Autonomous Multi-Agent Systems for Enterprise Decision-Making
TL;DR
An integrated conceptual framework is presented which maps layers of the MAS architecture to decision postures in the enterprise, a cross domain performance synthesis, and a research agenda for the next generation of enterprise-scale autonomous agent systems are presented.
Abstract
As enterprise settings evolve in complexity and volume, there is a need for systems to make decisions that are autonomous, adaptive, and at scale. Self-directed software agents operating in shared organisational environments that can perceive, reason and engage in action to collectively fulfill this demand have evolved as an appealing paradigm both theoretically and empirically, termed as autonomous multi-agent systems (MAS). Even though there exists much research work with regards to architecture and design of MAS, empirical performance comparisons have not been done systematically between its performance in different enterprise decision domains. This article is an empirical study on autonomous MAS in enterprise decision-making in three main areas such as supply chain management, manufacturing and industrial operations, and enterprise resource planning (ERP) and strategic planning. This study leverages state-of-the-art MAS research and successful benchmarks synthesised from published experimental research to assess the performance of MAS against traditional decision support systems (DSS) using four metrics: decision cycle time, operational throughput, forecast accuracy and adaptability to disruptions. The data shows a 15-47% performance improvement for MAS-based systems compared with traditional DSS systems in the areas of supply chain and manufacturing; and an 8-38% improvement in ERP and strategic planning scenarios, although smaller, but still significant. Hybrid BDI architectures and market-like coordination mechanisms proved best among design configurations. Key adoption challenges such as scalability beyond 1,000 agents legacy system interoperability and explainability gaps are identified and discussed. The article presents an integrated conceptual framework which maps layers of the MAS architecture to decision postures in the enterprise, a cross domain performance synthesis, and a research agenda for the next generation of enterprise-scale autonomous agent systems
