| Value | Interpretation in the context of recommenders | Example Indicator | Example design |
---|
1 | Usefulness | “The purpose of recommenders is often summarized as ‘help the users find relevant items’” [167] A recommender should provide a useful service to the user. | Long term user retention. User satisfaction surveys. | Show people more of what they rate highly. |
2 | Liberty | A platform should respect human dignity and protect human rights and freedoms [116:61]; recommenders should operationalize respect for autonomy [352]. A platform should not stop users from exploring certain types of information. Users should be able to pursue their own good in their own way. | Similar mechanisms as Freedom of Expression. | Protections against arbitrary removal of content. Eliminating threats of violence designed to compel individuals to form particular opinions, in violation of article 19 in Universal Declaration of Human Rights. |
3 | Freedom of Expression | “Everyone shall have the right to freedom of expression; this right shall include freedom to seek, receive and impart information and ideas of all kinds” [345 art .19] “The exercise of freedom of opinion, expression and information … is a vital factor in the strengthening of peace and international understanding” [342]. Platforms should not stop users from expressing their thoughts and opinions, or unduly suppress distribution of user posts. | “Conducting an ex-ante evaluation attempts to predict the future relationship between [human] rights and an on-going or proposed business activity.” [197] | Transparency around content which is suppressed. Appeals processes for removed content. |
4 | Control | Control “includes the system's ability to allow users to revise their preferences, to customize received recommendations, and to request a new set of recommendations” [267:18] “Ability to determine the nature, sequence and/or consequences of technical and operational settings, behavior, specific events, and/or experiences.” [162:17] The platform should give users ways to control the content selected for them and it should give users ways to control how the content they create is shared. | “I feel in control of my news feed.” [348] Locus of control Scale. Ex: “Other people usually control my life” [289] | Recommender controls [143, 147, 292] |
5 | Agency, Autonomy, Efficacy | “The sense of agency can be analyzed as a compound of more basic experiences, including the experience of intentional causation, the sense of initiation and the sense of control” [257] The platform should provide the capacity for intentional action and help users achieve their goals [352] The platform should not manipulate users for the benefit of other stakeholders. | General Self-Efficacy Scale (GSE) Ex: “It is easy for me to stick to my aims and accomplish my goals” [63] Autonomy Scale (AS): Measures components of autonomy including family loyalty autonomy, value autonomy, emotional autonomy, and behavioral autonomy [10] | Recommender controls [143, 147, 292] |
6 | Privacy | “The protection of personal data is of fundamental importance to a person's enjoyment of his or her right to respect for private and family life” [73] Users should be allowed to determine if and how their personal data is collected, processed and disseminated [116:21, 162:71] | GDPR Data Protection Impact Assessment [103] | Controls to limit data use. Federated recommendation algorithms [268, 373]. |
7 | Safety, Security | The principle of ‘safety’ requires that an AI system be reliable and that ‘the system will do what it is supposed to do without harming living beings or [its] environment.’” [116:38] “[System] use should not contribute to increasing stress, anxiety, or a sense of being harassed by one's digital environment.” [347] | Sense that most people can be trusted [372] | Hate speech moderation Parental controls Option to report harmful content. |
8 | Self-expression, authenticity | “Self-expression is a notion that is closely associated with a horde of positive concepts, such as freedom, creativity, style, courage, self-assurance, and even healing and spirituality” [186] The platform should empower users to express their identity (including personality, attributes, behavior) and to decide how it is presented to others. [183:19]. | Correlates with (but is not identical to) quantity and quality of user content creation | Easy-to-use, attractive content creation/modification tools, e.g., SnapChat filters, TikTok tools. Content moderation policies that allow self-expression (e.g., don't take down art because it has nudity) |
9 | Well-being | Well-being is a complex and multidimensional construct that encompasses many of the values we describe here and there is no universally accepted definition [88]. Well-being may be measured by both objective and subjective factors including satisfaction with life, emotions and affect, psychological well-being, social relationships, and meaning [205] On the platforms, users should see content that leads them to experience contentment, joy and pleasure, both ephemerally and over the long term. Platforms should help users feel satisfied with their lives. | Satisfaction with Life Scale (SWLS). Ex: “The conditions of my life are excellent” [87] Positive Affect and Negative Affect Scales (PANAS). Ex: “Indicate the extent to which you have felt interested, distressed, excited, upset, etc over the past week” [363] Scale of Psychological Wellbeing (SPWB). Assesses psychological functioning across domains such as autonomy, self-acceptance, purpose in life, and positive relationships [285] | Encourage active vs. passive use of social media [354]
