Predictive coding appears to offer strong explanatory power. Does this strong explanatory power affect its falsifiability?
Is there a unified consensus on the specific brain indicators that signify the occurrence of predictive coding and those that indicate its absence?
Or, has a common understanding emerged regarding what neural activities we can observe to validate predictive coding’s correctness and what neural activities can demonstrate its inaccuracy?
Based on the predictive coding theory, I got two predictions:
Prediction 1: For unexpected stimulus input, the feedback signals from higher-level cortex to lower-level cortex should increase, and the feedforward signals from lower-level cortex to higher-level cortex should decrease. This is because the feedback signals convey the predictions of the higher-level model, and the feedforward signals convey the prediction errors, or the mismatch between the predictions and the input. When there is an unexpected input, the prediction errors are high, and the model needs to update its predictions to reduce them.
Prediction 2: Lower-level brain regions exhibit weaker responses to expected stimuli, and higher-level brain regions show stronger representation of expected stimuli. This is because the lower-level regions signal the prediction errors, or the difference between the actual input and the higher-level predictions, and the higher-level regions update their predictions to minimize the prediction errors. When the stimuli are expected, the prediction errors are low, and the lower-level responses are suppressed by the feedback signals from the higher-level regions. When the stimuli are unexpected, the prediction errors are high, and the lower-level responses are enhanced by the feedforward signals from the lower-level regions.
Therefore, if we observe in an experiment that：
Phenomenon 1: Lower-level brain regions exhibit stronger responses to expected stimuli, and higher-level brain regions show weaker representation of expected stimuli.
Or
Phenomenon 2: For unexpected stimulus input, the feedback signals from higher-level cortex to lower-level cortex decrease, and the feedforward signals from lower-level cortex to higher-level cortex increase.
Can we conclude that predictive coding is not happening in this experiment?
Or alternatively, if predictive coding is happening, does it always follow these two predictions?
Does this mean that predictive coding theory can be falsified in this case?
MANAGED BY INCF