1. Introduction
A new era marked by artificial intelligence’s (AI) ubiquitous influence across many sectors has begun with AI technology’s rapid growth. Many industries, including healthcare, aerospace, banking, entertainment, and many more, have been impacted by this technological transformation, which is sometimes referred to as the “fourth industrial revolution.” These businesses are all trying to increase productivity and efficiency while cutting costs. In this sense, artificial intelligence (AI) describes computer programs that mimic human intelligence processes, carrying out or even surpassing human performance (1).
The application of AI technology is challenging, though. Biases from training data are known to be inherited by AI systems, which can have unforeseen repercussions and promote inequality in a variety of domains. Examples of this problem include instances of gender bias in research publishing and racial prejudice in healthcare projections (2). These biases have sparked questions regarding the reliability of AI systems and their opaque decision-making procedures, especially because sophisticated AI technologies like deep learning are still difficult for people to understand (3).
To ensure responsible use and shape the development of AI technology, it is imperative to define ethical rules and guidelines in light of these challenges. Notably, leading technology corporations have taken action to regulate their AI endeavors, such as Microsoft with its Responsible AI framework (4). Recognizing the strategic significance of AI for innovation, equity, and security, the US government has also joined the AI standards and regulatory space through the National Institute of Standards and Technology (NIST) (5). Furthermore, through its High-Level Expert Group on AI (AI HLEG), the European Union has been actively involved in creating ethical standards for AI, with an emphasis on an approach to AI ethics that is human-centric (6).
These many pioneering institutions’ conceptualization of these institutional ethical principles for AI technology provides insights into regulating AI’s social and technological advancement (7). Understanding the guiding concepts behind AI development and deployment is essential to ensure that these technologies remain reliable, open, and consistent with human values as they become increasingly integrated into our daily lives (4).
The proliferation of ethics guidelines by multiple organizations has fractured the debate on AI ethics, making it difficult to fully understand the field and making the pursuit of equitable implementation more difficult (8). Many organizations, such as user groups, government agencies, and developers, have published AI ethics principles (9). As a result, there are a lot of similarities and discrepancies between their efforts to create practical rules for the benefit of society (10). There needs to be a broad agreement on normative frameworks and standard norms for AI ethics (11). The central question is how to define “common good” and “social benefit” in an increasingly globalized and digitalized world (12). This calls for clear definitions of justice, human rights, and widely acknowledged values, as well as how to identify potential risks in AI applications that have the potential to support or contradict these values in various social and economic contexts (4).
This research is important because it offers a semi-systematic overview of governance, legislation, and ethics in AI and sheds light on how the area of AI ethics is developing (13). It tackles ethical issues and conflicts in formulating and disseminating ethical AI principles by classifying AI guidelines and pointing out institutional overlaps and omissions (14). As AI technology continues to advance in societal use cases, research helps to bring hidden tensions, fresh viewpoints, and tech-business social agendas to the fore (15). This promotes conflict resolution and progress. By offering insightful information for regulatory strategies and assurance services, this study adds to the continuing conversation on AI ethics (16). It guarantees stakeholders’ comprehension of AI technology’s performance, risk, and compliance (17). Additionally, by using framing theory to study institutional AI ethics principles and norms, it highlights the crucial roles that trust and understanding play in sophisticated AI technologies and their communication (18).
1.1 Literature review
1.1.1 Framing theory literature: a viewpoint for research and instrument for communicating AI ethics
One of the first academics to define the term “framework” was (19), who described frames as “schemata of interpretation” for understanding what has happened (20) frames assist in bringing seemingly unrelated occurrences into coherent wholes. The intricacy of framing was emphasized by pointing out that there might be frames inside frames (21). According to (3) framing is the process of choosing which parts of reality to highlight in a communication to support particular problem definitions, causal interpretations, moral assessments, or therapeutic suggestions (22).
The conceptualization and communication of climate change in Swedish agriculture were examined by (11), emphasizing the discrepancy between farmers’ perceptions and media portrayals of the issue. (20) used framing analysis to examine how the news media covered the IPCC Fifth Assessment Report on climate change to find dominating frames (23).
