Source-linked AI summary
Ethics of AI: A Systematic Literature Review of Principles and Challenges
Arif Ali Khan, Sher Badshah, Peng Liang, Bilal Khan, Muhammad Waseem, Mahmood Niazi, Muhammad Azeem Akbar
TL;DR
AI ethics research has produced principles, but their practical implications and adoption challenges remain insufficiently understood. This paper conducts a systematic literature review of primary studies and identifies 22 principles and 15 challenges, with transparency, privacy, accountability, fairness, lack of ethical knowledge, and vague principles most prominent.
Problem
AI ethics principles and guidelines are debated, while tools, methods, and frameworks for translating them into practice remain lacking.
Method
The paper uses a systematic literature review guided by Kitchenham and Charters to analyze relevant primary studies addressing AI ethics principles and challenges.
Results
The review identifies 22 AI ethics principles and 15 challenges, with transparency, privacy, accountability, and fairness most common and lack of ethical knowledge and vague principles most frequently reported.
Takeaways & Limitations
The findings provide initial inputs for a maturity model intended to assess ethical capabilities and support best practices for improving AI ethics practice.
Takeaways & Limitations
The findings summarize 27 primary studies, and the authors caution that this sample may not be strong enough to generalize the results.
Abstract
from arXiv · showhide
Ethics in AI becomes a global topic of interest for both policymakers and academic researchers. In the last few years, various research organizations, lawyers, think tankers and regulatory bodies get involved in developing AI ethics guidelines and principles. However, there is still debate about the implications of these principles. We conducted a systematic literature review (SLR) study to investigate the agreement on the significance of AI principles and identify the challenging factors that could negatively impact the adoption of AI ethics principles. The results reveal that the global convergence set consists of 22 ethical principles and 15 challenges. Transparency, privacy, accountability and fairness are identified as the most common AI ethics principles. Similarly, lack of ethical knowledge and vague principles are reported as the significant challenges for considering ethics in AI. The findings of this study are the preliminary inputs for proposing a maturity model that assess the ethical capabilities of AI systems and provide best practices for further improvements.
1 Introduction
AI systems offer broad benefits but raise ethical risks because their autonomous decisions can affect human control, cultural values, and societal life. This study therefore reviews AI ethics principles and challenges to guide further investigation.
- AI technologies support industries including health, manufacturing, banking, and retail, but may produce ethical harm alongside economic benefits.
- Autonomous AI systems raise questions about potential risks, appropriate behavior, control, and the handling of AI-based decisions.
- AI development and use are culturally and ethically embedded, requiring economic, political, societal, intellectual, and legal considerations beyond technical work.
- The study conducts a systematic literature review to identify AI ethics principles and challenging factors that discourage consideration of ethics.
- The review asks which AI ethics principles are key and which challenges hinder adopting ethics in AI.
2 Background
Organizations have produced AI ethics guidelines and standards, but evidence indicates that principles are not effectively translated into practice. The background therefore identifies a gap in tools, methods, frameworks, and evaluation standards for implementation.
- Organizations including IEEE, ISO, IEC, technology companies, and civil-society groups have developed AI ethics principles, guidelines, and standards.
- IEEE’s Ethically Aligned Design framework addresses ethical and technical values through principles and recommendations for AI development and implementation.
- Existing AI ethics guidelines have been reported as ineffective or insufficiently adopted in practice, while the ACM code reportedly had no impact on software developers’ ethical decisions.
- No tools, methods, or frameworks currently fill the gap between AI principles and their practical implementation, motivating evaluation standards or models.
3 Research Method
The study uses a Kitchenham-and-Charters-guided systematic literature review to formulate research questions, search and filter studies, assess quality, and synthesize evidence. The process combines repository searches, inclusion and exclusion criteria, quality assessment, and author review meetings.
- 3 Research Method: The systematic literature review follows Kitchenham and Charters’ guidelines to explore primary studies and address the research questions.
- 3 Research Method: The review process includes planning, conducting, and reporting phases, with study selection, quality assessment, research-question mapping, synthesis, and validity analysis.
- 3.1 Research questions (RQs): Research questions were developed after studying relevant literature and were finalized around identifying AI ethics principles and challenges.
- 3 Research Method: The authors selected digital repositories through team discussions and SLR experience, then finalized search terms and strings through group discussion and pilot searches.
- 3 Research Method: Search results were filtered using inclusion and exclusion criteria, yielding 24 shortlisted primary studies after an initial retrieval of 811 studies and full-text review of 60 studies.
- 3.6 Quality assessment (QA): Quality assessment used QA1–QA6 criteria, assigning scores of 1, 0.5, or 0 according to whether each study comprehensively, partially, or insufficiently addressed the questions.
4 Reporting the review
The review analyzed 27 primary studies, with publications increasing from 2018 through early 2021 and journals serving as the dominant publication venue.
- 4.1 Temporal distribution: 27 primary studies were analyzed, including 2 published in 2021 through 5 February, 19 in 2020, 4 in 2019, and 2 in 2018.The search was executed on 5 February 2021, so 2021 coverage includes only the first two months.
- 4.2 Publication type: 19 studies (70%) were published in journals, compared with 3 (11%) in conferences, 4 (15%) in book chapters, and 1 (4%) in a magazine.Journals were the most active venues for relevant studies.
