Скачать или смотреть Generative AI collective behavior needs an interactionist paradigm

This document advocates for an interactionist paradigm to understand the collective behavior of agents based on large language models (LLMs). It highlights that comprehending these emergent behaviors is crucial due to their significant societal implications in terms of both risks and benefits. The authors argue that traditional frameworks, like multi-agent reinforcement learning for `tabula rasa` agents, are inadequate for LLM-based generative agents given their extensive pre-trained knowledge and ability for in-context learning. The proposed interactionist paradigm aims to systematically examine how prior knowledge and embedded values interact with social context to shape emergent phenomena in multi-agent generative AI systems. This new framework is built upon four crucial pillars: an interactionist theory focusing on individual traits and emergent dynamics, causal inference for system monitoring, information theory for quantitative studies, and a sociology of machines to transcend human-centric social theories. This comprehensive approach is vital for the design, deployment, and alignment of LLM-based collectives, addressing complexities that individual agent analysis cannot fully explain. The paradigm's relevance extends beyond current transformer-based models, emphasizing the necessity of accounting for both individual and social dimensions in future AI collectives.
#GenerativeAI #LLM #CollectiveBehavior #MultiAgentSystems #InteractionistParadigm #AISafety #SociologyOfMachines

paper - https://arxiv.org/pdf/2601.10567v1
subscribe - https://t.me/arxivpaper
donations:
USDT: 0xAA7B976c6A9A7ccC97A3B55B7fb353b6Cc8D1ef7
BTC: bc1q8972egrt38f5ye5klv3yye0996k2jjsz2zthpr
ETH: 0xAA7B976c6A9A7ccC97A3B55B7fb353b6Cc8D1ef7
SOL: DXnz1nd6oVm7evDJk25Z2wFSstEH8mcA1dzWDCVjUj9e
created with NotebookLM