Research on framing in political science and sociology looks at the words, pictures, sentences, and ways that news items are presented, as well as the processes that shape them (24). Diverse theoretical and methodological approaches to framing have been given by many scholars [(25), Matthes, 2009 #629].
While framing and agenda-setting are similar, framing concentrates on the substance of issues rather than particular subjects (10). Discourse analysis and the idea of the explanatory theme are connected to framing (20). Four framing processes were distinguished by (26):
• Frame creation
• Frame Placement
• The consequences of frames at the individual level
• The audience role of journalists
1.1.2 TRUST framings serve as the study’s academic framework
Transparent and understandable AI systems are required to solve the “black box problem” in AI (4). To reduce dangers and improve confidence in AI decision-making processes, academics and organizations are developing technological and moral regulation strategies (9).
AI development and application heavily depend on the public dissemination of AI principles and guidelines (25). These published AI ethics principles do, however, have some notable distinctions, similarities, and conflicts (9). The project’s goal is to find important TRUST framings in texts, including AI concepts and guidelines (27).
AI principles and guidelines writings that incorporate issues such as interpretability, transparency, comprehensibility, and explainable AI are called transparent and understandable AI framing (The Royal Society, 2019, 28).
Safe and Reliable AI Framing: Covers safety management procedures, public reporting of issues and future goals, and reliability (4).
Human augmentation, user control, autonomy, and consent are the main topics of the User Control and Autonomy Framing [(4, Endsley, 2018) #634].
Data security, privacy, and the requirement for secure AI systems are all covered under the “Secure and Privacy AI Framing” (29).
Changing narratives surrounding the complexity, risks, and issues surrounding artificial intelligence, such as ethical conundrums, human resources, employment, rights, accessibility, fairness, non-discrimination, justice, inclusion, diversity, solidarity, accountability, whistleblowers, and AI audits; additionally, hidden costs associated with AI and responsible research funding (30). These scholarly frameworks provide a basis for comprehending the various facets of communication on AI ethics (31).
1.1.3 Research questions
RQ1: What kinds of frameworks are included in the text of the selected organizations’ AI principles and guidelines?
RQ2: How much do the framings that these institutions use correspond with or mimic the TRUST framings that are explained in this study? These frameworks include The Other Framings, User Control and Autonomy, Secure and Privacy AI, and Transparent and Comprehensible AI.
2. Methodology
The goal of the study is to examine AI ethics communication in the context of leading AI organizations’ AI principles and guidelines—Microsoft, NIST, and AI-HLEG, in particular— and to distinguish different framings in their communication about AI ethics. The TRUST is used to identify these framings—Framings from the AI literature review that were developed in the preceding part. The selection of these AI firms for analysis was done with great care to reduce the possibility of author bias. Other prominent AI organizations were not included in the analysis because of unclear institutional approaches to AI research, innovation, and self-regulation, ongoing ethical disputes that have been covered in the media recently (like Google’s Project Maven), or past ties to the author. The processes for gathering textual data and the researcher’s approach to locating frames in the AI messages of the selected universities are described in the part that follows.
Phase 1: The researcher gathered the text data from the open-access AI principles and standards published on the websites of the chosen three institutions. Table 1 contains the source links for this text data.
Table 1. Artificial Intelligence (AI) principles data for textual analysis as downloaded in Dec 2021.
2.1 Data sources
Phase 2: As Matthes (2009) noted in their methodical examination of media framing studies published in prestigious communication journals, frame analysis is an essential technique for closely examining the selection and prominence of particular components of a problem {(Guenther, 2023) #745}. The framings within the textual data were manually identified using the (3) concept of framing and the academic sources cited in the literature study. The framings included in the AI principles language of the chosen firms were identified using inductive and deductive methods (7). Based on the qualitative paradigm of frame analysis, which holds that frames are visible through particular words, this study explores framings using direct quotations taken from the selected AI pioneers’ recently developed and published AI principles and guidelines, making connections with different aspects of the current scholarly debate on AI ethics (32). During the textual study of Microsoft, NIST, and AI-HLEG’s AI principles and guidelines, the identification of frames was led by the systematic processes described by (22) in ‘Frames in Communication’.
