The Rise Of Explainable AI (XAI) - Bridging The Gap Between Legal Compliance And Managerial Transparency

Author: LawSphereX

Category: Technology, Cyber & Privacy

Published: 2026-08-23T21:28:43.409354+00:00

Updated: 2026-08-26T17:05:26.194+00:00

THE RISE OF EXPLAINABLE AI (XAI): BRIDGING THE

GAP BETWEEN LEGAL COMPLIANCE AND MANAGERIAL

TRANSPARENCY

Ms. Honey Verma, Ph.D. Scholar, Manipal University, Jaipur Mr. Uddhav Soni, BBA LLB (H), UPES, Dehradun

ABSTRACT:

Artificial Intelligence (AI) systems are becoming more and more prevalent in modern life. Despite their expanding sophistication, many AI systems remain "black boxes" that hide their decision-making from users and regulators. The GDPR mandates that all AI systems must meet certain standards of explainability and openness, which makes it challenging to comply with standards, especially GDPR-mandated openness in automated decision making. This paper explores the gaps between the legal and technical perspectives on AI transparency, and how can these differences be reconciled to develop effective regulatory frameworks for AI systems. We find that the transparency gap results from significant disagreements between technological norms and legal requirements for what constitutes openness.

Transparency, interpretability, and explainability are three fundamental concepts that should be specified and scoped in order to enhance legal certainty and guide the development of XAI systems. Since the legislation does not specify how to verify algorithmic transparency, it is unclear how providers may be made transparently certified while yet allowing providers to self-regulate.

1. INTRODUCTION

Explainable AI (XAI) advances AI to reconcile open management with legal compliance. Despite their expanding sophistication, many AI systems remain "black boxes" that hide their decision-making from users and regulators. Lack of transparency makes it challenging to comply with standards, especially GDPR-mandated openness in automated decision-making. XAI solves these problems by describing AI model logic. Fairness and transparency are XAI's legal guarantees. By doing so, businesses may show their AI systems follow all rules, including anti-discrimination legislation. To ensure ethical and legal compliance, businesses may analyze AI data to test algorithms for biases and errors. Openness is needed to win over customers, workers, and regulators. 1 Additionally, XAI enhances managerial transparency.

Managers may understand AI-driven processes better and make informed decisions about adapting them to corporate operations. This insight helps align AI with organizational goals to boost efficiency and effectiveness. Additionally, when AI solutions are easy to grasp and use, it promotes a trusting and collaborative work environment since workers are more inclined to embrace and use them. The advent of XAI heralds a sea change towards AI systems that are more open, responsible, and reliable. More responsible and sustainable AI deployment is possible because to XAI's efforts to remove the mystery from AI decision-making, which not only complies with legal and ethical requirements but also improves management supervision and strategic alignment.

1.1 Research Questions

1. How can the concept of "right to explanation" in the GDPR be practically implemented

in AI systems to ensure meaningful transparency for non-technical end-users?

2. What are the key differences between the legal and technical perspectives on AI

transparency, and how can these differences be reconciled to develop effective regulatory frameworks for AI systems? 1 Julija Sardelic, ‘Artificial Intelligence and the ́ Right to Explanation as a Human Right’ (2021) 25(2) IEEE Internet Computing 116–120.

3. In what ways can stakeholder-directed strategies be incorporated into the design of

explainable AI (XAI) systems to enhance trustworthiness and accountability?

1.2 Statement of Problem

By ignoring the socio-technical context and legal compliance in favor of a laser-like concentration on describing complicated algorithmic characteristics, XAI treats transparency as an objective in itself. We refer to this disparity as the "transparency gap". If XAI doesn't fix the transparency gap, many other transparency problems will remain unsolved. In order to start filling this gap, it is necessary to provide a fundamental explanation of transparency and explain the language used by XAI, European legislation, and the associated General Data Protection legislation (GDPR).

1.3 Research Methodology

The researcher has used a Doctrinal research methodology to procure information and data. The researcher has consulted various books, journal articles, judgments and commentaries to gather awareness and knowledge about the subject.

2. BRIDGING THE AI TRANSPARENCY GAP: PERSPECTIVES FROM LAW AND

TECHNOLOGY

2.1 AI Transparency Gap between Law and Technology

Everyone from business customers to non-technical hobbyists may now use Artificial Intelligence (AI) technologies, which are becoming more and more prevalent in modern life. Artificial intelligence (AI) may seem like a new area due to public demonstrations of its capabilities, but in reality, AI systems are already in common usage. Applications that have been using AI for a long time include medical diagnostics, recommender systems, and facial recognition. But concerns about the social impact of black-box AI have surfaced as its use in mission-critical applications grows. Officials have been busy formulating plans to meet the problems of the "AIera" after realizing how critical it is to do so. In the United States, the "Blueprint for an AI Bill of Rights" has been released by the White House, outlining five primary areas where AI regulation is proposed.

