Peak Shift, explained.
Peak shift is a displacement of strongest generalised responding away from a nonreinforced stimulus and beyond a reinforced stimulus under some discrimination-training conditions.
Why it happens
A Gaussian excitatory gradient is centred at 60, an inhibitory gradient at 40. The displayed response is max(0, excitatory−a×inhibitory). Change shared width and inhibitory strength; the peak is the maximum over integer stimulus positions 0–100.
Peak shift is a displacement of strongest generalised responding away from a nonreinforced stimulus and beyond a reinforced stimulus under some discrimination-training conditions.
Read the result
The reinforced cue stays at sixty. Only the modelled response maximum moves as the nearby inhibitory gradient suppresses one side more strongly.
A worked example
A response stronger than the trained cue
Set inhibitory strength to zero. The excitatory Gaussian peaks at the rewarded stimulus sixty.
Increase inhibition centred at forty while retaining the same excitatory curve and compare the new response maximum.
Change the common width to explore overlap. The shift comes from the teaching functions' interaction, not from measured preferences collected here.
OPTIONAL DEEPER DETAILGo deeper: inside the model
Inside this model
A Gaussian excitatory gradient is centred at 60, an inhibitory gradient at 40. The displayed response is max(0, excitatory−a×inhibitory). Change shared width and inhibitory strength; the peak is the maximum over integer stimulus positions 0–100.
Where this idea is useful
A practical use
Learning to distinguish two nearby cues can change responses to new cues, rather than merely reproducing the trained preference.
A common misconception
“The most rewarded example must produce the strongest response to every new cue.”
Generalisation after discrimination training can peak beyond the reinforced cue. Whether that happens depends on training and testing conditions.
What this explanation leaves out
- The gradients are authored teaching functions, not animal or human measurements. Peak shift depends on training and testing conditions, and gradient subtraction is one model rather than a universal explanation.
Are these curves a fitted account of Hanson's original data?
No. They are authored Gaussian teaching curves illustrating one gradient-interaction explanation; they do not reproduce the original measurements or every theory of generalisation.
How could learning what to avoid change a response to something never directly rewarded?
Associated thinkers
Associations marked provisional are awaiting source review.
Further reading
Explore the original research or the teaching reference behind this experiment.