【作者】 余柳涛； 陈清华； 吴思； 王大辉；

【Author】 Liutao Yu;Qinghua Chen;Si Wu;Dahui Wang;School of Systems Science Beijing Normal University;School of Brain and Cognitive Sciences Beijing Normal University;National Key Laboratory of Cognitive Neuroscience and Learning Beijing Normal University;

【机构】 北京师范大学系统科学学院； 北京师范大学脑与认知科学研究院； 北京师范大学认知神经科学与学习国家重点实验室；

【摘要】 在人类社会生活中,通常需要综合多个个体的意见以形成最后的群体决策,其决策结果往往优于个体决策,这种"群体智能"是人类社会发展的重要基础。群体决策过程中的信息如何整合以获得更具优势的决策长期以来都是众多研究领域的热点,但至今还没有得到实质性的进展。2010年,Bahrami等人基于双人知觉决策这一群体决策研究的简单范式,通过实验和理论相结合的方法发现:从自信度分享的角度看,双人决策可能是个人决策的贝叶斯优化形式的整合。贝叶斯优化形式的整合被证实广泛存在于人类大脑的各种信息处理过程中,如人类大脑通过贝叶斯优化的形式整合不同知觉模态的信息来获取对外界事物更为准确的感知(最小化感知的变差)。但双人决策中的信息交互涉及复杂的神经机制,认为它是贝叶斯优化整合还缺乏充足的证据。本文以双人信息整合为研究对象,设计了简单的颜色感知任务,检验其是否也是贝叶斯优化形式的整合。实验采用的视觉刺激如下:屏幕中心首先显示一个圆形的色斑,呈现3 s;在1 s的空屏之后呈现一个颜色渐变的色环,被试通过鼠标点击汇报颜色感知。实验中,两个被试首先在独立地观察同样的刺激后独立地汇报自身的颜色感知;随后两个被试同时观察同样的刺激,在进行口头讨论之后汇报群体的颜色感知。实验结果表明:1)颜色感知误差服从高斯分布,而且双人汇报误差的方差小于两个被试独立汇报的误差;2)双人汇报误差的方差与个体汇报误差的方差的关系并不完全符合贝叶斯优化整合的理论预期,大部分时候劣于贝叶斯理论预期(6组被试),但也有时候更优(2组被试)。基于数据分析的结果,文章建立了一个双人动态交互决策模型:在汇报双人决策结果之前两个人可进行多次交互,每个个体依次综合考虑自己的独立决策和上一次经过交互形成的临时双人决策以形成新的双人决策,直到对方无异议时讨论终止。在这些简单可信的假设下,模型较好地重现了实验数据,对模型的分析也可以揭示如何更有效地进行双人决策。更多还原

【Abstract】 In human society,people usually need to integrate many individuals’ opinions to form a group decision,which is usually better than individuals’ decisions,called "group intelligence",and it is a basis of the development of human society.The information integration process in group decision making is a hot topic in many research fields,but no substantive progress have been made for now.Applying a dyadic perceptual discrimination task as a simple way to explore group decision making,Bahrami et al.2010 demonstrated that dyadic decision making in their task can be a Bayesian optimal integration process from an aspect of confidence sharing experimentally and theoretically.Bayesian optimal integration have been observed in kinds of information processes in our brain.For example,human brain integrates information from different sensory modalities to get more accurate perception about the external world(minimize the variation of perception).However,dyadic interaction involves complex neural mechanisms,so the question whether information integration during dyadic decision making is the same as the mechanism underlying multisensory integration still needs more work.In this research,we applied a simple color estimation task to study how two persons integrate their individual estimations to form a dyadic estimation,and whether this integration is Bayesian optimal or not.The stimuli used in a task are as follows:firstly,a colored disc appears in the center of the screen which lasts for 3 s;and then,after al s’ delay,a graduated color wheel appears on the screen,and the participant needs to report his/her estimation by a single mouse click.In a task,two participants first report their estimations independently after observing the same stimuli independently;then one randomly chosen participant reports the dyadic estimation after oral discussions.Experimental results show that:1) the errors of color estimation obey a Gaussian distribution,and variation of the dyadic report is smaller than both participants’ reports;2) the relationship between variation of the dyadic report and those of the two individuals’ reports doesn’t meet the expectations of a Bayesian optimal process,with 6 dyads’ results worse and 2 dyads’ results better than corresponding theoretical expectations.To explore the integration mechanism underlying the dyadic color estimation task,we built a model taking the dynamic interaction process into account based on experimental observations.The model assumes that there exist multiple interactions before consensus reached,and each member makes new report based on his/her own estimation and dyadic estimation from previous round as a new dyadic estimation,and the interaction process ends until the teammate agree with the new dyadic estimation.With these simple and reliable hypotheses,the model well repeated experimental observations.Moreover,analysis of the model can help to reveal how to make better dyadic decisions.更多还原