Towards Neural Network based Cognitive Models of Dynamic Decision-Making by Humans

TL;DR

The paper addresses modeling dynamic human decision-making by combining cognitive instance-based memory with neural networks to capture heterogeneity across individuals. It introduces two memory-augmented architectures, TL-PMIM and IL-PMIM, that personalize predictions by conditioning on past experiences, and evaluates them against IBL and GPT-3.5 on phishing and cyber-security attacker datasets. Results show neural models, especially TL-PMIM, better reproduce individual decision patterns while maintaining interpretability (via attention in IL-PMIM) and offering insights into human cognition. The work advances cognitive modeling by enabling high-fidelity, data-driven personalization with potential applications in training and decision support.

Abstract

Modeling human cognitive processes in dynamic decision-making tasks has been an endeavor in AI for a long time because such models can help make AI systems more intuitive, personalized, mitigate any human biases, and enhance training in simulation. Some initial work has attempted to utilize neural networks (and large language models) but often assumes one common model for all humans and aims to emulate human behavior in aggregate. However, the behavior of each human is distinct, heterogeneous, and relies on specific past experiences in certain tasks. For instance, consider two individuals responding to a phishing email: one who has previously encountered and identified similar threats may recognize it quickly, while another without such experience might fall for the scam. In this work, we build on Instance Based Learning (IBL) that posits that human decisions are based on similar situations encountered in the past. However, IBL relies on simple fixed form functions to capture the mapping from past situations to current decisions. To that end, we propose two new attention-based neural network models to have open form non-linear functions to model distinct and heterogeneous human decision-making in dynamic settings. We experiment with two distinct datasets gathered from human subject experiment data, one focusing on detection of phishing email by humans and another where humans act as attackers in a cybersecurity setting and decide on an attack option. We conducted extensive experiments with our two neural network models, IBL, and GPT3.5, and demonstrate that the neural network models outperform IBL significantly in representing human decision-making, while providing similar interpretability of human decisions as IBL. Overall, our work yields promising results for further use of neural networks in cognitive modeling of human decision making.

Figures (20)

• Figure 1: Overview of Models for the Phishing Task. (Left) The IBL model and our newly proposed models all process inputs that consist of past memory and the current email. The interpretability of these models increases progressively from TL-PMIM to IL-PMIM and is highest in IBL. The IBL model derives interpretability from an explicit fixed-form formula, whereas IL-PMIM relies on attention weights to determine the contribution of each email. TL-PMIM, which processes each token in the email in a highly non-linear manner, offers the least interpretability. (Right) The memory maintains a set of email instances, each represented by a tuple consisting of state, action, and utility.
• Figure 2: Phishing confusion matrix with 50:10 split
• Figure 3: Phishing confusion matrix with 10:50 split
• Figure 4: IAG confusion matrix
• Figure 5: Model-human alignment over trials in the test set. (Left) Phishing 10:50; (Middle) Phishing 50:10; (Right) IAG