Learning about people’s perception toward food in India and devising Bias-free recommendation system

Abstract

One concept of fairness commonly discussed in ethical literature suggests that all individ uals, regardless of their background, should have equal access to the products and services they request. The realm of food, being crucial for human survival and integral to any econ omy, has witnessed the application of various Artificial Intelligence (AI) techniques in areas such as food recommendations, recipe generation, and wellness-oriented approaches. It is important to examine whether fairness concerns are relevant in this context, given that food choices can significantly impact an individual’s health, personal finances, and other aspects of life. Therefore, any AI system introduced in this domain should be evaluated responsibly, not only in terms of performance but also with respect to non-functional requirements, in cluding their impact on users. To begin our journey towards this objective, this work aims to examine how individuals perceive food in relation to their backgrounds. We intend to utilize this understanding as a framework for evaluating future AI systems centered around food. To accomplish this, we have conducted a unique survey among participants at educa tional institutions in India, encompassing individuals associated with various regions both within and outside the country. Our findings indicate that there is an existing imbalance in access to food from different regions, which may be considered unfair according to prevail ing definitions of group fairness. Additionally, we discovered that many individuals exhibit preferences for food that transcend their own regions, and there is an increasing reliance on digital tools for food-related matters. These results suggest that AI developers have the po tential to foster broader and fairer access to food by employing a combination of strategies. This entails embracing food choices that transcend regions, leveraging available food data (exploiting), as well as exploring and introducing new choices that are specific to individual regions (exploring), thus accommodating users’ regional associations. Also, our objective is to address the problem of bias in the Top-k Ranking recommenda tion system. The goal is to select a subset of k candidates from a large pool of n candidates while maximizing utility and ensuring adherence to group fairness criteria.