Human-machine Interaction Mechanisms Design And Prediction Efficiency In The Hybrid Human-machine Information Market

Accurate predictions for future events are valuable for the individuals,firms,and government to make important decisions in the areas of economics and finance.To improve forecasting efficiency,decision-makers can train machine learning models based on historical data to identify potential patterns.While machine models contribute to decision making,humans can also make inference and judgment based on domain knowledge and tacit experience.So,human-machine complementarity forms the basis for developing a hybrid human-machine system.Traditional approaches to integrate humans and machines intelligence generally include machines as assistants for humans,ignoring the value of their independent decisions in a competitive context which may improve prediction accuracy.In this study,market mechanism is adopted to integrate human and machine intelligence.Specifically,the participants,humans and machine models,are introduced into information market,and they buy and sell contracts that are related to future events.The market price of specific contract represents the human-machine collective expectations of the probability of corresponding event.Currently,there is no well accepted design principles and theoretical framework to scientifically develop a hybrid human-machine information market.This paper studies the novel hybrid prediction system and its forecasting efficiency from the perspective of human-machine interaction mechanism design.There are three main research motivations for this study.First,the trading behavior of machines in the market itself is an implicit interaction between humans and machines.Machines make decisions based on their own private information,which is not affected by the fluctuations of market prices.Hence,machines’ transactions that are not interfered by emotions can improve information market which only includes human traders.Second,based on business ethics,the hybrid humanmachine information markets are required to disclose to participants whether machines exist in the market.Therefore,compared with the financial market with algorithmic trading,human traders can easily perceive the presence of machines in the hybrid human-machine information market.Finally,in order to further improve the forecasting efficiency of the hybrid human-machine information market,it is necessary to calibrate the market price based on the complementary advantages of humans and machines by building a virtual hybrid human-machine information market.Regardless of whether the human traders feel the presence of machines in the information market,the emotion-free transactions made by machines can optimize the traditional information markets that involve only humans.When the existence of the machine is disclosed,there is an explicit interaction between the human and the machine.In the context of explicit interaction,the disclosure of machine presence and other human-machine interaction designs will directly affect the psychological states and behavior of the human traders.Based on multi-agent modeling simulation and online experiment methods,this paper investigates four related topics of human-machine interaction mechanisms in the hybrid humanmachine information market which uses stock fluctuation prediction in financial market as the forecasting task.First,this paper designs the hybrid human-machine information market and experimental system based on the design science principles.Following the design science approach,this study firstly summarizes the design principles for improving the hybrid human-machine information market.In addition,we conduct multi-agent modeling and simulation studies to evaluate the effectiveness of introducing machine transactions.The simulation results verified that the implicit interaction between humans and machines benefits the information market in three ways: it improves forecast efficiency,reduces market price deviation,and weakens the interference of noisy traders.The results indicated that the performance of hybrid human-machine information market remain robust under different parameter settings of machine trading time period,machine number ratio,and others.Based on the basis of the above design principles and simulation evaluation,this paper further develops a basic experimental platform for subsequent experiments on human-machine interaction mechanism design.Second,this study investigates the impact of disclosing machine presence on the efficiency of human-machine hybrid information market.The introduction of machine trading can bring machine intelligence.However,the impact of disclosure of machine presence on the participants’ decision-making quality and the overall efficiency of the market is an important human-machine interaction research topic.Based on decision theory,situation awareness,and autonomous agent teammatelikeness theory,this paper uncovers the impacts of disclosing machine presence by conducting an online experiment.The results showed that the introduction of machine trading will result in intense competition in the market which will reduce the deliberate effort of human traders,and reduce their belief updating frequencies meanwhile.Therefore,how to reduce the negative effects of machine presence and harmonize the relationship between people and machines has become an urgent problem to be solved.Third,this paper designs the human-machine interaction goal structure and the machine reward sharing method in the hybrid human-machine information market,which may harmonize the human-machine relationship.The experimental research results showed that machines sharing the reward can motivate experienced human traders continuously participate in the prediction market,and improve the overall decision-making quality.Human traders’ decision-making quality is much higher when humans and machines have a cooperative goal.In addition,when they have cooperative goal,humans perceive less threat from machines.The results also suggested that performance-based allocation approach amplify the incentive effect of machine reward-sharing design.Fourth,this study designs a price calibration tool for the market price improvement from human-machine complementary perspective.Using the real online experimental data set,we further filter participants dynamically based on human-machine complementary indicators to build an artificial hybrid humanmachine information market.The results showed that in the hybrid human-machine information market,price calibration based on the complementary advantages of humans and machines can improve the price efficiency in the information market,especially in dynamic calibration setting.The results also suggested that humans’ complementarity with machines are positively correlated with both of their historical decision-making quality and prediction experience.Based on the above studies and conclusions,we contribute to this area by verifying the effectiveness of the implicit human-machine interaction mechanism and designing reasonable explicit human-machine interaction mechanisms.Specially,this study makes the following three contributions.Firstly,based on the principles of design science,the design elements of the human-machine interaction mechanism in the hybrid human-machine information market are proposed.The effectiveness of the hybrid human-machine information market is verified based on multi-agent simulations,which proves that the human-machine hybrid information market is more efficient than the traditional information market composed of only human traders,and the hybrid market is more resistant to the influence of noise traders.Secondly,this research reveals the potential mechanisms of the negative effect of disclosing machines’ presence on humans’ decision-making quality and forecasting efficiency of the hybrid information market.Specifically,the disclosure of machines’ presence increases the competition and reduces human traders’ deliberation efforts.Moreover,it is verified that the appropriate design of humanmachine interaction goal structure and machines’ reward-sharing behavior may harmonize human-machine relationships and improve the quality of humans’ decision-making.Thirdly,from the perspective of human-machine complementarity,a market price calibration tool for the hybrid human-machine prediction market is designed,which is verified to be helpful for improving forecasting efficiency significantly.This finding not only provides a new perspective for the calibration method of the hybrid human-machine prediction system,but also reveals that human-machine complementarity is the foundation of hybrid human-machine information market.