Lessons learned from AI ethics principles for future actions

“[7] review 84 AI ethics guidelines publicly released.”

“[8] 2020 study of a different sample of documents shows, while there are certainly points of convergence, by no means is there unanimity across the 36 principle documents reviewed.”

“In reviewing the principles and frameworks, and in building a roadmap and choosing our next actions, we need to remember the questions Cath [54] raised, “who sets the agenda for AI governance?”

“Such principles slot neatly into existing corporate playbooks, rarely questioning “the business culture, revenue models, or incentive mechanisms that continuously push these products into the markets” [31, 43].”

“The third wave aims at promoting practical accountability mechanisms [24].”

“In addition, many individual journal articles [260, 261, 262, 263, 264, 251, 265, 266, 267], and a number of academic journals dedicated special issues to contribute to AI ethics.”

“According to a review of publications on AI ethics, these documents share similar values and priorities in terms of content (Hickok, 2021).”

“…accountability, and societal impact, addresses the surge in ethical challenges posed by AI’s increasing complexity and potential misalignment with human values, and requires ethical governance frameworks, multidisciplinary collaborations, and technological solutions [214], [235], [15], [236].”

“1 The third wave of scholarship on ethical AI, which builds on the first two by exploring how to turn AI principles into actionable data science practice and governance (see for instance [10, 12, 24, 29, 62]), examines as such the spectrum of applicability of AI ethics in real-world cases [45], highlights issues of st…”

“…2020; Estrada 2020; Mittelstadt 2019b; Hagerty and Rubinov 2019; LaGrandeur 2020; Lauer 2020; Sloane 2018), as well as the contextual applicabil‑ ity in some areas of these ethical guidelines (Ryan and Stahl 2020; Morley et al. 2020; Hickok 2020; Floridi 2019; ÓhÉigeartaigh et al. 2020).”

“However, very few show the principles’ practical implementation to focus on actionable items inside the organisations (Canca 2020; Floridi 2019; Hickok 2021; Mark and Anya 2019; Stahl et al. 2021; Vakkuri and Kemell 2019).”

“Hickok (2021) also calls for greater diversity amid a need to progress from high-level abstractions and concepts in favour of applied ethics which establish accountability.”

“Regulation and establishing solid ethical standards are essential to ensure that AI is used responsibly and with respect for people’s rights and well-being (Hickok 2021).”

“…structures proposed by an individual or group), and legislation (legal rules established and enforced by a governing body) for AI transparency around the world, taking a narrower and updated view of this specific issue that pairs with broader analyses of global AI policy such as Hickok [ 27].”

“Lastly, our findings offer practical implications for policy makers who are to assess the impact that AI has on various stakeholders in society (Hickok, 2021).”

“Although these recommendations have served as building blocks of a common ethical framework for AI, less guidance exists on how these ideals must be translated into practical considerations [3–6].”

“Such a determination must be made by engaging, internally and externally, multidisciplinary public officials and impacted communities in the decisions (Hickok, 2021).”

“NGOs, research institutes, companies, and government agencies have published more than 100 AI ethics guidelines [15], and research analyzing AI ethics guidelines is also actively underway.”

“Algorithmic systems which violate fundamental human rights and human dignity should not be legitimized by principles [101] or through the use of risk management systems.”

“…for the specification of AI ethics guidelines (Alshammari and Simpson 2017; Gasser and Almeida 2017; Butcher and Beridze 2019; Jobin et al. 2019; Leslie 2019; Krafft et al. 2020; Hickok 2021; Metcalf et al. 2021; Seppälä et al. 2021; Stix 2021; Ayling and Chapman 2022; Georgieva et al. 2022).”

“This is a much deeper engagement than post-facto judgement using abstract ethical principles as in most AI ethics debates (see Jobin, Ienca, and Vayena 2019; Fjeld et al. 2020), without any clear result (Hickok 2020).”

“…based on self-regulation that permits one to avoid regulations or to retrofit one’s ethical standards by ethics shopping one’s principles, viz., when someone cherrypicks the principles that better suit so to make one’s AI uses ethically acceptable rather than changing one’s behaviour [45, 93, 119].”

“…offer valuable orientation but cannot necessarily provide the specific perspective needed for individual organisations with their unique digital challenges, needs and values, or for translating high-level principles into robust practice (Mittelstadt 2019; Floridi 2019; Hickok 2020; Blackman 2020).”

