Among the seven key requirements to achieve trustworthy AI proposed by the High-Level Expert Group on Artificial Intelligence (AI-HLEG) established by the European Commission, the fith requirement (“Diversity, non-discrimination and fairness”) declares: “In order to achieve Trustworthy AI, we must enable inclusion and diversity throughout the entire AI system's life cycle. [...] This requirement is closely linked with the principle of fairness”. In this paper, we try to shed light on how closely these two distinct concepts, diversity and fairness, may be treated by focusing on information access systems and ranking literature.
Diversity and fairness concepts should not be used interchangeably because they do represent
two diferent values, but they also cannot be considered totally unrelated or divergent. Having
diversity does not imply fairness, but fostering diversity can efectively lead to fair outcomes,
an intuition behind several methods proposed to mitigate the disparate impact of information
access systems, i.e. recommender systems and search engines [
nEvelop-O fairness relates to the concept of coverage-based diversity, an aggregated diversity metric often used in Recommender Systems literature. Indeed, such metric is maximised when diferent groups of items are represented in the most heterogeneous way.
Second, both fairness and diversity relate to the treatment of, and consequently the impact on, protected/disadvantaged/minority groups (or classes). The definition of protected class is usually dependent upon laws and policies which may vary between countries, aiming at preventing any form of discrimination towards such classes. For instance, the EU Charter of Fundamental Rights states that: “Any discrimination based on any ground such as sex, race, colour, ethnic or social origin, genetic features, language, religion or belief, political or any other opinion, membership of a national minority, property, birth, disability, age or sexual orientation shall be prohibited” [Article 21, EC, 2012].
As argued by Castillo [
So far we have discussed some intersections between diversity and fairness concepts, but in order to better clarify their nature it is useful to focus also on the diferences between them. Early quantitative definitions of both values have been proposed several decades ago, but in their rationale we note a substantial diference. Indeed, whilst since the beginning fairness metrics have been proposed to tackle societal issues [14], most of the diversity indexes still widely used have been proposed in disparate fields, e.g., Simpson’s Index in Ecology [ 15], and they have been originally formulated to measure diversity intended as heterogeneity, variety or entropy, e.g., Shannon’s Index [16]. Even if this does not undermine their use in measuring diversity, it is also true that their application needs to be contextualised for supporting the validity of the inferred results. Similarly, a lack of a value-oriented approach can be found in the design of the diversification techniques [ 17, 18]. Indeed, looking at the early proposals of the Information Retrieval and Recommender Systems communities, the main goal for diversifying is to tackle the problem of ambiguity of a query or the redundancy of the results, and also to deal with uncertainty. Great advancements have been made in this direction [19], but this utility-oriented definition has partly created ambiguity over the concept of diversity itself.
Whilst the aforementioned points are just a few among the several aspects that link diversity and fairness, we conclude by stressing their relevance in recent policies proposed in the European context. The Digital Service Act (DSA) [20] mandates that digital services powered by technologies such as recommender systems and search engines should be monitored to guarantee the avoidance of unfair or arbitrary outcomes. Under a diferent lens, the Artificial Intelligence Act (AI Act) proposal [21] also refers to the need for bias monitoring as part of the mandatory requirements for high-risk AI systems. Moreover, in terms of diversity the AI Act explicitly states that providers of AI systems should be encouraged to create code of conduct covering aspects such as accessibility, stakeholders participation and ensuring diversity of development teams. These two goals considered above, i.e. system-centric (ensuring bias and fairness in algorithmic systems) and a people-centric view (ensuring diversity of persons involved in the AI design process), are strongly related. Only fostering the diversity of development teams, and therefore embedding diferent perspectives, could lead to a future where Information Access Systems act in a trustworthy and fair way.
This work has been adapted from Lorenzo Porcaro’s PhD dissertation [22]. It is partially supported by the HUMAINT programme (Human Behaviour and Machine Intelligence), Joint Research Centre, European Commission. The project leading to these results received funding “la Caixa” Foundation (ID 100010434), under agreement LCF/PR/PR16/51110009, and from EUfunded projects “SoBigData++” (grant agreement 871042) and “FINDHR” (grant agreement 101070212). [7] M. Zehlike, K. Yang, J. Stoyanovich, Fairness in ranking, part I: Score-based ranking, ACM
Computing Surveys (2022). doi:10.1145/3533379. [8] G. K. Patro, L. Porcaro, L. Mitchell, Q. Zhang, M. Zehlike, N. Garg, Fair ranking: A critical review, challenges, and future directions, in: 2022 ACM Conference on Fairness, Accountability, and Transparency, 2022, p. 1929–1942. doi:10.1145/3531146.3533238. [9] M. Drosou, H. Jagadish, E. Pitoura, J. Stoyanovich, Diversity in big data: A review, Big
Data 5 (2017) 73–84. doi:10.1089/big.2016.0054. [10] M. Mitchell, D. Baker, N. Moorosi, E. Denton, B. Hutchinson, A. Hanna, T. Gebru, J. Morgenstern, Diversity and inclusion metrics in subset selection, in: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, AIES ’20, 2020, pp. 117–123. doi:10.1145/3375627.3375832. [11] D. Steel, S. Fazelpour, K. Gillette, B. Crewe, M. Burgess, Multiple diversity concepts and their ethical-epistemic implications, European Journal for Philosophy of Science 8 (2018) 761–780. doi:10.1007/s13194-018-0209-5. [12] EC, European Commission, The charter of fundamental rights, Oficial Journal of the
European Communities (2012). URL: https://eur-lex.europa.eu/eli/treaty/char_2012/oj. [13] N. Helberger, K. Karppinen, L. D’Acunto, Exposure diversity as a design principle for recommender systems, Information, Communication and Society 21 (2018) 191–207. doi:10.1080/1369118X.2016.1271900. [14] B. Hutchinson, M. Mitchell, 50 years of test (un)fairness: Lessons for machine learning, in: Proceedings of the Conference on Fairness, Accountability, and Transparency, FAT* ’19, 2019, pp. 49–58. doi:10.1145/3287560.3287600. [15] J. Simpson, Measurements of diversity, Nature 163 (1949) 688. [16] C. Shannon, A Mathematical Theory of Communication, The Bell System Technical
Journal Vol.27 (1948) 379–423. [17] J. Carbonell, J. Goldstein, Use of MMR, diversity-based reranking for reordering documents and producing summaries, 1998, pp. 335–336. doi:10.1145/3130348.3130369. [18] B. Smyth, P. McClave, Similarity vs. Diversity, in: Proceedings of the International Conference on Case-Based Reasoning (ICCBR) 2001, 2001, pp. 347–361. doi:10.1007/ 3-540-44593-5_25. [19] P. Castells, N. J. Hurley, S. Vargas, Novelty and diversity in recommender systems, in: F. Ricci, L. Rokach, B. Shapira (Eds.), Recommender Systems Handbook, third ed., Springer, 2022, pp. 603–647. [20] EC, European Commission, Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market For Digital Services and amending Directive 2000/31/EC (Digital Services Act), 2022. [21] EC, European Commission, Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act).COM(2021) 206, 2021. [22] L. Porcaro, Assessing the impact of music recommendation diversity on listeners, Ph.D. thesis, Universitat Pompeu Fabra, 2022.