Describing the process for identifying certain framings is crucial before providing the research and findings (33). “When researchers employ computer programs for analyzing large volumes of text, they must identify the universe of words that signal the presence of a frame,” according to guidelines (34). This study’s academic framing literature review phase found theme words indicative of the identified framings in the sample text on AI principles and guidelines. It is important to remember that identifying “frames in communication” entails being aware of the important points highlighted in a speaking act. In the methodology, which lacks uniform measuring standards, persuasive communication research adheres to four essential steps: (1) Identifying a particular problem, occasion, or person; these components define communication frames. (2) Isolating particular attitudes to understand how frames shape public opinion (32). (3) Determining an issue’s starting set of frames inductively to create a coding scheme. (4) Using the original set of frames that have been identified to select the content sources for analysis.
All of the methods above for finding framing were followed, except the second stage, which examined how frames influence public opinion, given the study’s goals and scope {(Mhlanga, 2020) #744}. Previous sections identified and explained specific topics, pertinent events, examples, AI actors, and the chosen sample institutions. The academic framing literature review portion identified and elaborated on an initial set of framings corresponding to the concerns covered. Regarding the last phase (32), the study’s introductory part detailed the textual selection of AI principles and guidelines taken from three institutional sources for analysis.
3. Results or finding
As was already mentioned, every institution’s AI principles should encourage risk reduction and problem-solving related to this new technology. This insight is related to Goffman’s person-role formula, which states that an AI actor’s social role is closely related to Its type. The framings of the AI principles and guidelines are soft (because there is no legal obligation) but strong (as they take into account each position’s/society’s role’s priorities) (35). The following two research problems are addressed by the AI ethics principles and guidelines text analysis:
RQ1: What framings can be found in the AI principles and guidelines text of the chosen institutions?
The High-Level Expert Group on Artificial Intelligence (AI HLEG) was established by the European Commission to foster trust in the AI system’s entire life cycle (from development to deployment, from planning and communication to policy and investment recommendations). They produced a comprehensive guiding document that is currently influencing Europe’s overall AI approach to empower, benefit, and safeguard European citizens (18). In addition to the guidelines, which are referred to as the “Ethics Guidelines for Trustworthy AI,” the expert group produced three other deliverables: the AI Ethics Guidelines document itself included Sectoral Considerations on the Policy and Investment Recommendations, Assessment List for Trustworthy AI (ALTAI), and Policy and Investment Recommendations for Trustworthy AI. The AI ethics standards serve as the cornerstone upon which more comprehensive texts are constructed. Following the foundation chapter on Ethics Guidelines, each extension above receives a full chapter treatment.
RQ2: Which of the institutional framings are the same as or similar to TRUST framings explained in this study? (Where TRUST Framings indicate Transparent and Comprehensible AI Framing, Reliable and Safe AI Framing, User Control and Autonomy Framing, Secure and Privacy AI Framing, and The Other Framings).
The principles that underpin the guidelines drafted by the AI high-level expert group are rooted in Ethics in Science and New Technologies and the Fundamental Rights Agency (36). These three components are adhering to legal requirements, upholding ethical principles, and providing assurance of “robustness” (specifically, “technical robustness” combined with safety measures for humans, animals, and the environment in a variety of settings, as well as fallback plans)—all taken from AI HLEG’s EU documents and assessment list for trustworthy-AI.
According to (9), the standards specify essential requirements that are not legally binding. Although the seven conditions don’t impose any new legal duties, they offer developers and stakeholders thorough guidance in persuading them to comply (6). Developing and implementing AI systems that meet the seven specified characteristics of AI HLEG would create reliable AI systems. The guidelines state that if AI applications respect the following:
1. Human agency and oversight.
2. Technical robustness and safety.
3. Privacy and data governance.
4. Transparency.
5. Diversity, non-discrimination, and fairness.
6. Societal and environmental well-being.
7. Accountability, then they will be considered trustworthy.
The guidelines’ text and their communication to the European Parliament (18) are related to the study’s other framings (diversity, non-discrimination and fairness, accountability) as well as the following: transparent and understandable AI framing, reliable and safe AI framing, user control and autonomy framing, secure and privacy AI framing, and user control and autonomy (26). Table 2 provides some sample quotes from chosen AI principles and guidelines data documents linked to the TRUST framings of this study (37). Refer to Appendix A, Tables 3, 4 in the ensuing sections for further AI ethical language framing examples from Microsoft, the EU’s AI HLEG, and NIST’s AI principles and guidelines (38).
