IAB-i: Bi-Modal Artificial Intelligence Architecture

IAB-i: Bi-Modal Cognitive-Robotic Architecture for Simulating Internal State and Emotion-Informed Decision-Making in Conversational Agents Abstract This work introduces the IAB-i (Integrated Bi-modal Artificial Intelligence Architecture), a software framework designed to simulate the modularity of human cognition within a conversational agent. The primary objective is to move beyond the purely reactive paradigm of LargeLanguage Models (LLMs) by endowing the agent with an InternalState that directly influences its response generation. The IAB-i architecture operates on two parallel processing paths, inspired by psychological theories 1. The Conscious Path: Represented by a large language model (LLM,Phi-3 in this implementation) responsible for action generation andverbal output. 2. The Subconscious Path: Managed by a local vector database (ChromaDB), serving as a memory of past events and experiences that silently modulates the agent's internal state. The agent's core relies on four dynamic psychological scales (Security, Creativity, Connection, Integration) initialized and updated by semantic analysis (using spaCy) of the user stimulus (detecting danger, social opportunity). These scales quantify the situation and provide an emotionalcontext (CARE FREE , SUSPICION , prompt ALERT ) that is injected into the LLM's Innovation and Results: IAB-i demonstrates superior behavioral coherence, where an identical stimulus can lead to different actions depending on accumulated memory and emotional state.This model paves the way for more credible and resilient agents, capable of adapting not only their textual content but also their robotic outputs (voice modulation via gTTS and visual mood changes) to their simulated internal status. The repository contains the complete source code (Python/Flask) and modules, serving as a foundation for research in artificial cognitive architectures. So email to [email protected] to get it. References A. Tools and Implementation [1] Ollama (Local LLM Inference Engine) - Ollama: Run large language models locally. (https://ollama.com) [2] ChromaDB (AI Native Embedding Database) - ChromaDB: The AI-native embedding database. (https://www.trychroma.com) [3] spaCy (Text Analysis/NLP Framework) - Honnibal, M., Montani, S. (2017). spaCy: Industrial-strength Natural Language Processing in Python. (https://spacy.io) [4] Phi-3 (LLM used) - Microsoft. (2024). Phi-3 Technical Report. (https://arxiv.org/abs/2404.14219) B. Theoretical and Architectural Concepts [5] Cognitive Architectures (General) - Langley, P., Laird, J. E., & Rogers, S. (2009). Cognitive architectures: Research issues and challenges. Cognitive Science, 33(1), 141-149. [6] LLMs and RAG (Memory Integration) - Gao, Y., Ma, X., Geng, J., et al. (2024). Retrieval-Augmented Generation for Large Language Models: A Survey. arXiv preprint arXiv:2403.11189. [7] Conscious/Subconscious Models (Modular) - Baars, B. J. (2002). The conscious access hypothesis: Origins and recent evidence. Trends in Cognitive Sciences, 6(1), 30-36. [8] Model of Fear/Security - LeDoux, J. (1998). The Emotional Brain: The Mysterious Underpinnings of Emotional Life. Simon & Schuster.