Theory protection: Do humans protect existing associative links?

Area of Science:

• Cognitive Psychology
• Learning and Memory
• Behavioral Neuroscience

Background:

• Traditional associative learning theories posit learning is proportional to prediction error.
• Spicer et al. (2020) proposed "theory protection": learning focuses on uncertain cues to preserve confident associations.
• Previous findings supported theory protection over prediction error models.

Purpose of the Study:

• To further test the theory protection model against prediction error accounts.
• To evaluate Holmes et al.'s (2019) prediction-error framework in explaining prior results.
• To investigate learning in inhibitory associative learning paradigms.

Main Methods:

• Four experiments involving inhibitory associative learning with letter cues and outcome presence/absence.
• Participants trained on cue-outcome associations, then tested with compound cues (e.g., AB-).
• Measured learning about cues with varying prediction error and confidence levels.

Main Results:

• Participants learned more about ambiguous cues (B) than cues with high prediction error (A) in compound trials (AB-).
• Experiments 3 and 4 showed greater learning for cues with no prediction error compared to those with large prediction errors.
• Experiment 4 findings aligned with theory protection and contradicted prediction error models.

Conclusions:

• The results support the theory protection model of associative learning.
• Selective learning about uncertain cues helps maintain confidence in established associations.
• This challenges standard prediction error accounts and alternative frameworks like Holmes et al.'s.