For a long time, scientists have sought to grasp how humans make decisions — whether we’re selecting what to eat for lunch or navigating high-stakes clinical treatments. Traditional computational models of decision-making often rest on fixed assumptions about how people learn from rewards and punishments. Yet these assumptions can struggle to reflect the wealthy, adaptive ways through which humans actually behave.
In an effort to tackle this complexity, Dezfouli and colleagues introduced a novel framework based on recurrent neural networks (RNNs) of their paper: Models that learn the way humans learn: The case of decision-making and its disorders.
Their approach goals to capture the nuanced processes behind human learning by training an RNN to mimic the subsequent motion a participant would soak up a decision-making task. Critically, the researchers tested this model on each healthy individuals and people living with unipolar or bipolar depression.
By comparing these groups, the study not only revealed the RNN’s capability to model complex behaviors more accurately than traditional reinforcement-learning methods, but in addition opened…
