Mechanisms of Adaptive Choice History Biases in the Human Brain

许多行为学研究表明，人类的决策深受强烈偏见的影响。这些偏见的一个重要来源是之前选择的历史。例如，在判断模棱两可的证据时，人们经常比偶然预期更频繁地重复相同的选择。最近在认知、系统和理论神经科学方面的工作，从一系列决策过程中大脑内部状态的动态变化中，提供了对选择历史偏差出现的一瞥。具体来说，我的团队在DFG研究资助的第一个资助期的工作导致了两个假设，我们的目标是在该资助的第二个资助期进行测试：(i)选择历史偏见对内部大脑状态的影响是由随后决策中与选择相关的神经信号的缓慢和情境依赖积累介导的；（ii）由此产生的神经偏差通过对选择性注意的偏倚，塑造了对新决策相关证据的解释。为此，我们将对选择历史偏差和选择性注意进行系统的操作，然后在完整的人类大脑中，在行为、算法和神经水平上量化大脑内部状态的变化。我们将详细分析选择行为的历史模式，这些行为模式的替代算法模型的拟合，以及大规模脑动力学的瞳孔测量和脑磁图（MEG）记录。我们将重点关注不确定性下初级感知决策中的选择历史偏差，使我们能够利用表征良好的神经群体测量方法来编码感觉输入（“证据”）、不断演变的决策和最终的行动计划。反过来，这将有助于在后续决策之间的这些测量中识别偏见，这在行为层面上调解了选择历史偏见。我们还计划基于对选择历史偏见的新兴理解，为选择历史偏见开发一个综合的理论框架。该框架将在不同研究途径的见解之间架起桥梁：从计算层面的见解以及潜在的动态神经机制，在局部微电路和全脑网络的层面；以及从感性和基于价值的决策研究中获得的见解。综上所述，该项目将深入了解不确定性下人类判断和决策的生物学基础。

Numerous behavioral studies have revealed that human decision-making is profoundly shaped by strong biases. One important source of those biases is the history of previous choices. For example, when judging ambiguous evidence, people often repeat the same choice more frequently than expected by chance. Recent work in cognitive, systems, and theoretical neuroscience provides a glimpse into the emergence of choice history biases from the dynamics of internal brain states during sequences of decisions. Specifically, my team’s work during the first funding period of this DFG research grant has led to two hypotheses, which we aim to test in the second funding period of this grant: (i) the effect of choice history biases on internal brain state is mediated by the slow, and context-dependent, accumulation of choice-related neural signals across subsequent decisions; and (ii) the resulting neural biases shape the interpretation of new decision-relevant evidence, by biasing selective attention. To this end, we will perform systematic manipulations of choice history biases as well as of selective attention, and then quantify the resulting changes in internal brain states at the behavioral, algorithmic, and neural levels, in the intact human brain. We will perform detailed analyses of history patterns in choice behavior, fits of alternative algorithmic models to these behavioral patterns, and pupillometric and magnetoencephalographic (MEG) recordings of large-scale brain dynamics. We will focus on choice history biases during elementary perceptual decisions under uncertainty, enabling us to exploit well-characterized neural population measures of the encoding of sensory input (“evidence”), the evolving decision, and the resulting plan to act. This, in turn, will facilitate the identification of biases in these measures between subsequent decisions, which mediate the choice history bias at the behavioral level. We also plan to develop an integrated theoretical framework for choice history bias, based on the emerging understanding of choice history bias. The framework will bridge between insights from distinct avenues of research: insights from the levels of computation as well as underlying dynamical neural mechanisms, at the levels of local microcircuits and brain-wide networks; and insights from studies of perceptual and value-based decisions. Taken together, this project will provide deep insights into the biological basis of human judgment and decision-making under uncertainty.