Artificial Emotion Framework

A potential integral component of the Bicameral AGI Project: Designing Affect for Intelligent Systems

This repository implements an artificial emotion and semantic memory system integrated into a Flask web application, utilizing OpenAI's API and sentence embeddings for semantic memory retrieval.

Click the image to watch the demo on YouTube

Overview

The system consists of two main components:

1. Emotional System

Simulates an evolving emotional state with dynamic intensity influenced by user interaction, memories, and internal background thoughts. Emotion dynamics include:

• Emotion dimensions: joy, sadness, anger, fear, surprise, trust, disgust, anticipation
• Time-based Decay: Gradually reduces emotion intensity over time to simulate natural emotional transitions.
• Micro-fluctuations and Noise: Adds subtle, randomized variations for realistic emotional states.
• Thought Generation: Background generation of contextually influenced thoughts using the OpenAI API, which subtly affect emotional states.
• Emotion-driven Responses: User interactions affect and are influenced by the current emotional state, producing naturally varying responses.

2. Memory System

Stores and retrieves memories based on semantic similarity, with special handling for emotionally significant memories.

• Semantic Memory: Uses embeddings (via sentence-transformers) to store and retrieve semantically related memories.
• Emotional Memories: Associates memories with significant emotional responses.
• Dynamic Memory Management: Memories are maintained based on recency, relevance, and recall frequency, limiting memory size for efficient retrieval.

Installation & Setup

Clone the repository:

git clone https://github.com/alanh90/BICA-ArtificialEmotions.git
cd BICA-ArtificialEmotions

Install dependencies:

pip install -r requirements.txt

Configure environment variables by creating a .env file:

OPENAI_API_KEY=your_openai_api_key_here

Run the Flask server:

python app.py

Future Directions & Potential Improvements

• Emotion Vector Representation: Implement emotions as vectors in latent emotional spaces for increased flexibility and realism.
• Emotional Conflict Resolution: Introduce methods for resolving conflicting emotions, such as weighted blending or context-based prioritization.
• Emotion-Influenced LLM Attention: Incorporate emotional states directly into attention mechanisms or prompts to more effectively guide AI responses.
• Optimizations: Cache embeddings for faster semantic searches, reduce computational overhead in background emotional updates.
• Mutualistic Human-AI Interaction: Further develop methods for adaptive emotional responses specifically tailored to improve human-AI interactions and empathy.

Future Directions & Improvements

• Develop adaptive mechanisms allowing the AI to learn and modify its emotional responses based on experiences.
• Expand ethical considerations, ensuring transparency and interpretability of artificial emotional states.
• Implement dynamic emotion vector spaces for greater flexibility and realism.

License

Distributed under the MIT License. See LICENSE for more details.

---

This implementation serves as a foundational structure, aligning closely with the theoretical "Latent Emotion Spaces" concept. Future iterations will deepen these capabilities to create more nuanced, adaptive, and realistic emotional interactions.