Understanding Human Decision-Making: Individuality Transfer Explained

Humans (and other animals) exhibit substantial commonalities in their decision-making processes. However, considerable variability is also frequently observed in how individuals perform perceptual and cognitive decision-making tasks (Carroll and Maxwell, 1979; Boogert et al., 2018). This variability arises from differences in underlying cognitive mechanisms. For example, individuals may vary in their ability or tendency to retain past experiences (Duncan and Shohamy, 2016; Collins and Frank, 2012), respond to events with both speed and accuracy (Wagenmakers and Brown, 2007; Spoerer et al., 2020), or explore novel actions (Frank et al., 2009). If these factors can be meaningfully disentangled, they would enable a concise characterization of individual decision-making processes, yielding a low-dimensional, parameterized representation of individuality. Such a representation could, in turn, be leveraged to predict future behaviors at an individual level. Shifting from population-level predictions to an individual-based approach would mark a significant advancement in domains where precise behavior prediction is essential, such as social and cognitive sciences. Beyond prediction, this approach offers a framework for parameterizing and clustering individuals, thereby facilitating the visualization of behavioral heterogeneity, which has applications in psychiatric analysis (Pedersen et al., 2017; Dezfouli et al., 2019a). Furthermore, this parameterization offers a promising pathway toward computational modeling at the individual level—replicating the cognitive and functional characteristics of individuals in silico (Shengli, 2021). Cognitive modeling is a standard approach for reproducing and predicting human behavior (Navarro et al., 2006; Busemeyer and Stout, 2002; Yechiam et al., 2005), often implemented within a reinforcement learning framework (e.g. O’Doherty et al., 2007; Daw et al., 2011; Wilson and Collins, 2019). However, because these cognitive models are manually designed by researchers, their ability to accurately fit behavioral data may be limited (Fintz et al., 2022; Song et al., 2021; Miller et al., 2023; Eckstein et al., 2022). A data-driven approach using artificial neural networks (ANNs) offers an alternative (Dezfouli et al., 2019b; Radev et al., 2022; Schaeffer et al., 2020). Unlike cognitive models, which rely on predefined behavioral assumptions (Rmus et al., 2024), ANNs require minimal prior assumptions and can learn complex patterns directly from data. For instance, convolutional neural networks (CNNs) have successfully replicated human choices and reaction times in various visual tasks (Kriegeskorte, 2015; Rajalingham et al., 2018; Fel et al., 2022). Similarly, recurrent neural networks (RNNs; Siegelmann and Sontag, 1995; Cho et al., 2014) have been applied to model value-guided decision-making tasks such as the multi-armed bandit problem (Yang et al., 2019; Dezfouli et al., 2019a). A promising approach to capturing individual decision-making tendencies while preserving behavioral consistency is to tune ANN weights using a parameterized representation of individuality. This idea was first proposed by Dezfouli et al., 2019a, who employed an RNN to solve a two-armed bandit task. Their study utilized an autoencoder framework (Rumelhart and McClelland, 1987; Tolstikhin et al., 2017), in which behavioral recordings from a single session of the bandit task, performed by an individual, were fed into an encoder. The encoder produced a low-dimensional vector, interpreted as a latent representation of the individual. Similar to hypernetworks (Ha et al., 2016; Karaletsos et al., 2018), a decoder then took this low-dimensional vector as input and generated the weights of the RNN. This framework successfully reproduced behavioral recordings from other sessions of the same bandit task while preserving individual characteristics. However, since this individuality transfer has only been validated within the bandit task, it remains unclear whether the extracted latent representation captures an individual’s intrinsic tendencies across a variety of task conditions. To address this question, we aim to make the low-dimensional representation—referred to as the individual latent representation—robust to variations across individuals and task conditions, thereby enhancing its generalizability. Specifically, we propose a framework that predicts an individual’s behaviors, not only in the same condition but also in similar yet distinct task conditions and environments. If the individual latent representation serves as a low-dimensional representation of an individual’s decision-making process, then extracting it from one condition could facilitate the prediction of that individual’s behaviors in another. In this study, we define the problem of individuality transfer across task conditions as follows (also illustrated in Figure 1). We assume access to a behavioral dataset from multiple individuals performing two task conditions: a source task condition and a target task condition. We train an encoder that takes behavioral data from the source task condition as input and outputs an individual latent representation. This representation is then fed into a decoder, which generates the weights of an ANN, referred to as a task solver, that reproduces behaviors in the target task condition. For testing, a new individual provides behavioral data from the source task condition, allowing us to infer his/her individual latent representation. Using this representation, a task solver is constructed to predict how the test individual will behave in the target task condition. Importantly, this prediction does not require any behavioral data from the test individual performing the target task condition. We refer to this framework as EIDT, an acronym for encoder, individual latent representation, decoder, and task solver. We evaluated whether the proposed EIDT framework can effectively transfer individuality in both value-guided sequential decision-making tasks and perceptual decision-making tasks. To assess its generalizability across individuals, meaning its ability to predict the behavior of previously unseen individuals, we tested the framework using a test participant pool that was not included in the dataset used for model training. To determine how well our framework captures each individual’s unique behavioral patterns, we compared the prediction performance of a task solver specifically designed for a given individual with the performance of task solvers designed for other individuals. Our results indicate that the proposed framework successfully mimics decision-making while accounting for individual differences.