Table 2. Examples of Identified Framings in the Institutional AI Ethics Principles and Guidelines Text data (EU’s AI HLEG, Microsoft, NIST).
Table 3. Identified Framings in the Institutional AI Ethics Principles and Guidelines Text Data.
Table 4. Examples of identified framings in the institutional AI ethics principles and guidelines text.
Transparent and Comprehensible AI Framing Because advanced artificial intelligence (AI) systems in social settings can be complicated, NIST, a federal non-regulatory agency under the U.S. Department of Commerce whose goal is to foster innovation and industrial competitiveness in the country, places a strong emphasis on “transparency” in its AI principles (39). The transparency of AI systems and their understandability by human recipients of the information are the foundations of three of the four NIST AI principles (40). NIST’s AI principles, which elaborate on the kinds, meanings, and precision of explanations, support The Royal Society (2019) assertion that there are several explainability approaches, which are covered under the Transparent and Comprehensible AI Framing in this study’s literature review. The principles of NIST reaffirm that the nature and specifics of an explanation would differ based on the application in question and the kind of AI technique created and implemented in a social context (41). The text under AI principles in Microsoft’s published case studies and video transcripts covers three AI framings: Secure and Privacy (words: Privacy and Security), Fairness, Inclusiveness, and Accountability, and Transparent and Comprehensible (words: Transparency and Explainability) (42). These are discussed in the academic frames section of this study’s literature review (for data examples, refer to Tables 3, 4)
3.1 Safe and dependable AI framing
AI ethical guidelines published by an organization are considered soft law or non-legislative policy tools with persuasive language but no legal force behind them (9). Through its three offices/committees—the Office of Responsible AI (ORA), the Aether Committee (which stands for AI, Ethics, and Effects in Engineering and Research), and the Responsible AI Strategy in Engineering (RAISE)—Microsoft operationalizes its AI principles, which it has dubbed “Responsible AI.” While the Aether Committee advises Microsoft’s senior leadership on responsible AI issues, technologies, processes, and best practices, RAISE is an initiative and engineering team designed to facilitate the implementation of Microsoft’s responsible AI rules and processes across its engineering groups (44). In summary, committees that advise Microsoft’s leadership, engineering, and all other teams inside the organization provide direction as it implements its responsible AI principles. Thus, the text’s six key AI principles come first.
4. Discussions
The debate highlights the significance of word choices and framing within AI principles and standards when examined through the prism of framing theory. The results of this study support the notions put out by (19) and (3) on the existence of frames within frames by showing that these frames might function as “signs of priorities” within these documents. For instance, Microsoft prioritizes some framings by partner needs. Still, it withholds the weight given to these framings across different industries, creating a lack of transparency in the process of deciding how certain settings will turn out. Contrastingly, the approach taken by the European Union, as described in the AI ethics document by (18), treats all framings equally. The research also emphasizes how convincing these documents are, despite not having legal force behind them, and how they add to the conversation about global AI ethics, governance, and legislation [(9), (18) #639].
The conversation emphasizes how international AI stakeholders must come together to create a single database with ethical norms and principles unique to AI. According to (13) this convergence is necessary to handle the difficulties and possible conflicts that may occur when giving particular AI principles, like fairness and priority. As it prepares the way for the creation of formal AI norms and laws for various societal scenarios, convergence in the framing of AI ethics principles is essential for building faith in the technology’s transformative potential (16). This talk emphasizes the importance of framing theory in understanding how AI ethical discourse impacts our future and the necessity for convergence to protect the common good in the setting of a global digital society.
This study, which focused on pioneering organizations like the European Commission and NIST in developing AI principles and standards, was confined to AI ethics draft texts available until December 2021. However, many actors from many sectors—including enterprises, academic institutions, national and international organizations, and more—are involved in the quickly changing field of artificial intelligence and are working on reports and frameworks related to AI ethics (45). Future studies should, therefore, take into account the dynamic field of AI ethical principles and delve further into the implications of these frames at the personal level (45). It should consider the difficulties that come with putting these ideals into reality and the diversity of values that exist among various socioeconomic classes and geographic regions. The three components of this research framework—developing AI ethics principles, applying them in particular contexts, and examining their effects on individuals and society as a whole—can greatly support moral behavior and just (14).
