Behavioral flexibility is manipulatable and it improves flexibility and problem solving in a new context: post-hoc analyses of the components of behavioral flexibility

Abstract

Behavioral flexibility, adapting behavior to changing situations, is hypothesized to be related to adapting to new environments and geographic range expansions. However, flexibility is rarely directly tested in a way that allows insight into how flexibility works. Research on great-tailed grackles, a bird species that has rapidly expanded their range into North America over the past 140 years, shows that grackle flexibility is manipulatable using colored tube reversal learning and that flexibility is generalizable across contexts multi-access box). Here, we use these grackle results to conduct a set of posthoc analyses using a model that breaks down performance on the reversal learning task into different components. We show that the rate of learning to be attracted to an option (phi) is a stronger predictor of reversal performance than the rate of deviating from learned attractions that were rewarded (lambda). This result was supported in simulations and in the data from the grackles: learning rates in the manipulated grackles doubled by the end of the manipulation compared to control grackles, while the rate of deviation slightly decreased. Grackles with intermediate rates of deviation in their last reversal, independently of whether they had gone through the serial reversal manipulation, solved fewer loci on the plastic and wooden multi-access boxes, and those with intermediate learning rates in their last reversal were faster to attempt a new locus on both multi-access boxes. These findings provide additional insights into how grackles changed their behavior when conditions changed. Their ability to rapidly change their learned associations validates that the manipulation had an effect on the cognitive ability we think of as flexibility.

Risk-sensitive learning is a winning strategy for leading an urban invasion

“…their speedier reversal learning performance via pronounced risk-sensitive learning (low 𝜙 and high Logan et al, 2022b;Lukas et al, 2023;Logan et al, 2023b,c). Heeding previous calls (Dukas, 1998;McNamara and Houston, 2009;Fawcett et al, 2013), our study provides an analytical solution to facilitate productive research on proximate and ultimate explanations of seemingly flexible (or not) behaviour: because we publicly provide step-by-step code to examine individual decision making, two core underlying learning mechanisms, and their theoretical selection and benefit (see https://github.com/alexisbreen/Sex-differences-in-grackles-learning), which can be tailored to specific research questions.…”

“…Of these approaches, reversal learning and serial—multiple back-to-back—reversal learning tasks are the most common experimental assays of behavioural flexibility (non-exhaustive examples of each assay in bees; Strang and Sherry 2014; Raine and Chittka 2012; birds; Bond et al 2007; Morand-Ferron et al 2022; fish; Lucon-Xiccato and Bisazza 2014; Bensky and Bell 2020; frogs; Liu et al 2016; Burmeister 2022; reptiles; Batabyal and Thaker 2019; Gaalema 2011; primates; Cantwell et al 2022; Lacreuse et al 2018; and rodents; Rochais et al 2021; Boulougouris et al 2007). We have shown, however, at least for our grackles, faster reversal learning is governed primarily by risk-sensitive learning, so: firstly, these go-to reversal learning assays do not necessarily measure the unit they claim to measure (a point similarly highlighted in: Aljadeff and Lotem, 2021; Fed-erspiel et al, 2017); and secondly, formal models based on the false premise that variation in learning speed relates to variation in behavioural flexibility require reassessment (Lea et al, 2020; Blaisdell et al, 2021; Logan et al, 2022b; Lukas et al, 2023; Logan et al, 2023b,c). Heeding previous calls (Dukas, 1998; McNamara and Houston, 2009; Fawcett et al, 2013), our study provides an analytical solution to facilitate productive research on proximate and ultimate explanations of seemingly flexible (or not) behaviour: because we publicly provide step-by-step code to examine individual decision making, two core underlying learning mechanisms, and their theoretical selection and benefit (see https://github.com/alexisbreen/Sex-differences-in-grackles-learning), which can be tailored to specific research questions.…”

Implementing a rapid geographic range expansion - the role of behavior changes

“…This change also appears to better reflect the performance of the Santa Barbara grackles, because they had higher 𝜙 values, which, in turn, meant lower 𝜆 values to reflect the performance during their initial learning. These lower 𝜆 values better reflected the birds' behavior during the first reversal trials: a large 𝜆 value means that birds continue to choose the now unrewarded option almost 100% of the time, whereas the lower 𝜆 values mean that birds start to explore the rewarded option relatively soon after the switch of the rewarded option (Lukas et al, 2022).…”

“…With the previous initial attraction scores set to zero, a bird would be expected to choose the rewarded option in 100% of the trials after the first time it chose that option (attraction cannot be lower than zero, and choice is shaped by the ratio of the two attractions so that when one option is zero and the other is larger than zero, the ratio will be 100% for the rewarded option). 2) We changed the updating so that an individual only changes the attraction toward the option they chose in that trial [either decreasing their attraction toward the unrewarded option or increasing their attraction toward the rewarded option; see Lukas et al (2022) for more detail]. Previously, both attractions were updated after every trial, assuming that individuals understand that the experiment is set up such that one option is always rewarded.…”

“…For our birds, we instead assumed that individuals will focus on their direct experience rather than making abstract assumptions about the test. Our modification resulted in needing a higher 𝜙 to have the same learning rate as a model where both attraction scores update after every trial (Lukas et al, 2022). This change also appears to better reflect the performance of the Santa Barbara grackles, because they had higher 𝜙 values, which, in turn, meant lower 𝜆 values to reflect the performance during their initial learning.…”