Abstract
Predicting guests' food preferences while simultaneously reducing food waste has become an increasingly pressing challenge in the hospitality industry, with significant implications for operational efficiency, sustainability, and profitability. This study explores a novel approach to optimising food purchasing and minimising waste by integrating cognitive biases into AI-driven decision-making systems.
Traditionally, AI systems are trained on data generated by human decision-making, which inherently reflects human cognitive biases. Machine learning bias refers to the distortions that arise when the training dataset is influenced by these biases, leading to imbalanced or skewed outcomes. Such biases can result in reduced accuracy, discriminatory effects, and, in some cases, harmful or ineffective outcomes. However, recent literature presents a growing argument for strategically instilling cognitive biases into machine learning architectures (Battaglia et al., 2018; Goyal & Bengio, 2021). The relationship between cognitive biases in humans and those embedded in AI systems is complex and intertwined. For instance, Hagendorff and Fabi (2024) suggest replicating human cognitive biases as inductive biases within machine learning, arguing that these biases can bring valuable benefits to algorithmic decision-making. This approach contrasts with the traditional pursuit of fairness in AI, advocating instead for the transparent and intentional embedding of ethical principles through biases in the design, training, and deployment of AI systems. When purposefully designed and carefully controlled, cognitive biases can enable AI systems to compensate for the limitations of imperfect real- world data and decision-making contexts.
The study proposes the development of a hybrid AI system that combines machine learning with expert knowledge (e.g., from chefs) to inform food purchasing decisions. Fashionable statistical optimisation methods often struggle with issues like scarce, noisy, or highly seasonal data, but the deliberate introduction of a bias such as “conservatism” could act as a safeguard, helping the system to reduce overproduction and, consequently, minimise food waste. Hybrid AI systems combine the strengths of machine learning and symbolic AI, drawing on both the statistical power of neural networks and the reasoning capabilities of expert systems. These systems are particularly effective in environments characterised by imprecise information and uncertainty. Machine learning, as a core domain of AI, focuses on developing algorithms that can identify patterns in data and generalise to new, unseen instances. Expert systems, on the other hand, rely on structured repositories of domain-specific knowledge (knowledge bases) and employ inference engines to simulate human reasoning. In hybrid systems, supervised learning techniques optimise rule selection and adapt parameters based on expert input, enhancing the system’s ability to make informed decisions.
In contexts like tourism and hospitality, training datasets may be limited due to factors such as the small scale of businesses, seasonality, and time constraints on data collection. In these situations, both managers and customers often face time-sensitive decisions under uncertainty, frequently influenced by emotions, personal experiences, and situational pressures. Cognitive biases, which are evolutionary and adaptive, may serve as a useful design feature for facilitating natural adaptation (Haselton, Nettle, & Andrews, 2005). Rather than opposing the heuristic approach, integrating cognitive biases with machine learning can offer an effective shortcut for more efficient learning and decision-making. By exploring how cognitive biases can be purposefully integrated and controlled within AI, this research contributes to the ongoing development of next-generation AI applications in the tourism and hospitality sectors.