GenAINet: Enabling Wireless Collective Intelligence via Knowledge Transfer and Reasoning

• DOI: 10.48550/arxiv.2402.16631
• Publication Date: 2024-02-26
• Authors: Hang Zou; Qiyang Zhao; Lina Bariah; Yu Tian; Mehdi Bennis; Samson Lasaulce; Mérouane Debbah; Faouzi Bader
• Keywords: Collective Intelligence; Knowledge Transfer

Generative artificial intelligence (GenAI) and communication networks are expected to have groundbreaking synergies in 6G. Connecting GenAI agents over a wireless network can potentially unleash the power of collective pave way for general (AGI). However, current designed as "data pipe" not suited accommodate leverage GenAI. In this paper, we propose GenAINet framework which distributed communicate knowledge (high-level concepts or abstracts) accomplish arbitrary tasks. We first provide architecture integrating capabilities manage both protocols applications. Building on this, investigate effective reasoning problems by proposing semantic-native GenAINet. Specifically, extract semantic from multi-modal raw data, build knowledgebase representing their relations, is retrieved models planning reasoning. Under paradigm, an agent learn fast other agents' experience making better decisions with efficient communications. Furthermore, conduct two case studies where device query, show that extracting transferring improve query accuracy reduced communication; control, via collaborative Finally, address developing hierarchical level Telecom world model key path towards intelligence.