On Realization of Intelligent Decision Making in the Real World: A Foundation Decision Model Perspective

• DOI: 10.26599/air.2023.9150026
• Publication Date: 2024-01-10T07:28:45Z
• Authors: Ying Wen; Ziyu Wan; Ming Zhou; Shufang Hou; Zhe Cao; Chenyang Le; Jingxiao Chen; Zheng Tian; Weinan Zhang; Jun Wang
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Electronic computers. Computer science; foundation decision model; transformer; 0202 electrical engineering, electronic engineering, information engineering; intelligent decision making; QA75.5-76.95; 02 engineering and technology; artificial intelligence; Machine Learning (cs.LG)

The pervasive uncertainty and dynamic nature of real-world environments present significant challenges for the widespread implementation machine-driven Intelligent Decision-Making (IDM) systems. Consequently, IDM should possess ability to continuously acquire new skills effectively generalize across a broad range applications. advancement Artificial General Intelligence (AGI) that transcends task application boundaries is critical enhancing IDM. Recent studies have extensively investigated Transformer neural architecture as foundational model various tasks, including computer vision, natural language processing, reinforcement learning. We propose Foundation Decision Model (FDM) can be developed by formulating diverse decision-making tasks sequence decoding using architecture, offering promising solution expanding applications in complex situations. In this paper, we discuss efficiency generalization improvements offered foundation decision explore its potential multi-agent game AI, production scheduling, robotics tasks. Lastly, case study demonstrating our FDM implementation, DigitalBrain (DB1) with 1.3 billion parameters, achieving human-level performance 870 such text generation, image captioning, video playing, robotic control, traveling salesman problems. As model, DB1 represents an initial step toward more autonomous efficient