5 Detail results and analysis
The review identified 21 AI ethics principles and found that transparency, privacy, accountability, and fairness received the most attention. The discussion describes these principles as interconnected requirements concerning system visibility, data protection, responsibility, and nondiscrimination.
- 5.1 RQ1 (AI Ethics Principles): 21 AI ethics principles were extracted from 27 primary studies, with transparency, privacy, accountability, and fairness mentioned most often.The four leading principles were reported 17, 16, 15, and 14 times, respectively.
- 5.1.1 Transparency: Transparency addresses how and why an AI system makes decisions and supports interpretability, explainability, and trustworthiness.The review states that transparency should apply to both system operations and the technical process, with levels varying across stakeholders.
- 5.1.2 Privacy: Privacy concerns users’ control over information and becomes more challenging when data-driven systems clean, merge, and interpret user data.The passage frames privacy as a lifecycle requirement for AI and autonomous systems.
- 5.1.3 Accountability: Accountability focuses on liability by assigning responsibility, safeguarding justice, and preventing harm across system development, implementation, and operation.The review links accountability to transparency because liability decisions require the system to be understood.
- 5.1.4 Fairness: Fairness concerns discrimination by AI decision-making systems and seeks to avoid unfair bias that can affect dignity, justice, and social fairness.The discussion also connects fairness with transparent decision-making and identification of accountable entities.
- Analysis: The prominent principles align with the accountability, responsibility, and transparency framework, while responsibility may be less cited because it is associated with accountability.A related framework adds fairness to the ART constructs.
5.2 RQ2 (Challenges)
The review identifies 15 AI ethics challenges, with lack of ethical knowledge and vague principles among the most prominent barriers to applying ethics in practice. Conflicting interpretations, limited technical understanding, and limited evidence also constrain implementation and motivate a proposed maturity model.
- Vague principles: Vague, general, and ambiguous principles make it difficult for organizations to apply concepts such as fairness and human dignity in real-world AI settings.Organizations may interpret principles according to their own understandings because universally agreed principles are absent.
- Conflict in practice: Conflicting views among organizations and committees create divergent interpretations of how AI ethics should be implemented in practice.The review contrasts calls for human guidance of robots with autonomous decisions in hospital diagnosis and surgery.
- Lack of technical understanding: Limited technical understanding among policymakers and ethicists widens the gap between AI system design and ethical thinking.The review states that ethicists need to grasp technical knowledge through their ethical frameworks.
- Lack of ethical knowledge: Lack of ethical knowledge is identified as the most common challenge to mature AI ethics practice.The review links this gap to limited moral awareness, insufficient ethical standards, and practitioners’ limited attention to ethical aspects.
- Evidence and future development: The challenge evidence base is limited because few studies directly or indirectly discuss barriers to AI ethics.Low-frequency challenges receive little detail because of page limitations, although the complete factor list appears in Table 4.
- Proposed maturity model: The findings provide initial inputs for a proposed maturity model evaluating the ethical capabilities of organizations developing AI systems.Figure 8 presents a preliminary model structure incorporating the review’s principles and challenges.
6 Threats to validity
The study reports threats to validity concerning the completeness and quality of the selected literature and the generalizability of findings based on a small primary-study sample. The authors describe search and screening safeguards, while noting that broader industrial evaluation is still planned.
- Construct validity: The primary-study selection process may affect the quality of the synthesized data.The authors used a formal search strategy, pilot searches with multiple strings, consensus meetings, and backward snowballing to reduce this threat.
- Construct validity: The review follows formal SLR guidelines, with its process and phases described methodically.The study uses the Kitchenham and Charters guidelines and provides an SLR process plan.
- External validity: Findings based on 27 primary studies may not be strong enough to generalize across the field.The authors plan an industrial study to evaluate the findings and assess practitioners’ perceptions.
7 Conclusions and future directions
The review identifies a convergent set of AI ethics principles and challenges, highlighting four frequently emphasized principles and two major barriers. Future work will examine ethics in practice and evaluate a maturity model through industrial studies.
- The systematic review identifies 22 AI ethics principles and 15 challenging factors across 27 relevant primary studies.
- Transparency, privacy, accountability and fairness are the four principles most frequently emphasized for AI system designers.
- Lack of ethical knowledge and vague principles are the most frequently reported barriers to implementing AI ethics.Ethical knowledge is described as important for both management and technical teams.
- Future work will use an industrial survey to study practical understanding and best practices, followed by industrial case studies of the proposed maturity model.
S. No. Selected Primary Studies Q1 Q2 Q3 Q4 Total
The selected primary-study set includes work on AI ethics principles, operationalization, governance, privacy, fairness, accountability, and implementation challenges. The supplied records also show studies spanning healthcare, autonomous systems, business, and cross-cultural governance.
- The selected studies include work on AI ethics and ethical AI, such as Siau and Wang’s treatment of AI ethics concepts.
- Several selected studies address operationalizing principles and implementing AI ethics in practice.
- The study set covers domain-specific and governance concerns, including healthcare, privacy, China’s AI governance, autonomous weapons, and cross-cultural cooperation.