The United Kingdom is now developing a "National AI Strategy" that lays out its unique pro-innovation stance on

artificial intelligence legislation. The European Union is now debating a bold new law titled the AI Act. Providers of artificial intelligence (AI) systems are obligated by law to meet certain standards of explainability and openness. In order to define these standards, it uses a proportionate risk-based methodology and suggests thorough conditions for conformance evaluation. 2 Also, scientists agree that trustworthy AI that is open and honest about its intentions and that takes social and legal factors into account while still being able to handle real-world challenges effectively is necessary. Research into explainable AI (XAI) has recently shifted its emphasis to finding social explanations as a means to human-aligned AI, since researchers see explanations as crucial to transparency in human-AI interaction.

Nevertheless, technical alternatives do not agree on the extent, aim, or definitions of regulatory transparency, much less on how these concepts translate into more reliable systems. In XAI's narrow and technologically oriented perspective, transparency is just an algorithmic quality that provides practical answers. In contrast, openness is seen by the law as an attribute of the intricate socio-technical relationships that exist between AI and its many stakeholders, including users, developers, owners, and society at large. The "transparency gap" describes this discrepancy in perspective. Having a clear understanding of the law is not an end in itself, but it is a necessary tool to promote various principles. Therefore, EU legislation distinguishes between technical, enabling, and protecting transparency, each of which may be computationally tractable to different degrees.

On the other hand, XAI treats algorithmic transparency—also known as black box or interpretable systems—as a feature of computers on its own. According to this point of view, the complexity of AI systems makes transparency an objective in and of itself. Because of this, the transparency gap may also be defined as the divergence between the two approaches to openness. A number of the legal criteria have been criticized by authors from the computer science community for being too ambitious, technically impossible, or ineffectual, and the transparency gap is a contributing factor to this criticism. The naiveté here is to think that the law's lofty transparency standards are just technical directives that can be satisfied by making the right design choices.

2 Anil Kumar, Artificial Intelligence Enabled Management 154-167 (De Gruyter 2024).

However, instead of directly equating algorithmic explainability with legal transparency, it is more helpful to examine which design choices help achieve and which ones impede the desired level of openness. This kind of discrepancy has been around for a while. Law, according to system theory, is intellectually open but normatively closed, much like any other social organization. For example, this means that the scientific understanding of "causation" as a descriptive term is "normatively constrained"; in other words, not all scientific causations are also legal ones, thus scientific explanations cannot be used as legal justifications.3

2.2 Addressing Concerns About Opaque AI Systems before XAI

It is necessary to know the origins of the present worries about opaque AI systems before addressing XAI. The foundation of the first AI systems was symbolic logic, which involves the representation of information about the world using mathematical symbols. Expert systems, the first AI to be commercially viable, were developed using this paradigm. Because they are structured like a causal chain of reasoning, these representations of information are inherently interpretable and, hence, understandable by humans. On the other hand, the current dominant deep learning models do not rely on such concrete data representations. Many millions of parameters make up these models, which are often built on neural networks. These parameters are used to mathematically convert the input data into an output prediction or classification; learning their values requires massive quantities of data.

The inherent lack of interpretability in this family of techniques is a result of their built-in parameterization and abstraction. The public is increasingly worried about the social, economic, ethical, and political impacts of these models, despite their good performance. This is caused, in part, by a lack of autonomy and a socio-technical blindness that is worsened by ambiguous scientific public discourse. The area of XAI was born out of the need to explain AI models in response to the growing mistrust of these systems caused by a lack of transparency. The majority of literature reviews characterize XAI as a set of practices that, when applied to AI models, provide results that are both understandable and logical. To further assist regulators in defining XAI, we also provide a viable lower-level description based on four features. 3 Mark O.

Riedl, 'Expanding Explainability: Towards Social Transparency in AI systems' (2021) 21 JSTOR 96- 102.

The outputs produced by various AI systems are optimally suited for explanations to varying degrees due to their intrinsic features. Interactions among interpretability, explainability, justification, and transparency have sparked heated controversy in XAI. These concepts are not synonymous, and many people have different ideas about what AI systems are responsible for and what they can do. 4 Therefore, it's vital to define them clearly. Even without an explanation from the AI, a human user with average intelligence may comprehend the system's output and how it was obtained from the input. A basic linear regression function that uses historical data to forecast a country's population number is interpretable because any user with a reasonable level of expertise can see how the data influences the result.

