Second-Order Conditioning and Conditioned Inhibition in Different Moments of the Same Training: The Effect of A+ and AX− Trial Number

Abstract

The feature negative discrimination (A+/AX−) can result in X gaining excitatory properties (second-order conditioning, SOC) or in X gaining inhibitory properties (conditioned inhibition, CI), a challenging finding for most current associative learning theories. Research on the variables that modulate which of these phenomena would occur is scarce but has clearly identified the trial number as an important variable. In the set of experiments presented here, the effect of trial number was assessed in a magazine training task with rats as a function of both the conditioning sessions and the number of A+ and AX− trials per session, holding constant the total number of trials per session. The results indicated that SOC is most likely to be found at the beginning of training when there are many A+ and few AX− trials, and CI (as assessed by a retardation test) is most likely to be found at the end of training when there are few A+ and many AX− trials. Both phenomena were also found at different moments of training when the number of A+ trials was equal to the number of AX− trials. These results cannot be predicted by acquisition-focused associative models but can be predicted by theories that distinguish between learning and performance.

Conditioned inhibition, inhibitory learning, response inhibition, and inhibitory control: Outlining a conceptual clarification.

“…Such a further assumption seems to distance the Rescorla–Wagner model from its original simplicity, which is one of the main attributes that make it attractive for theorists. For this reason, observations of excitatory conditioning from feature-negative discrimination protocols are routinely regarded as a failure of the Rescorla–Wagner model 7 (e.g., Lee, 2021; Miller et al, 1995; Muñiz-Diez et al, 2021; Urcelay, 2017).…”

“…Another result regarding conditioned inhibition that cannot be accounted for by the original Rescorla–Wagner model is that the canonical feature-negative discrimination protocol (A+/AX−) may promote either excitatory or inhibitory associative states in the target stimulus (X; see Figure 3h). This outcome is known to depend on the number of excitatory (A+) and inhibitory (AX−) trials experienced by subjects (Stout et al, 2004; Yin et al, 1994) as well, as on other factors (see Muñiz-Diez et al, 2021). This is at odds with the prediction of the Rescorla–Wagner model that the initially neutral target stimulus would start accruing a negative associative state as soon as it is presented in compound with a mildly excitatory companion stimulus (A) in inhibitory trials.…”

Second-Order Conditioning in Humans

“…According to traditional associative models, X should accrue negative associative strength and become a conditioned inhibitor ( Rescorla and Wagner, 1972 ). Empirically, SOC is typically found early in training while conditioned inhibition emerges with additional training ( Herendeen and Anderson, 1968 ; Yin et al, 1994 ; Stout et al, 2004 ; Muñiz-Diez et al, 2021 ). Rescorla (1973a , 1980) proposed that both effects could be captured using a single dimension of associative strength if it is assumed that SOC is a transient and earlier phase of conditioned inhibition, with second-order excitatory learning gradually being erased or overridden by the developing inhibitory learning.…”

“…Future studies could test whether parameters known to promote SOC over conditioned inhibition have similar effects in humans. For instance, SOC tends to be found early in training, using a small number of training trials ( Herendeen and Anderson, 1968 ; Rashotte et al, 1981 ; Yin et al, 1994 ; Stout et al, 2004 ; Muñiz-Diez et al, 2021 ). While SOC has been demonstrated with simultaneous presentation of the XA compound ( Rescorla, 1973a ), serial presentation of the XA compound tends to be better than simultaneous presentation at promoting SOC ( Stout et al, 2004 ), while intermixing or blocking the feature negative contingencies seems to have no effect when the number of trials is small in both animals ( Yin et al, 1994 ) and humans ( Karazinov and Boakes, 2007 ).…”