Explainable Artificial Intelligence and Legal Ethos: Developing Key Performance Indicators for ‘G20 Giants’

Introduction

The accountability, transparency and the legal liability of Artificial Intelligence (AI) systems have also evolved with the growing usage of these systems due to their complexities and autonomy. In cases of AI failures, assigning responsibilities and understanding how AI systems make decisions has brought to the forefront the question of its “explainability” (Gilpin et al., 2018; Hacker et al., 2020). To address this concern, the EU’s General Data Protection Regulation (GDPR) allows individuals to seek insights into decisions of AI systems (Gilpin et al., 2018). Conversely, for a country like India having complex and heterogeneous cultural and social contexts, applying this right to regulate AI systems poses significant challenges.

Globally, despite advances in the research to enhance the explainability of AI systems, the hitherto proposed frameworks are still devoid of due considerations for diversity in cultural, social and ethnic fabric of stakeholders. Most of the studies indicate that Western models apply universally; this necessarily does not take into account the non-Western, collectivist societies (Peters & Carman, 2024). Existing frameworks also incorporate transparency at the cost of political and economic ideologies on the explainability of AI systems. As a result, such systems are culturally biased and may lead to inconsistencies, if used globally (Prabhakaran et al., 2022). Globally relevant and harmonized AI regulations must embody the core principles of transparency, accountability, security and dynamic societal adaption (Bhatt, 2025).

Culturally adaptive and stakeholder-sensitive AI systems are the need of the hour. AI regulating frameworks must consider cultural, socio-political, ethical and legal heterogeneity across different regions. To ensure equitable and purposeful AI explanations, we need to shift our focus to development of culturally adaptive and stakeholder-sensitive ‘Explainable AI’ (XAI) models. The ‘PEEC Doctrine’ propounded by Bhatt & Bhatt in 2023 (Bhatt & Bhatt, 2023) is one promising idea. This proposal not only integrates universally accepted theories of Public Interest, Environmental Sustainability, Economic Development, and Criminal Law (PEEC) to create a realistic approach to development of XAI, but also focuses on transparency considering the broader social, economic, political and legal impacts of AI decisions. The theory has the potential to address AI explanations by duly considering the ‘PEEC’ elements to promote sustainability and ease of access to explainability while ensuring sufficient accountability with a multi-dimensional perspective on XAI systems to serve the real-world societal needs.

AI/ XAI systems are complex algorithms blending social, ethical and human values. Human perceptions, values and interpretations are crucial in determining the success of these systems. However, conflicting goals and objectives, dynamic and unpredictable and an unforeseen environment, value-laden issues, and a complex interplay between human values, technology, and societal norms calls for a structured and iterative methodological approach besides purely technological or legal strategy to deal with the issue.

In these contexts, the study aims to investigate ‘Right to Explanation’ and ‘PEEC Doctrine’ by duly considering diverse cultures and values of ‘G20 Giants’ (USA, Germany, Japan, India, Brazil, & Russia) using Soft System Methodology (SSM) to develop Key Performance Indicators (KPIs) for adaptable, transparent and culturally sensitive XAI regulations that would enhance the trust and efficacy of AI systems worldwide.

1. Ethical and Cultural influences on XAI in ‘G20 Giants’

Sizeable differences in the cultural and ethical core values across many nations especially in terms of individualism (personal autonomy & self-determination), collectivism (Prioritization of Group Solidarity and Communal Well-being), trust in technology (Confidence in Digital Innovations and Automated Systems) and respect for authority (Adherence to Institutional Hierarchies and Governance Structures) has been highlighted by many recent cross-cultural research (Triandis, 2018; Jan et al., 2024). For instance, the United States and Germany are considered individualistic societies, operationalizing personal autonomy and self-reliance (Triandis, 2018). Conversely, cultures like Japan and India, also known as collectivist cultures, emphasize group well-being and social harmony (Eckhardt, 2002). Research evidence suggests that technological diffusion in developed countries rarely reaches the technology adoption rates of developed countries, largely due to socio-economics and digital literacy constraints (Comin & Hobijn, 2011). These cultural features influence policy decisions, societal behaviors, and international relations.

Deeply ingrained societal norms, historical contexts, and national/regional ideologies and policies form the basis for variance in cultural and ethical values across various countries. Table 1 illustrates how different nations uniquely deal with key cultural dimensions.

These elements profusely influence how XAI regulating policies are framed, which shall promote cross-cultural collaboration for universally relevant XAI regulating models simultaneously. This would also guarantee its strict alignment to corresponding societal expectations and values.

