Two kinds of attention in Pavlovian conditioning: Evidence for a hybrid model of learning.

Abstract

Four appetitive Pavlovian conditioning experiments with rats examined the rate at which the discrimination between compounds AY and AX was solved relative to the discrimination between compounds AY and BY. In Experiments 1 and 2, these discriminations were preceded by training in which A and B were continuously reinforced and X and Y were partially reinforced. Consistent with the Pearce and Hall (1980) model, the results showed that the AY/AX discrimination was solved more readily than the AY/BY discrimination. In Experiments 3 and 4, the discriminations were preceded by feature-positive training in which trials with AX and BY signaled food but trials with X and Y did not. Consistent with the Mackintosh (1975) model, the results showed that the AY/BY discrimination was solved more readily than the AY/AX discrimination. These results are discussed with respect to a hybrid model of conditioning and attention.Keywords: partial reinforcement, continuous reinforcement, feature positive, predictive validity, attention, associabilityCentral to many theories of learning is the assumption that variation in stimulus processing determines the degree to which an association between a conditioned stimulus (CS) and an unconditioned stimulus (US) will be modified. Perhaps most influentially, Rescorla and Wagner (1972) proposed that associative learning is determined by the extent to which the CS and the US are processed. According to Rescorla and Wagner, US processing is variable and is a function of the discrepancy between the magnitude of the US and the total associative strength of all stimuli present ( Ϫ ⌺V). Effects such as blocking (Kamin, 1968), conditioned inhibition (Pavlov, 1927), and the influence of degrading the contingency between the CS and US (Rescorla, 1968) were all cited as phenomena that supported the notion that learning is influenced by variations in US processing. In contrast, processing of the CS (␣) was assumed to be fixed and determined by the physical properties of the CS. Although sympathetic to the idea that CS processing, or attention, 1 may vary as a consequence of experience , no attempt was made to formalize this mechanism.Since the publication of the Rescorla-Wagner (1972) model, however, there has been considerable interest in the idea that CS processing, like US processing, varies as a consequence of learning, and efforts have been made to specify the conditions under which this variation takes place. Two theories, in particular, have come to dominate the study of CS processing. The first, by Mackintosh (1975), proposed that a CS will be processed more when it is a good predictor of a US than when it is a poor predictor of a US (see also Kruschke, 2001;Lovejoy, 1968;Sutherland & Mackintosh, 1971;Zeaman & House, 1963). The second, by Pearce and Hall (1980; see also Pearce, Kaye, & Hall, 1982;Schmajuk, Lam, & Gray, 1996), proposed the opposite: that a CS will be processed more when it is followed by unpredictable events (such as the US) than when it is followed by predictable events. As we sha...

Prediction and Uncertainty in Associative Learning: Examining Controlled and Automatic Components of Learned Attentional Biases

“…Our 'pure exploitation' predator would now be at a disadvantage, since it would remain hunting for a scarce resource when an easier alternative is available. In contrast, a predator that occasionally changes its foraging strategy and explores other potential prey items (about which it is currently uncertain) would rapidly learn to increase its reliance on red spiders, and may thus be more likely to thrive.Animal experimentation has shown results consistent with the operation of uncertainty-driven attention (Haselgrove, Esber, Pearce, & Jones, 2010;Kaye & Pearce, 1984a;Swan & Pearce, 1988;Wilson, Boumphrey, & Pearce, 1992), and the neural basis of this mechanism has been delimited to critical involvement of the prefrontal and amygdala regions (Fiorillo, Tobler, & Schultz, 2003;Roesch, Esber, Li, Daw, & Schoenbaum, 2012). (2015) showed that overt attention to stimuli (measured via eye-tracking) could be determined by both predictiveness and uncertainty within the same task.…”

Overshadowing and associability change.

“…However, associative strength itself cannot be the sole determinant of the associability of CSs. As will be recalled from our description of the experiments conducted by Haselgrove et al (2010), which used the same stimuli, apparatus and species as in the current experiments, following the solution of an AXϩ, BYϩ, XϪ, YϪ discrimination, the associability of A and B (that commanded strong conditioned responding and which, presumably, had high associative strength) was greater than X and Y (that commanded weaker conditioned responding and that, presumably, had lower associative strength). A related result was also reported in autoshaping experiments with pigeons conducted by .…”

“…Although this is a logical possibility, we have strived to use a variety of techniques that, in theory, should be sensitive to variations in the associability of a CS, while at the same time reduced the impact of alternative mechanisms on our results. Furthermore, we, and others have used these techniques in the past to provide evidence for Mackintosh's theory (e.g., Duffaud et al, 2007;Haselgrove et al, 2010). If an advocate of Mackintosh's theory were to object to the current techniques for assessing associability, then the same advocate would, presumably, also have to object to these past experiments.…”

“…Interestingly, the analysis we have just provided generates a way of testing the Pearce-Mackintosh analysis. If A, B, X, and Y entered the experiment with relatively low associability values, rather than high values as we and others have suggested (e.g., Haselgrove et al, 2010;Le Pelley, 2004) then during ABϩ, Xϩ, Yϩ training, ␣ for X and Y will climb to its maximum value, while ␣ for A and B will remain at its low starting value. The consequences of this for the model may be substantial, for now, the total associability of A and B could end up being either equivalent to, or lower than X and Y.…”

“…This omission presumably stems from the difficulty in assessing the associability of CSs that have very different levels of associative strength. However, in a recent series of conditioning experiments conducted with pigeons and rats, we have circumvented this problem and have demonstrated a number of techniques that permit the evaluation of the associability of a CS (Haselgrove, Esber, Pearce, & Jones, 2010;Pearce, Esber, George, & Haselgrove, 2008). For example, Haselgrove et al gave rats two feature-positive discriminations of the form: AXϩ, BYϩ, XϪ, YϪ.…”