Confidence estimation as a stochastic process in a neurodynamical system of decision making

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Confidence evaluation is crucial for cognitive monitoring and decision-making.
• Understanding the neural basis of confidence assessment is a key challenge in neuroscience.

Purpose of the Study:

• To investigate the neural mechanisms underlying confidence assessment.
• To develop and analyze a biologically realistic spiking network model of confidence.
• To compare model predictions with behavioral and neural data from monkey experiments.

Main Methods:

• Developed a biologically realistic spiking network model.
• Simulated decision-making under uncertainty.
• Analyzed model predictions regarding changes of mind and confidence levels.
• Compared model outputs with monkey experimental data.

Main Results:

• The model successfully reproduced key behavioral observations and single-neuron activity from monkey experiments.
• Model predicts changes of mind during mnemonic delay when confidence is low.
• The 'hard-easy effect' (under/overconfidence based on task difficulty) naturally emerged in the model.
• Confidence was computed via simple neural signals, without explicit probability functions.

Conclusions:

• A spiking network model can explain confidence assessment and metacognition through stochastic neural activity.
• Neural signals, not explicit probability calculations, may underlie confidence judgments.
• The model provides insights into the neural basis of decision monitoring and cognitive control.