2. Existing Provisions of ‘Right to Explanation’ across ‘G20 Giants’

The ‘Right to Explanation’ has been brought into sharp focus due to increasing universality of legal and ethical debates on AI systems. Many researchers are of the view that this right of explanations may not always be practical and sufficient (Edwards & Veale, 2018; Taylor, 2023; Doshi-Velez et al., 2017). The developed nations, particularly the EU, have established frameworks to deal with complexities in AI decision-making by way of ‘Right to Explanation’. Whereas, the developing world is facing a lot of hurdles including legal and technical intricacies in accommodating ‘Right to Explanation’ into existing frameworks. The practical implementation of the right remains a challenge for developed countries. While a statutory ‘Right to Explanation’ is a potent mechanism empowering an individual to comprehend and challenge automated systems, its effectiveness depends upon establishing complementary mechanisms like impact assessment and judicial review. To safeguard potential biases and discrimination in automated decision-making, some EU member states have incorporated mandatory impact assessment into their national legislation (Malgieri, 2019). Judicial review provides an additional layer of control and accountability and ensures fairness in automated decision-making (Gacutan & Selvadurai, 2020; Malgieri, 2019).

The complexities of machine language limit the ability of AI developers and operators to provide meaningful and comprehensible explanations for laymen. This calls for a balanced approach whereby neither excessive control nor non-interference circumvents the development of AI systems. Use of AI systems for sectors such as public administration and healthcare ought to meet the standards of safety, transparency and accountability in diverse socio-technical and legal contexts.

Table 2 provides comprehensive information on ‘G20 Giants’ having provisions of ‘Right to Explanation’ and corresponding sectors where XAI are currently being employed or planned to be employed.

3. Assessing Applicability of ‘PEEC’ Doctrine for XAI in ‘G20 Giants’

A qualitative assessment of the ‘PEEC’ framework proposed by Bhatt & Bhatt, 2023 is imperative to validate the diverse approaches to AI regulations and context-specific policies that reflect each country’s unique socio-cultural and political landscape. AI regulating policies across countries are shaped by their respective socio-cultural priorities and governance ideologies. The elements of ‘PEEC’ framework, viz. public interest, environmental sustainability, economic development and criminal law have to be evaluated accordingly.

Public interest focuses differ from country to country. The United States prioritizes consumer protection and Germany emphasizes privacy of data, while countries like India and Japan are more inclined to social harmony and equitable access. Countries like Brazil and Russia intend to address governance failures and ensure state security. When it comes to environmental sustainability, countries like the United States and Germany aim to leverage AI for private sector innovation and improving industrial productivity and efficiency. Japan and India are more inclined to achieve long-term goals in smart city planning and water management. Both Brazil and Russia understand that AI systems can help attain environmental sustainability. Brazil focuses on combating climatic issues while the Russian approach is more focused on the energy sector.

Economically, Russia and Brazil thrive to drive state-led innovations and technological upgradations, while the United States and Germany endorse innovation with structured labour protection and concerns. Japan and India are both keen on developing robotics and finance technologies. AI use for criminal laws also varies significantly. While the United States balances security and personal liberty, Germany emphasizes oversight. Japan employs AI with checks and balances, the Indian approach is to develop safeguards. The Brazilian approach is all about tackling whereas Russia prioritizes security through surveillance. This diverse set of considerations highlights the intricate interplay that ought to be considered for XAI development worldwide.

It is worthwhile assessing the fitness of ‘PEEC’ principles as proposed by Bhatt & Bhatt, 2023 for ensuring that the XAI regulations shall remain effective, contextual and aligned with societal expectations worldwide. Table 3 shows the fitness of ‘PEEC’ principles for development of XAI across ‘key G20 economies’.

4. Developing Integrated Key Performance Indicators (KPIs) for XAI

A purely technical or quantitative approach cannot entirely cover all the inherent complexity, subjectivity, and ethical dimensions of the XAI regulating framework. To forge a truly comprehensive and robust framework, it is imperative to actively engage and incorporate the diverse perspectives of all stakeholders involved in AI systems. Soft systems methodology (SSM) is one interesting approach for tackling problematic and messy situations of diverse varieties, particularly those involving human systems (Checkland & Poulter, 2020).

The SSM allows users to deal with complex technical, political and socio-cultural problems in an organized manner and forces them to look for a holistic solution. The integration of ‘PEEC’ principles and ‘SSM’ can provide a potent tool for development of integrated KPIs for practical policy regulations on XAIs.

To truly tap the intricacies of the explainability of AI systems, one must look beyond just the technicalities involved. The authors proposes a novel approach of developing a comprehensive ‘Key Performance Indicators’ (KPIs) that would bring on board due considerations of Public Interest, Environmental Sustainability, Economic Development, Criminal Law through the structured procedure of ‘Soft System Methodology’ for a holistic assessment of AI’s implications on social, economic, political and legal regimes. Figure shows the conceptual model adopted for the said purpose.