On the other hand, explainability is a quality of AI systems that automatically generates output in the grammatical form of an explanation. Put simply, it's the capacity of an AI system to articulately explain the logic that behind its decisions. Displaying model weights to an uninitiated individual is likely to be incomprehensible since a linear regressor is inherently unexplainable. On the other hand, when coupled with the right XAI method, a linear regressor may explain its decision-making process by communicating the relevant weights in an understandable way. Rather than providing a mechanical explanation, a justification offers a teleological one. It aims to explain why a choice is beneficial, but it does not necessary seek to describe the actual decision making process, according to Miller.

A self-driving vehicle may state that stopping for a pedestrian is the "lawfully correct and safe action" without providing any details about the processing that led to that conclusion. Finally, transparency and interpretability are synonymous when considering just the algorithmic features of an AI system. But the Act's top-down approach to transparency and other more expansive conceptions of openness make this reading overly narrow.

3. ASSESSING GAP AND EXPLORING XAI AND THE GDPR'S 'RIGHT TO

EXPLANATION

While XAI does provide algorithmic transparency, the emphasis on technical details in its methods ignores the need for a stakeholder-directed strategy that prioritizes trustworthiness, as we saw earlier. The transparency gap is therefore revisited. When it comes to the law, openness

4 Ibid.

is usually seen as a way to get to bigger picture objectives, including algorithmic accountability, but not as a guarantee of reliability. An accountable system enables the user to assign responsibility properly in the event that the result is inaccurate, in contrast to a trustworthy system that provides good reasons for the user to accept the output as correct. When it comes to accountability, legislation often have a dual purpose. Not only do they work to prevent bad things from happening, but they also figure out who's to blame when bad things do happen. Though various definitions of openness may be necessary to achieve each goal, transparency is essential to achieving both.

In this setting, the transparency gap encompasses both the intradisciplinary friction that XAI experiences when trying to determine the purpose of its transparent system development and the inter-disciplinary means-end friction that we saw in the introduction. 5 Understanding the transparency standards in the AI Act is crucial for closing the transparency gap; to do so, it is helpful to place these requirements in their historical perspective. The GDPR, a watershed piece of EU data protection legislation, was the first to test out a "right to explanation" for automated judgments. One way to understand the breadth and depth of this provision and its relationship to the transparency gap is to trace its origins back to the GDPR, when the idea of explainability was first introduced.

By comparing and contrasting the two sets of laws, we may better grasp the similarities and contrasts between them. A data subject has the right to "express his or her point of view and to contest the decision" that

is "based solely on automated processing" and to obtain "meaningful information about the logic involved" in the processing of personal data. This "right to explanation" can be derived from Article 22(3) and Articles 13–15 GDPR. The explanation obligation is also made clear in the non-binding Recital 71. According to the "GDPR establishes multiple layers of transparency," the "kind of individualized transparency it requires" is directly related to the "individual rights the GDPR establishes— contestation, correction, and erasure. " The "right to explanation" was further developed by a number of scholars who sought to differentiate between varying degrees of detail by merging the rights to clarity and openness. The recommendations of the Article 29 Data Protection Working Party (WP29), the previous EU data protection advisory body, provide support for many of these claims.

Among them is a need for a "right to be informed" and procedures to provide notice before automated decisions are

5 Thomas Wischmeyer, Regulating Artificial Intelligence 214-232 (Springer 2019).

made. Furthermore, the GDPR's codification of algorithmic audits and "data protection by design" approaches is the true foundation of a "right to explanation. " Another reoccurring problem with automated profiling is discrimination resulting from computer prejudice; in such circumstances, a "right to explanation" might be drawn from human rights. A "right to explanation" has been proposed, although its practicality and very existence have been questioned. The purposefully nebulous wording of the GDPR makes the interpretation of Article 22 hard, and it is contended that such a right has both practical and legal implications. The GDPR does not specify whether an explanation must be provided ex ante or ex post.

No "complex mathematical explanations about how algorithms or machinelearning work" are required; rather, what is required are "clear and comprehensive ways to deliver information to the data subject," as stated in WP29. Because of the technological challenges in providing useful explanations, the Act runs the risk of allowing low-quality automated decisions with some sort of explanation, even though the affected party won't actually be able to use this explanation to challenge the outcome. In its place, Edwards and Veale propose a number of practical ways to guarantee a "right to better decisions," such as via data protection impact assessments. 6 The reasoning for the "illusion of a legal right" is reinforced by highlighting that the focus should be on concrete methods to attain transparency, since it is a "right ̆ to contest" in Article 22(3).

Transparency, according to this point of view, is no longer an objective in and of itself but rather a tool for achieving successful contestation, and one should judge a transparency tool by how much it helps accomplish this goal. We will see that this is the position that the Act mostly maintains. The "right to explanation" has been a common justification for XAI design decisions, but researchers should keep in mind that the GDPR introduces a new "right to explanation" that is focused on algorithms, the extent of which is still up for debate, and there is currently no case law interpreting it in this way. But in a historic judgment that might lean toward the view that a "right to explanation" exists, the Court of Justice of the European Union (CJEU) has lately added fuel to the fire.

