Predicting human decision making in psychological tasks with recurrent neural networks

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Human decision-making involves complex cognitive processes (beliefs, intentions, theory of mind), making prediction challenging.
• Traditional time series methods are insufficient for capturing the nuances of human psychological mechanisms in action sequences.
• Predicting human behavior in interactive scenarios requires advanced modeling techniques.

Purpose of the Study:

• To develop and apply a recurrent neural network (RNN) architecture, specifically long short-term memory (LSTM) networks, for predicting human decision-making trajectories.
• To evaluate the efficacy of LSTM networks in modeling human actions within the Iterated Prisoner's Dilemma and Iowa Gambling Task.
• To explore potential interpretations of LSTM network weights related to performance differences in decision-making strategies.

Main Methods:

• Collation of human behavioral data from 8 published Iterated Prisoner's Dilemma studies (168,386 decisions) and 10 Iowa Gambling Task studies (617 trajectories).
• Post-processing of data into behavioral trajectories for training and testing LSTM-based prediction networks.
• Comparative analysis against state-of-the-art methods for prediction accuracy in single-agent and multi-agent scenarios.

Main Results:

• LSTM networks demonstrated a clear advantage over existing methods in predicting human decision-making trajectories in both the Iowa Gambling Task and Iterated Prisoner's Dilemma.
• The model successfully predicted action sequences in both single-agent (Iowa Gambling Task) and multi-agent (Iterated Prisoner's Dilemma) settings.
• Analysis revealed differences in LSTM network weight distributions between top and poor performers, suggesting insights into strategy adoption.

Conclusions:

• Recurrent neural networks, particularly LSTMs, offer a powerful tool for modeling and predicting complex human decision-making processes.
• The findings highlight the potential of computational approaches to uncover underlying psychological mechanisms influencing human behavior in strategic interactions.
• Further research into LSTM weight interpretations may provide deeper understanding of cognitive strategies and performance variations.