This website requires JavaScript.

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

Ying WenZiyu WanMing Zhou ...+6 Jun Wang
Dec 2022
Our situated environment is full of uncertainty and highly dynamic, thushindering the widespread adoption of machine-led Intelligent Decision-Making(IDM) in real world scenarios. This means IDM should have the capability ofcontinuously learning new skills and efficiently generalizing across widerapplications. IDM benefits from any new approaches and theoreticalbreakthroughs that exhibit Artificial General Intelligence (AGI) breaking thebarriers between tasks and applications. Recent research has well-examinedneural architecture, Transformer, as a backbone foundation model and itsgeneralization to various tasks, including computer vision, natural languageprocessing, and reinforcement learning. We therefore argue that a foundationdecision model (FDM) can be established by formulating various decision-makingtasks as a sequence decoding task using the Transformer architecture; thiswould be a promising solution to advance the applications of IDM in morecomplex real world tasks. In this paper, we elaborate on how a foundationdecision model improves the efficiency and generalization of IDM. We alsodiscuss potential applications of a FDM in multi-agent game AI, productionscheduling, and robotics tasks. Finally, through a case study, we demonstrateour realization of the FDM, DigitalBrain (DB1) with 1.2 billion parameters,which achieves human-level performance over 453 tasks, including textgeneration, images caption, video games playing, robotic control, and travelingsalesman problems. As a foundation decision model, DB1 would be a baby steptowards more autonomous and efficient real world IDM applications.