The lack of openness would persist even if this reading is confirmed, as XAI would continue to use the law to justify its own interests. "Choices are made which significantly affect us but which we 6 Praveen Thunikoi,'Explainable AI in Data Science - Enhancing Model Interpretability and Transparency' (2021) 1(1) AJAIS 1-8.

do not understand," and regardless of whether robots are involved, there is frequently a moral and/or legal right to an explanation. Since this right would apply in any situation where a human or machine made a decision, we could say that replacing a human with an AI shouldn't compromise this right since the AI "inherits" the right to an explanation from the human. We also need to think about how to make AI transparent for all the people who have a stake in it. There is a feedback loop between the stakeholders (developers, users, deployers, and impacted parties) and the explanation process, according to users. Their classification is in line with the Act's definitions, which take comparable stakeholder groups into account. Article 3 of the Act defines terms such as "provider," "user," "importer," and others in a legal sense.

Moreover, a number of academics have proposed three separate XAI end-user categories: data experts, AI experts, and AI novices. 7 The first group is crucial because it includes end-users who have very little knowledge of the system's inner workings. Regulators are tackling problems like prejudice, privacy, and confidence in the Act, and they hope that more openness will help with them. Common methods for easing the minds of those new to artificial intelligence include cognitive modeling, conversational agents, multi-modal explanations, and natural language communication. Additional multidisciplinary research focusing on stakeholders is desperately required, while there has been a lot of theoretical and practical success in constructing social XAI that addresses these issues.

4. CONCLUSION AND SUGGESTIONS

We discovered that the transparency gap results from significant disagreements between technological norms and legal requirements for what constitutes openness. XAI employs a specialized lexicon to tackle the unique and specific problem of algorithmic transparency as an objective in and of itself. On the other hand, regulations take a broader perspective on openness, seeing it as just one of many ways to uphold broader ideals. While courts and regulators may be unable to evaluate algorithmic transparency in the absence of a shared conceptual definition of the term, expert discourse (such as those found in XAI literature) would be unable to cover all societal concerns. Indeed, these governing bodies of law are entrusted with the responsibility of ascertaining suitable interpretations of the normative requirements imposed by supreme

7 Ibid.

laws, with the help of expert deliberations. The Act and XAI run the danger of failing to have a positive effect due to the transparency gap, which is caused by the fact that computer science and law have different perspectives on openness. One possible outcome would be for the technical discourse to supersede the sociolegal one, which would mean that the broader legal idea of transparency would be equated with the far more limited computer science term. As a result, many broader damages to openness would go unchecked. Another risk is that lawmakers will put too much emphasis on XAI's potential, leading to regulations that are too complex and impossible to implement. This will have the unintended consequence of helping users and impacted third parties, not because lawmakers fail to generate transparency, but because the legal framework does not equip them to use what they've learned.

In the one case, we have unrealistically low expectations of the rule of law; in the second, we have unrealistically high expectations of the capabilities of computers. To avoid the colonization of one system by the other and instead ask XAI to learn how the law works internally so that it can create new tools and methods to promote transparency while remaining faithful to its roots, rather than asking it to "solve" the problem of legal transparency. Suggestions- • Transparency, interpretability, and explainability are three fundamental concepts that should be specified and scoped in order to enhance legal certainty and guide the development of XAI systems. • Methods that conform to the legal definition of an AI system are often the foundation of XAI approaches.

Laws may fail to take XAI tools and AI systems into account as a unified whole, leading to confusion about transparency standards and their effects. • Since the legislation does not specify how to verify algorithmic transparency, it is unclear how XAI systems may be made transparently certified while yet allowing providers to self-regulate. • Strict standards for data control and quality are outlined in the legislation. Data openness is something that XAI has failed to address so far, but it should expand to include how dataset characteristics impact AI performance.

REFERENCES

Julija Sardelic, ‘Artificial Intelligence and the ́ Right to Explanation as a Human Right’ (2021) 25(2) IEEE Internet Computing 116–120. • Anil Kumar, Artificial Intelligence Enabled Management 154-167 (De Gruyter 2024). • Mark O. Riedl, 'Expanding Explainability: Towards Social Transparency in AI systems'

(2021) 21 JSTOR 96-102.

Thomas Wischmeyer, Regulating Artificial Intelligence 214-232 (Springer 2019). • Praveen Thunikoi,'Explainable AI in Data Science - Enhancing Model Interpretability and Transparency' (2021) 1(1) AJAIS 1-8.