Usage controls, e.g., screen time limits |
10 | Inspiration, Awe | “In times of uncertainty, others are sought for guidance, inspiration or motivation, to seek ideas, goals or possibilities, which can influence ambitions, choices, and achievements.” [183:11] Platforms should show users content that will inspire, motivate, or guide them. | Inspiration scale: Measures the experience of being inspired, feeling inspired and motivation to do something with that inspiration [336] | Uprank content identified by Inspiration classifier [163] |
11 | Mental Health | Platforms should help users protect and improve their mental health. Platforms should not encourage unhealthy types or amounts of use. “[A] state of well-being in which the individual realizes his or her own abilities, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to his or her community” [371] | Warwick-Edinburgh Mental Well-Being Scale (WEMWBS) [331] Positive Mental Health Scale (PMH) Ex: “I feel that I am actually well equipped to deal with life and it's difficulties” [211] | Features that help users manage their screen time, e.g., “take a break” notifications, screen time limits. Option to report harassment or bullying. |
12 | Physical Health | Platforms should help users protect and improve their physical health, such as by providing accurate health information and encouraging behaviors that contribute to physical health. | WHO Wellbeing Index (WHO-5) Ex: “I have felt active and vigorous” and “I woke up feeling fresh and rested” [338] | Help users develop healthy habits (exercise, diet, etc.) Clear platform policies on removing health misinformation, ads for fake cures, etc. |
13 | Self-actualization, Personal Growth | Platforms should enable users to reach their full potential. “Personal growth is a continuous improvement in life in order to find purpose and meaning” [183:22] “[R]ather than have people choose the easiest option, we wish to have them develop a strong sense of determination of having selected the right path” [187] | “Feeling that the things one does are worthwhile” [340] European Social Survey (ESS). Ex: “To what extent [do you] learn new things in life?” [247] | Educational recommender systems [81] |
14 | Recognition, Acknowledgment | Platforms should provide ways for other people to recognize a user's contributions or worth. “Whilst some esteem needs can be met by having self-acceptance, self-worth and self-value, validation from others is also important” [183:28] | Social Inclusion Scale (SIS). Ex: ‘I have felt accepted by my friends’’ [370] | “Celebrate” reaction button (as LinkedIn has) |
15 | Knowledge, Informativeness | Users should see items that keep them informed about topics they care about or might care about. “[P]ersonalised news recommendations allow the media not only to help users find relevant information, but also to inform them better and more effectively.” [148:995] “Curiosity drives individuals to seek stimulation, information or new experiences, serving a purpose to increase knowledge and build skills.” [183:18] | News knowledge quizzes [6, 12] Curiosity and exploration inventory (CEI). Ex: “I would describe myself as someone who actively seeks as much information as I can in a new situation’’ [180] | News recommender systems [179] Trending topics |
16 | Connection | “Individuals are driven to interact and seek social closeness … The benefits of social connections are far-reaching.” [183:17] | Perceived Social Support scale (PSS). Ex: “I have friends with whom I can share my joys and sorrows” and “I can talk about my problems with my friends” [391] Online Social Support Scale (OSSS). Ex: “Online, I belong to groups of people with similar interests” and “Online, people make me feel like I belong” [245] | Recommendations for people and groups with similar interests [59] |
17 | Civic Engagement | “Working to make a difference in the civic life of one's community and developing the combination of knowledge, skills, values and motivation to make that difference.” [97:vi] “Creating a public forum and optimal conditions for engagement” [148:1005] | Voter turnout [252] Attendance of peaceful demonstrations [372] Approximate total hours a month active in voluntary organizations [372] Donations to a charity in a month [242] | Encourage people to vote [38] |
18 | Community, Belonging | “[T]he feeling that people matter to one another and to the group, and a shared faith that members’ needs will be met through their commitment to be together.” [219] “The concept of members sharing mutual concern and/or love for one another” [183:15]. | Sense of belonging to a neighborhood [340] The General Sense of Belongingness Scale [212]. Ex: “I have a place at the table with others” | Recommend community-oriented groups User affordances for donating to charitable causes. |
19 | Accessibility, Inclusiveness | Inclusiveness in design means that diverse people and perspectives should be involved in the design of AI systems. Inclusiveness in impact calls for a just distribution of AI benefits and harms.[116:51–53] | Assessing the experience of different user demographics through surveys and other methods, such as visually impaired or hard of hearing users [31, 233] | Implement Web Content and Accessibility Guidelines (WCAG) Variety of formats for cognitive impairments (e.g., screen readers, transcripts, captions) |
20 | Tolerance, Constructive Discourse | Tolerance “creates the opportunity for a wide range of political groups to express their ideas and to participate in public life” [327] Polarization divides society into “us” and ”them” camps and contributes to the erosion of democracy [218] | Polarization measures such as issue-position or affective polarization [43, 256, 378] | Increase ideological diversity, select for civil conversations, optimize polarization measures [321] |