5. Conclusion
To sum up, this research explores the quickly changing field of AI ethics standards and principles, concentrating on trailblazing organizations like NIST and the European Commission. The study’s limitations, which only included draft texts accessible through December 2021, draw attention to the necessity for continued research in this rapidly developing sector. The significance of examining developing AI ethics frameworks is highlighted by the spread of AI technology and its interactions with diverse industries and societies. Future ethical studies in AI should consider the varied values found in various social groups and geographic areas, in addition to monitoring modifications to guiding principles and guidelines and investigating their consequences at the individual level.
Furthermore, since these are only the first steps, it is crucial to address the difficulties that come up when putting AI ethics concepts into practice. The present study underscores the significance of a thorough research methodology that encompasses three fundamental domains:
• Devising ethical guidelines for AI
• Executing them in particular situations or settings
• Examining their influence on individuals and society as a whole
In an AI environment that is always evolving, such research can substantially contribute to moral behavior and the fair application of AI ethics concepts.
In the end, as AI technology continues to change society, it will be vital for everyone to work together to create, modify, and apply AI ethics principles to make sure that AI upholds ethical standards, advances justice, and respects a variety of values.
References
2. Obermeyer Z, Mullainathan S. Dissecting racial bias in an algorithm that guides health decisions for 70 million people. Paper presented at the Proceedings of the conference on fairness, accountability, and transparency. New York, NY (2019).
4. Nagar N. Framing TRUST in Artificial Intelligence (AI) Ethics Communication: Analysis of AI Ethics Guiding Principles through the Lens of Framing Theory. Rochester: Rochester Institute of Technology (2022).
5. Sivan-Sevilla I. Complementaries and contradictions: National security and privacy risks in US federal policy, 1968–2018. Policy Internet. (2019) 11:172–214.
6. Parviala T. EU Entering the Era of AI: A qualitative Text analysis on the European Union’s Policy on Artificial intelligence. Brussels: European Commission (2019).
7. de Greeff J, de Boer MH, Hillerström FH, Bomhof F, Jorritsma W, Neerincx MA. The FATE System: FAir, Transparent and Explainable Decision Making. Paper presented at the AAAI Spring Symposium: Combining Machine Learning with Knowledge Engineering. New York, NY (2021).
8. Sarwar H, Ishaq MI, Amin A, Ahmed R. Ethical leadership, work engagement, employees’ well-being, and performance: a cross-cultural comparison. J Sustain Tour. (2020) 28:2008–26.
9. Jobin A, Ienca M, Vayena E. The global landscape of AI ethics guidelines. Nat Mach Intell. (2019) 1:389–99.
11. Asplund T. Climate change frames and frame formation: An analysis of climate change communication in the Swedish agricultural sector. London: Linköping University Electronic Press (2014).
12. Benefo EO, Tingler A, White M, Cover J, Torres L, Broussard C, et al. Ethical, legal, social, and economic (ELSE) implications of artificial intelligence at a global level: a scientometrics approach. AI Ethics. (2022) 2:667–82.
13. Binns R. Fairness in machine learning: Lessons from political philosophy. Paper presented at the Conference on fairness, accountability and transparency. New York, NY (2018).
14. Holton R, Boyd R. ‘Where are the people? What are they doing? Why are they doing it?’(Mindell) Situating artificial intelligence within a socio-technical framework. J Sociol. (2021) 57:179–95.
15. Caliskan A. Beyond Big Data: What Can We Learn from AI Models? Invited Keynote. Paper presented at the Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security. New York, NY (2017).
16. Friedler SA, Scheidegger C, Venkatasubramanian S. The (im) possibility of fairness: Different value systems require different mechanisms for fair decision making. Commun ACM. (2021) 64:136–43.
17. Caplar N, Tacchella S, Birrer S. Quantitative evaluation of gender bias in astronomical publications from citation counts. Nat Astron. (2017) 1:0141.
18. Hleg A. Ethics guidelines for trustworthy AI. B-1049 Brussels. Brussels: European Commission (2019).
19. Goffman E. Frame analysis: An essay on the organization of experience. Cambridge, MA: Harvard University Press (1974).
