Variants 1 and 3 also realize such function by providing bonuses
for actions leading to both stimuli (see Figure S2). Only providing
bonus for sign-tracking behaviour – as in Dayan’s model (Variant
2) – does not fit well with the attribution of incentive salience to
both stimuli. It would suggest that we should not observe incentive
salience towards the magazine in any rats, which is in discrepancy
with the experimental data. Thus, the important mechanism here
is that stimuli are not processed differently. Any stimulus is
attributed with its respective bonus, which is pertinent in regard to
the attribution of incentive salience.
Conditioned Reinforcement Effect (CRE). An important
question about the difference in observed behaviours is about the
properties acquired by the lever that makes it more attractive to
STs than to GTs. To answer this question, Robinson and Flagel
studied the dissociation of the predictive and motivational
properties of the lever [22]. Part of their results involves asking
whether the Pavlovian lever-CS would serve as a conditioned
reinforcer, capable of reinforcing the learning of a new instrumental
response [29,30]. In a new context, rats were presented
with an active and an inactive nose port. Nose poking into the
active port resulted in presentation of the lever for 2 seconds
without subsequent reward delivery, whereas poking into the
inactive one had no consequence. The authors observed that while
both STs and GTs preferred the active nose port to an inactive
one, STs made significantly more active nose pokes than GTs (see
Figure 6 B, see also [31]). This suggests that the lever acquired
greater motivational value in STs than in GTs.
Without requiring additional simulations, the model can explain
these results by the value that has been incrementally learned and
associated with approaching the lever in the prior autoshaping
procedure for STs and GTs. In the model, STs attribute a higher
value to interacting with the lever than GTs and should actively
work for its appearance enabling further engagement. Figure 6 D
shows the probabilities of engagement that would be computed at
lever appearance after removing the magazine (and related
actions) at the end of the experiment. Indeed, even though the
lever is presented only very briefly, upon its presentation in the
conditioned reinforcement test, STs actively engage and interact
with it [22]. Any value associated to a state-action pair makes this
action in the given state rewarding in itself, favouring actions (e.g.
nosepokes) that would lead to such state. Repeatedly taking this
action without receiving rewards should eventually lead to a
decrease of this value and reduce the original engagement.