21 | Duty | The notion of duty and obligation is defined in contrast to self-interest [158]. | Personal control and responsibility scale [220] | User affordances for donating to charitable causes. |
22 | Care, Compassion, Empathy | “The ethic of care emphasizes the importance of context, interdependence, relationships, and responsibilities to concrete others.” [188] and is “based on the recognition that self and others are interconnected” [309]. These notions also appear in Confucian and Buddhist traditions [350:132] and in Ubuntu values where “a person is ‘a person through others’” [243]. | Empathy quotient [22]. Ex: “I find it easy to put myself in somebody else's shoes’’ and ‘’Seeing people cry doesn't really upset me ‘’ | Best practices in suicide prevention [278] |
23 | Fairness, Equality, Equity | “A morally justifiable distribution of benefits, goods, harms, and risks” [351]. In multi stakeholder fairness, different stakeholders have different objectives and needs from the system [1]. | Fairness metrics in classification and recommendation [34, 271] | Add fairness measures to the recommender training or serving objectives [34, 222] |
24 | Diversity | “[D]iversity refers to the idea that in a democratic society, informed citizens construct their worldview from a diverse set of sources which helps them to make balanced and well-considered decisions.” [150:192]. “‘Cultural diversity’ refers to the manifold ways in which the cultures of groups and societies find expression.” [343 art. 4] | Slate and feed diversity metrics of various kinds (Kunaver and Požrl, 2017) Perceived diversity and user satisfaction (Kunaver and Požrl, 2017) | Methods to increase various types of diversity, including topical and source diversity, novelty, coverage, etc. [60, 194] |
25 | Accountability | Accountability is “a component of the state of being responsible, alongside being answerable and being attributable” [215] AI principles documents include “verifiability and replicability,” “impact assessments,” “environmental responsibility,” “evaluation and auditing requirements.” [116:28] | Ranking Digital Rights Corporate Accountability Index [273] | Explanations of decisions according to clearly defined principles. Processes for users to bring up problems and resolve them in a timely manner. |
26 | Transparency and Explainability | Transparency means that information provided about a system is a) meaningful, b) useful, c) accessible, d) comprehensive, and e) truthful [162:71] “Transparency is instrumental to uphold intrinsic values of human autonomy and justice” [56:20] Desirable properties of transparency metrics [312] “[162:49, 312] | System cards which provide stakeholders with an overview of different components of an ML system [4] | Explanations for recommendations, answering questions such as “why am I seeing this?” or “Why has this been removed?” Publicly disclose and discuss changes to moderation or ranking [322]. |
27 | Accuracy (Factuality) | Accuracy is “an AI system's ability to make correct judgements, for example to correctly classify information into the proper categories, or its ability to make correct predictions, recommendations, or decisions based on data or models” [365] AI systems should make “accurate information readily available” [133] and should not spread “untrustworthy information.” [347] | Credibility metrics including prevalence of misinformation items among recommendations, and user engagement on this material. | Misinformation classifiers [45] Prompts to consider accuracy before sharing [262]. |
28 | Tradition, History | “Heritage is a concept to which most people would assign a positive value. The preservation of material culture ... and intangible culture … are generally regarded as a shared common good by which everyone benefits.” [299]Practices, symbols, ideas, and beliefs that are developed by groups and represent their shared experience and fate. They often take the form of religious rites, beliefs, and norms of behavior [293:6]. Rituals or conventional norms are one of the five central virtues of Confucianism [77] | Percentage of locally produced content. | Promotion of cultural events [62] |
29 | Environmental Sustainability | “Respect for environment and natural habitat, efficiency, maintainability, operability, supportability, reliability, durability, resilience, forgiveness, robustness, redundancy, reusability, reconfigurability, simplicity, economy, renewability” [162:70] “Develop and scale up carefully assessed technologies, infrastructure and actions that reduce climate change and its associated risks.” [344] | Product level carbon footprints [224] | Recommender systems can potentially direct users toward climate-friendly options [279:25] Reporting on the energy used by ML models [152] |
30 | Progress | The International Covenant on Civil and Political rights includes the right of people to “freely pursue their economic, social and cultural development” [345] as expanded in the Declaration on Social Progress and Development [346]. | UN Human Development Index [313] | To the best of our knowledge, none |
31 | Labor | Meaningful work is both significant and positive in valence [281]. Job insecurity, low employability, and unemployment are all detrimental to well-being, beyond the effects of income loss [136] | Work and Meaning Inventory [314] | Ensure users see recommendations for jobs that they would want. |