Does signed prediction error drive declarative memory? Evidence from variable choice paradigms.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Machine Learning

Background:

• Prediction error (PE) quantifies the discrepancy between predictions and new information.
• PE can be signed (better/worse than expected) or unsigned (surprise).
• Previous research suggests PE influences one-shot learning and memory, with debate on signed vs. unsigned PE.

Purpose of the Study:

• To replicate and test the role of signed and unsigned prediction error in one-shot declarative learning using a variable choice paradigm.
• To identify and address potential confounds within the existing experimental paradigm.
• To evaluate the suitability of the variable choice paradigm for investigating PE in declarative memory.

Main Methods:

• Replication of De Loof et al.'s variable choice paradigm with and without financial reward (Experiments 1A-1B).
• Identification of a potential confounding factor in the original paradigm through simulations.
• Development and application of a modified paradigm to circumvent the identified confound (Experiment 2).

Main Results:

• Experiments 1A-1B showed memory was slightly better explained by unsigned PE than signed PE.
• A potential confound was identified, suggesting results could be explained without invoking PE.
• Results from the modified paradigm (Experiment 2) were inconsistent with the prediction error account.

Conclusions:

• Variable choice paradigms may be ill-suited for investigating the role of PE in one-shot declarative learning.
• The purported role of signed PE in declarative memory requires further empirical investigation.
• The findings challenge existing models that rely on PE to explain one-shot learning within this paradigm.