESHOLD: Control structural plasticity. Fine-tuning these parameters (via GUI or settings files) affects the dynamics and differentiation capability of the system.

📖 Analogy for Non-Scientists: How NeuroPersona "Thinks" Imagine you ask a calculator, "What is 2 + 2?" — you always get "4". That’s a deterministic system.

NeuroPersona is different. Imagine asking a person a complex question like: "Should we heavily invest in a new, risky technology?"

On Day 1, feeling optimistic and inspired by success stories, the answer might be: "Absolutely! Huge opportunities — we must innovate!" (Focus: innovation, opportunity). On Day 2, after reading about similar failures and feeling cautious, the answer could be: "Careful! We must assess risks first and ensure ethical responsibility." (Focus: safety, ethics, risk assessment). On Day 3, feeling highly analytical, the person might say: "Let's first analyze the fundamentals and long-term efficiency impacts." (Focus: fundamentals, efficiency). All these answers are plausible human reactions, depending on internal "mood" (emotions), "priorities" (values), and currently salient information.

NeuroPersona replicates exactly this kind of variability:

It has internal "moods" (emotions) that shift. It has "priorities" (values) that evolve. Randomness (noise) influences thought paths. Learning continuously reshapes connections. Thus, if NeuroPersona delivers different outcomes across runs, it’s not an error — it’s a feature. It simulates different but coherent cognitive perspectives, illustrating the diversity of plausible cognitive-emotional responses to a problem.

Reference

https://app.readytensor.ai/publications/neuropersona-A9Nex0aLF2Lp