Neural mechanism of subsequent decisions in multi-stage decision making

Background and Aim: In real life decision making, we frequently make a series of decisions to reach a certain goal. These decisions would be usually followed by no feedbacks. Therefore, the decision policy must be determined based on the internal estimate of the decision accuracy: Confidence. In the multi-stage decision making which reward is received when the series of correctly made decisions, the perceptual decision in each stage depended on not only the current sensory evidence but also to the previous confidence. In other words, previous confidence could adjust the speed-accuracy tradeoff of the subsequent decision. Methods: We designed a two-stage psychophysical experiment that contained two successive visual stimuli. Participates were required to make a binary decision regarding the direction of moving dots in each stage (right or left in the first stage and up or down for the second stage). For our computational approach, we employed an attractor network. This network consists of two populations, each selective to a choice, competing to reach a predefined threshold. We fitted our attractor model using behavioral data. For the second decision, we applied a top-down current, proportional to the previous decision's confidence, into both populations of the network. In other words, the second network received input as a function of the confidence in the first decision. Results: The theoretical and experimental accounts have been studied in multi-stage decisions. But yet the neural mechanisms of the confidence role in this type of decision making remain to be determined. Using attractor neural networks, we explained how confidence could make such alternations into the next decision. Conclusion: Our behavioral data revealed that the confidence of the first decision could modulate the reaction time of the second decision. In this way, higher confidence in the first decision caused higher reaction time in the second decision and vice versa. Our model simulations also showed that this effect could be produced by our main assumption in the attractor model: top-down current. Thus, we could conclude that in the absence of feedback, the confidence of the previous decision could act as an input current for the second decision network. This finding could enhance our understanding of the neural mechanisms of multi-stage decision making.

کلیدواژه ها:

Multi-stage decisions ، speed-accuracy tradeoff ، attractor network ، confidence