“These concerns appear to have stayed relatively consistent over time (Roberts et al., 2021), which is somewhat problematic given the limited set of stakeholder perspectives contained in the field of principles (Hickok, 2021).”

“Problems of systemic racism and other culturally deeply encoded discriminatory mechanisms multiply the severity of such conflicts of interest insignificant, perhaps non-linear and even counter-intuitive ways [22, 24, 36, 38, 39, 44].”

“…and challenges that are being created with the rise in the use of AI across sectors of society, many organizations, both public and private have come up with and published frameworks, principles, and guidelines on the ethical use of AI (Jobin et al., 2019; Hickok, 2020; Zicari et al., 2021).”

“Frameworks, tools, and checklists are some attempts at a technical fix to ethical issues raised in high-level AI ethics principles, but tangible action needs to move from high-level arguments to practice-based accountability and ethics mechanisms [16].”

“…example, argues that the “new field of AI ethics still needs a lot of work to incorporate the wisdom and values of different cultures” (Hickok 2021, p. 45), and “[t]he development of future work needs to stay true to the principles’ claims to be global and beneficial to all” (Hickok 2021 p. 46).”

“This trend also known as principalism (Clouser & Gert, 1990) has raised questions about the applicability of these principles in practice (Hickok, 2021; Mittelstadt, 2019), which has led to an increasing volume of research about operationalisation of ethics principles (Bleher & Braun, 2023; Morley…”

“Yet, there is uncertainty and debate about whether this frame is likely to influence policy in a meaningful way (Hickok, 2021; Morley et al., 2023; Taeihagh, 2021), given numerous barriers to operationalizing AI ethics and concerns about private and public sector “ethics-washing” and “ethics…”

“AI should be integrated into a governance framework that abides by ethical standards (Hickok, 2021; Samarawickrama, 2022) and ensures compliance with human rights norms through oversight and external regulation (Yeung et al., 2020).”

“…OECD, 3 as well as a general groundswell in academic and organisational interest in proposing and adopting different AI principles and guidelines [46], such principles are also beginning to court attention and critique, in some cases critiques extending to the whole endeavour of AI ethics itself…”

“While AI is often praised for its transformative potential, it also raises ethical concerns [10].”

“It will also be difficult to predict AI behaviours and identify errors that can lead to biased decisions, unforeseen risks, and further user trust concerns [4,18,27,59].”

“As important as such initiatives are for raising awareness for ethical issues of AI, they have been criticized as being too abstract [39, 57] and offering little to no practical applicability [66, 73, 75].”

“Future research could address gaps that arise in resolving these dilemmas.”

“In this spirit, future studies should further investigate how organizational constraints do influence XAI deployers’ alignment with specific ethical stances and their willingness to express dissenting views (Hickok, 2021; Ibáñez & Olmeda, 2021; Kitamura et al., 2021).”

“Yet even after a first wave of responses coalesced around Ethical AI principles (Hickok, 2021), a second wave cultivated Ethical AI technical tools, and now a third wave is motivating litigation and advocacy (Kind, 2020), published Ethical AI principles and technical tools still have limited impact on the daily practi…”

“To ensure that HCAI has a positive impact on patients, it is crucial that AI ethics reflects the values that are important to people [12, 14], but it remains unclear how this should be achieved.”

“Over the last years, substantial work has thus been dedicated to “lowering the level of abstraction” [10] and to translating ethical principles into actionable and specific practical requirements in AI governance [6, 11].”

“This is consistent with existing organizational efforts for AI trustworthiness (e.g., AI principles [59, 90, 138], toolkits [14, 77], alliances 4 , and pacts 5 ).”

“As the future unfolds, the field should remain vigilant by incorporating lessons learned from prior waves of AI research (Aliferis et al., 2024; Hickok, 2021).”

“Those fears have led to the surge of new reports, articles, and principles that seek more trustworthy AI (TAI) visions that provide a Human-centered vision of users’ trust and socio-ethical characteristics towards AI [23,38,83,85,86,88].”

“…by some for being (i) too abstract to be practical, (ii) reflecting mainly the values of the experts chosen to create them (hence, not being inclusive), and (iii) serving the priorities of the private entities which funded some of this work (‘ethics washing’) (Hagendorff 2020; Hickok 2021).”