Conceptual Model for Development of KPIs

To develop robust and universally acceptable KPIs for XAI, the SSM approach was employed to incorporate and integrate key dimensions of public interest, environmental sustainability, economic development, legal issues and governance. The structured and iterative process involved in SSM ensured that the developed KPIs would be apt and fitting for diverse cultural and legal contexts. Firstly, a thorough ‘Problem Identification’ for analysis of challenges to AI’s transparency across different landscapes was explored. Secondly, a comprehensive literature review of existing AI policies, academic research publications and news articles were critically analyzed for conducting a ‘Rich Picture Analysis’ for visually mapping the expectations of key AI stakeholders, viz. policy makers, public, industry and legal experts. Thirdly, for refinement of ‘PEEC Dimensions & Impact Areas’, ‘Root definition’ and ‘Conceptual Modelling’ approach was employed to ensure its alignment with ideologies relating to socio-economic, legal, ethical and environmental sustainability aspects. Fourthly, a ‘Comparative Analysis’ for validation of real-world applicability of AI regulation was undertaken. Lastly, an ‘Iterative Refinement Cycle’ helped ensure that the developed KPIs were not just attuned to the needs but were also practically implementable, rendering streamlined data collection and measurable criteria. Table 4 shows a proposed comprehensive KPI framework for XAI.

For high-risk decisions where the consequences of an unexplained decision are severe, the proposed ‘Clarity and Trust Index’ (CTI) can be kept at 90 percent to 100 percent depending upon the requirement. CTI value can be as low as 50 percent to 80 percent for routine automated decisions and 70 percent to 90 percent for strategic decisions.

Ideally, the proposed ‘Bias Reduction Index’ (BRI) shall be 100 percent, though above 90 percent it shall remain acceptable in most cases. Theoretically, the proposed ‘AI Carbon Footprint Index’ (AICFI) shall be as low as possible. However, the ‘AICFI’ can also be attuned to suit the UN Sustainable Development Goals (SDGs). Most countries would prefer the proposed ‘AI Socio-economic Benefit-Cost Ratio’ (ASEBC) higher than 1.0 or greater, however, efforts must focus on maximizing net tangible socio-economic benefits. The proposed ‘Cultural & Legal Accountability Score’ (CLAS) shall ideally be 1.0, though a value higher than 0.9 in most cases would suffice the public expectations. Thorough analysis of country-specific context, stakeholder engagement and necessary understanding of cultural contextual factors shall influence the feasibility and desirability of proposed KPIs indicator values. The onus is on policymakers to devise accurate ranges to reflect national circumstances while remaining globally acceptable in contributing and realizing the true potential of XAI systems for the betterment of mankind.

Effective country-specific careful adjustments to the proposed KPIs will optimize it for universal use. While all the five KPIs are relevant for ‘G20 Giants’, their relative importance hinges on country-specific socio-cultural-legal contexts. The long-running regulatory debates and corporate initiatives in the USA demand higher ‘CTI’. While, in Germany, the provisions of GDPR suggest a need for higher ‘BRI’ and lower ‘AICFI’. Brazilian and Indian policies are more centered on having higher ‘CLAS’ and ‘ASEBC’. Russian policies look to leverage AI in governance and thus have higher ‘ASEBC’ with a firm strategic focus to uphold national sovereignty and integrity.

Conclusions

Diversities in cultural, ethical, socio-economic and legal choices made by society pose a mammoth challenge before the policy makers to develop a regulating XAI framework that fits international requirements. These factors are in fact, the limiting determinant that have the potential to clinch the success or failure of XAI regulations. Finding a path forward requires attention not only to technological aspects, but most essentially the human dimensions to it. ‘Right to Explanation’, ‘Public Interest’, ‘Environmental Sustainability’, ‘Economic Development’, and ‘Criminal Law’ (PEEC) are all exigent and shall ever remain central to all our attempts to regulate AI technologies.

The study proposes a novel KPI based regulating model for XAI based on the principles of PEEC using a structured approach of ‘Soft System Methodology’. To truly tap the potential of the proposed KPI model, countries must set nationally relevant indicator ranges that hold value internationally. Besides deracinating the problems of regulating XAI across diverse cultural, ethical and legal landscapes, the proposed model ensures that the XAI regulating framework duly considers the human dimensions as it seeks to harness the true potential of AI. This approach will inspire AI-driven society in the future.

The study only considers the cultural complexities of ‘G20 Giants’. Further research or investigations shall encompass a wider spectrum to include other developed and developing nations to make the XAI regulating framework adaptable to diverse national and global demands.