Research On Human-Machine Hybrid Decision-Making Method Based On The Effectiveness Of Human Decision

With the development of artificial intelligence technology,the autonomous ability of machines has been continuously improved,enabling intelligent machines to be applied and developed in all walks of life.In this process,it is inevitable to encounter situations where intelligent machines cannot cope with the complexity and unpredictability of real tasks,and many systems will still require humans to continuously work with machines in supervision,goal setting,emergency response,etc.in the future.Therefore,it is particularly important and meaningful to study how to mix human decision-making and machine decision-making to achieve better decision-making effects in this scenario.In human-machine hybrid decision-making,whether human decision is effective,that is,whether human decision promotes the completion of tasks and effectively reflects human’s true intentions,affects the final decision-making performance from two aspects.On the one hand,the failure of one decision-making will lead to the decline of the hybrid performance.On the other hand,intelligent machines usually cannot directly know the intention of human,but need to infer the intention based on human decisions,and then make decisions to assist human to complete the intention.The failure of human decision-making may lead to the failure of intention inference,which in turn leads to the overall failure of machine decision-making and human-machine hybrid decision-making methods.Therefore,taking the reinforcement learning algorithms as the tool and the human-machine hybrid decision-making methods as the research object,this dissertation conducts research from two aspects:the full-time effective human decision-making and the non-full-time effective human decision-making,and proposes human-machine hybrid decision-making methods based on reinforcement learning to improve the decision performance.The research work of this dissertation mainly includes the following two aspects:(1)Aiming at the situation that human decisions are always effective,a humanmachine hybrid decision-making method based on reinforcement learning and following the principle of minimal intervention is proposed.It is noted that most of the existing methods only optimize the system performance,while ignoring the indicators of human satisfaction with the human-machine system.This method introduces the principle of minimal intervention into the human-machine hybrid decision-making,sets the adaptive threshold of human-machine decisions fusion,and provides the greatest help for humans with minimal intervention,so that the method can remain optimal in the changing environment,and improves the indicators of system performance and human satisfaction at the same time,providing a basic method for the subsequent optimization design scheme.(2)Aiming at the situation that human decisions may be invalid,a human-machine hybrid decision-making method based on reinforcement learning is proposed.It is noted that most of the existing methods assume that human decisions are always effective.This method considers the situation that human decisions may be ineffective.By judging the effectiveness of human decisions and identifying whether the human’s intention has changed,the method can make the machine complete the task alone when the human decision is ineffective and avoid ineffective decisions from damaging the system performance,so that the system can still complete the correct task goal in the case of long-term ineffective human decisions,which effectively improves the task success rate.