The added value of affective processes for models of human cognition and learning

Abstract

Building on the affectivism approach, we expand on Binz et al.'s meta-learning research program by highlighting that emotion and other affective phenomena should be key to the modeling of human learning. We illustrate the added value of affective processes for models of learning across multiple domains with a focus on reinforcement learning, knowledge acquisition, and social learning.

Binz et al. bid to establish an ambitious research program to model human cognition using a meta-learning framework. They effectively illustrate the potential and advantages of meta-learned models, showcasing the ability of such models to acquire inductive biases through experience. Notably, the authors outline a compelling blueprint for how these models could foster the development of a domain-general model of human learning. Here, we seek to complement this blueprint by highlighting a key element that is left mostly unexplored in the target article: Affective processes.

Affective processes – typically emotions, feelings, motivations, moods, or attitudes – are not only inherently linked to well-being, but also drive behavioral and cognitive processes such as attention, learning, memory, and decision-making (e.g., LaBar & Cabeza, 2006; Lerner, Li, Valdesolo, & Kassam, 2015; Phelps, 2006; Pool, Brosch, Delplanque, & Sander, 2016). This grants affective processes high explanatory power in understanding human behavior and cognition, a central argument of the affectivism approach (Dukes et al., 2021). As such, we suggest that considering affective processes is pivotal to the modeling of human cognition, and especially of learning. Affective processes are indeed central to – and exert a pervasive influence on – how humans learn (e.g., Öhman & Mineka, 2001; Vollberg & Sander, 2024; Wuensch, Pool, & Sander, 2021). Below, we illustrate how emotion and other affective phenomena are central to human learning across various domains, with a particular focus on reinforcement learning, knowledge acquisition, and social learning.

Affective processes emerge as important factors in reinforcement learning. This fundamental learning process enables individuals to attribute value to states or stimuli and actions via teaching signals such as rewards and punishments. These reinforcers and their associated stimuli typically evoke affective responses, which are core components of reward-seeking and threat-related behaviors (Levy & Schiller, 2021; Stussi & Pool, 2022). Affective processes also modulate how individuals learn from reinforcers. Studies on Pavlovian conditioning – a basic form of reinforcement learning – have shown that stimuli with heightened affective relevance, such as both threat-relevant (e.g., angry faces) and positive emotional (e.g., baby faces) stimuli, are more rapidly and persistently associated with an aversive outcome than neutral stimuli (Stussi, Pourtois, & Sander, 2018; Stussi, Pourtois, Olsson, & Sander, 2021). At the computational level, these studies indicate that affective relevance modulates how individuals learn from prediction errors: Affectively relevant stimuli were associated with a lower learning rate for negative prediction errors (i.e., when the aversive outcome was expected but omitted), enhancing the persistence of their association with the aversive outcome (Stussi et al., 2018, 2021). Similarly, substantial evidence has demonstrated that individuals learn differently about positive and negative outcomes in the instrumental domain (Dorfman, Bhui, Hughes, & Gershman, 2019; Lefebvre, Lebreton, Meyniel, Bourgeois-Gironde, & Palminteri, 2017). Positively valenced prediction errors are generally associated with a higher learning rate than negatively valenced prediction errors, providing a computational correlate of such learning asymmetry (Palminteri & Lebreton, 2022). Altogether, these findings highlight that affective mechanisms shape basic reinforcement learning processes.

Affective processes likewise support epistemic learning. Both positive and negative emotions have long been studied for their roles in the encoding, consolidation, and recall of episodic memories (Levine & Pizarro, 2004), as well as in academic learning in educational settings (see Pekrun & Linnenbrink-Garcia, 2014). Epistemic emotions are the key family of emotions supporting knowledge exploration and acquisition (see Muis, Chevrier, & Singh, 2018). Emotions such as interest, curiosity, confusion, surprise, wonder, or awe are drivers of learning (e.g., Chevrier, Muis, Trevors, Pekrun, & Sinatra, 2019; Vogl, Pekrun, Murayama, & Loderer, 2020). As an illustration of the central role of epistemic emotions in learning, the “trivia questions” paradigm is typically used to understand how epistemic curiosity enhances memory. Using this paradigm, research has shown that the more participants are curious to know the response to a question (e.g., “who is the most cited psychologist of the 21st century?”), the more they later remember the response (e.g., Kang et al., 2009; Marvin & Shohamy, 2016). The impact of curiosity on knowledge exploration and acquisition, partly relying on reward-related processes, is therefore a salient example of how the intrinsic value of information can enhance learning (Murayama, 2022).

While many things can be learned by exploring one's own environment, this individual approach has its limitation. Some information simply cannot be gleaned in this way and requires input from other (human) sources (Harris & Koenig, 2006). Critically, such social learning fundamentally relies on affective processes. Social learning has historically been seen as either a non-human primate phenomenon that explains how behavior can be learned from conspecifics (Zentall & Galef, 1988), or a human cognitive developmental phenomenon concerned with learning from others’ testimony (Harris, 2012). However, affective social learning not only points out that these two branches of social learning originate from the same tree (Gruber, Bazhydai, Sievers, Clément, & Dukes, 2022), but also that it is possible to learn from others’ affective attitudes about the value of objects (e.g., ideas, people, customs). An important part of who we are – our values, ethics, and morality – is based on our perception, attention, and memory of interaction with and learning from others, whether or not this information is communicated ostensively (Dukes & Clément, 2017; Egyed, Király, & Gergely, 2013). And indeed, what we perceive, attend to, and remember is largely defined by how important, valuable, and affective those objects of perception, attention, and memory are. Not only do we remember what is affectively relevant to us as individuals, but others, serving as proxy relevant detectors, can also signal what is more or less relevant, to be learned or forgotten (Dukes & Clément, 2019; see also Sorce, Emde, Campos, & Klinnert, 1985).

In conclusion, affective processes play a fundamental role in learning across various domains and their consideration is key to the modeling of human learning. Given that emotions are not immutable and static but flexibly arise from the interaction between an individual and their environment (Scherer & Moors, 2019), it could be particularly enlightening to conceptualize emotion as a kind of inductive bias attuned by experience within the meta-learning framework proposed by Binz et al. Such conceptualization could offer a promising way of modeling the effects of emotion on learning, thereby providing added value to the meta-learning research agenda.