“We hold that this contribution is a first step to address the call expressed in [12] “for tangible action to move from high-level abstractions and conceptual arguments towards applying ethics in practice and creating accountability mechanisms.””

“Recently, government and business entities have also started to emphasize the need to account for the ethical implications of using AI systems (Hagendor ff , 2020; Hickok, 2021).”

“In general, the guidelines do not rank their core principles a priori, which lets them appear to be of symmetrical value the undifferentiated call for XAI bears a disguising effect and encourages so-called “ethics shopping” [ 64, p. 42].”

“Recent work highlights the need to address emerging AI challenges through responsible development of AI and the need for participatory engagement in the creation of documents addressing the ethical implications of AI across organizations in all sectors [9, 10, 11].”

“Those who value AI ethics note that ethical principles can be the basis for solving complex ethical problems, expected along with advances in AI technologies (Hickok, 2021).”

“This observation aligns with the current state of AI ethics research which—as denoted by (Hickok, 2021)—is transitioning from principle affirmation to operationalization (Hagendorff, 2019; Morley et al., 2023).”

“Some view them as futile due to the lack of legal enforcement, while others recognize their positive role in setting expectations and standards for evaluating AI output (Hickok, 2021).”

“This requires operationalising high-level guidance to match an organisation’s structure, workflows and unique digital challenges (Floridi, 2019; Mittelstadt, 2019; Morley et al., 2020 ; Hickok, 2020; Blackman, 2020; Mökander et al., 2022b ).”

“…in trying to manage the ethical development of AI tools as evidenced by the wide proliferation of AI ethics principles and guidelines, as well as regional efforts to regulate it and protect individuals and society from potential harms (see the European Union’s upcoming AI regulation) (Hickok 2021).”

“These data suggest an under-representation of society’s needs and an over-representation of (mainly) US companies’ interests (Hickok, 2021)—a conjecture strongly reinforced by the lack in the final document of a declaration of no conflict of interest (Charlesworth, 2021).”

“…and Rodrigues [124, p. 2] argue, the “teeth” of ethics can be found in its “continuously refreshed and agile attention to reality as it evolves”. outline of a shared foundation upon which one can build and use as a benchmark to communicate expectations and evaluate deliverables” [32, p. 44].”

“Following the development of AI ethics guidelines and codes by multiple countries, private companies, and non-governmental organizations (NGOs), the task of practically implementing AI ethics has been rising to prominence as a pressing issue [21, 47, 79].”

“…the framework is consistent with the prevailing academic discourse that ethical guidelines such as the CoDE must be translated into practice to provide a valuable impact for organizations (Blackman, 2020; Floridi, 2019; Hickok, 2020; Mittelstadt, 2019; Mökander et al., 2022; Morley et al., 2020).”

“Thisisespeciallytrueforcertain disciplinessuchas‘AIethics’(Hickok,2020),wherepeoplelookforwaystogobeyondabstract principles—tomovefromtheorytopractice,or,asMorleyetal.”

“The lack of diversity in tech teams and a dominant focus on Western values have proven a major obstacle to developing AI applications that do not discriminate or exclude certain groups [22].”

“For instance, scholars working in the area of ethics for Artificial Intelligence (AI) have observed a particular ethical ‘frame’ that prioritises technical solutions [15, 16].”

“…and workflows (Mittelstadt and Floridi 2016; Rochel and Evéquoz 2021; Floridi et al. 2018; Jobin et al. 2019; Adams 2024; Amugongo et al. 2023; Kri-jger et al. 2023; Fjeld et al. 2020; Hagendorff 2020; Schiff et al. 2021; Floridi 2019; Morley et al. 2020; Hickok 2020; Giovanola and Tiribelli 2023).”

“…there are several consequences of applying ethical standards in the process of AI development and utilisation owing to the absence of agreed standards, which hinders the realisation of actual ethical standards through arbitrary and biased applications (Rességuier and Rodrigues 2020; Hickok 2021).”

“…aspects of AI systems, and to help steer the development and deployment of such systems in directions that are ethically and socially benefi - cial or preferable (Floridi 2019; Floridi et al. 2018; Hickok 2020; Jobin et al. 2019; Morley et al. 2020; Sætra and Dana-her 2022; Van de Poel 2020).”