Research On Human-Machine Hybrid Decision-Making Based On Cognitive Model In Board Games

In recent years,research on artificial intelligence based on deep learning has made leapfrog progress,and the emergence of various network models has enabled machines to handle problems in various fields such as vision,language,search,and decision-making.In the process of executing tasks,it is inevitable to encounter problems that are difficult for machines to understand and solve,requiring human intervention and guidance to jointly complete the task.This type of scenario is defined as a humanmachine hybrid scenario.Under the current technical conditions,to achieve higher levels of intelligence and to improving the computational accuracy and generalization ability of machines,it is necessary to conduct in-depth research on human-machine hybrid decision-making scenarios and explore the paradigm of human-machine hybrid decision-making.This article takes the human-machine hybrid decision-making method in board game scenes as the research object,and combines the symbolic reasoning ability of cognitive models to propose a human-machine hybrid decision-making framework.The research work of this article mainly includes the following two aspects:First,to address the problem of pure deep learning driven systems tending to increase the number of model layers and abstractions,which cannot meet the requirements of rapid dynamic adjustment of the system,a pruning search algorithm was designed and implemented to introduce human strategies into the training process.This method can quickly and effectively improve the model’s ability after a few iterations of training.Second,to solve the problem of existing human-machine hybrid decision-making models lacking the ability to adjust strategies online,the final strategy being untraceable and difficult to understand,and the operator being unable to fully control the direction of the system.Unsupervised reinforcement learning was designed and implemented for skill training,combined with the cognitive model’s symbolic reasoning ability,strategies were arranged,and the final decision was delivered to the operator.In summary,this article proposes a design scheme for a humanmachine hybrid decision-making framework aimed at rapidly improving the capabilities of intelligent agents,enhancing the traceability and comprehensibility of human-machine hybrid decision-making systems,and preserving the ability of operators to dynamically adjust strategies.This effectively solves the above problems in pure deep learning driven systems and provides new ideas for subsequent research.