20. O’Neill S, Williams HT, Kurz T, Wiersma B, Boykoff M. Dominant frames in legacy and social media coverage of the IPCC Fifth Assessment Report. Nat Clim Change. (2015) 5:380–5.
21. Carabantes M. Black-box artificial intelligence: an epistemological and critical analysis. AI Soc. (2020) 35:309–17.
22. Aftab J, Sarwar H, Kiran A, Qureshi MI, Ishaq MI, Ambreen S, et al. Ethical leadership, workplace spirituality, and job satisfaction: moderating role of self-efficacy. Int J Emerg Mark. (2022) doi: 10.1108/IJOEM-07-2021-1121 [Epub ahead of print].
23. Chien S, Doyle R, Davies AG, Jonsson A, Lorenz R. The future of AI in space. IEEE Intell Syst. (2006) 21:64–9.
25. D’angelo P. News framing as a multiparadigmatic research program: A response to Entman. J Commun. (2002) 52:870–88.
28. Xu W. Toward human-centered AI: a perspective from human-computer interaction. Interactions. (2019) 26:42–6.
29. Harris J, Anthis JR. The moral consideration of artificial entities: a literature review. Sci Eng Ethics. (2021) 27:53.
30. Hagendorff T. The ethics of AI ethics: An evaluation of guidelines. Minds Mach. (2020) 30:99–120.
31. Hernández D, Cano J-C, Silla F, Calafate CT, Cecilia JM. AI-enabled autonomous drones for fast climate change crisis assessment. IEEE Internet Things J. (2021) 9:7286–97.
32. Druckman JN. The implications of framing effects for citizen competence. Polit Behav. (2001) 23:225–56.
33. Markus AF, Kors JA, Rijnbeek PR. The role of explainability in creating trustworthy artificial intelligence for health care: a comprehensive survey of the terminology, design choices, and evaluation strategies. J Biomed Inform. (2021) 113:103655.
34. Došilović FK, Brčić M, Hlupić N. Explainable artificial intelligence: A survey. Paper presented at the 2018 41st International convention on information and communication technology, electronics and microelectronics (MIPRO). New York, NY (2018).
35. Datzov NL. The Role of Patent (In) Eligibility in Promoting Artificial Intelligence Innovation. UMKC L Rev. (2023) 92:1.
36. Hugosson B, Dinh D, Esmerson G. Why you should care: Ethical AI principles in a business setting: A study investigating the relevancy of the Ethical framework for AI in the context of the IT and telecom industry in Sweden. Brussels: European Commission (2019).
37. Saetra HS, Coeckelbergh M, Danaher J. The AI ethicist’s dilemma: fighting Big Tech by supporting Big Tech. AI and Ethics (2022) 2:15–27.
38. Schnack H. Bias, noise, and interpretability in machine learning: From measurements to features Machine learning. London: Elsevier (2020). p. 307–28.
39. Shneiderman B. Human-centered artificial intelligence: Reliable, safe & trustworthy. Int J Hum Comput Interact. (2020) 36:495–504.
40. Siau K, Wang W. Artificial intelligence (AI) ethics: ethics of AI and ethical AI. J Database Manage. (2020) 31:74–87.
41. von Eschenbach WJ. Transparency and the black box problem: Why we do not trust AI. Philos Technol. (2021) 34:1607–22.
42. Warner R, Sloan RH. Making artificial intelligence transparent: Fairness and the problem of proxy variables. Crim Just Ethics. (2021) 40:23–39.
43. Whittlestone J, Nyrup R, Alexandrova A, Cave S. The role and limits of principles in AI ethics: Towards a focus on tensions. Paper presented at the Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society. New York, NY (2019).
44. Pitney AM, Penrod S, Foraker M, Bhunia S. A systematic review of 2021 microsoft exchange data breach exploiting multiple vulnerabilities. Paper presented at the 2022 7th International Conference on Smart and Sustainable Technologies (SpliTech). New York, NY (2022).
45. Wilson N. Understanding the Battle for AI in Warfare through the Practices of Assemblage: A Case Study of Project Maven. Brussels: European Commission (2020).
© The Author(s). 2024 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.