According to Richard Susskind in his book, “How To Think About AI” it is stated that,
“Since the mid-1950s, when the term artificial intelligence was first coined, fans and critics have been equally vocal. The tension between the proponents and naysayers extends well beyond the usual conflicts about this. There is incalculably more at stake here because to put it starkly, AI will heavily shape the future. Will AI provide answers to mankind’s gravest challenges, from climate challenge to global health? We don’t have pat answers to these or many other pressing questions that AI is now prompting.”
That is where we face a gap between machine work and humans. Agentic AI has the capability of doing goal oriented work, efficiency and adaptability to the extent of working independently. This increasing capability of machines raises the question of the extent to which machines can be given freedom to work independently without human intervention.
Organizations need strategic human-in-the-loop oversight to keep AI systems transparent, accountable, and aligned with human values. Leaders face a significant choice between full automation, which runs complete workflows independently, and a system in which people’s decision-making authority remains intact while AI offers recommendations. This balance becomes especially important when regulations are evolving. The European Union’s AI Act specifically highlights human oversight requirements for high-risk AI systems. As stated in the European Union’s AI Act 2026:
“4)AI is a fast evolving family of technologies that contributes to a wide array of economic, environmental and societal benefits across the entire spectrum of industries and social activities. By improving prediction, optimising operations and resource allocation, and personalising digital solutions available for individuals and organisations, the use of AI can provide key competitive advantages to undertakings and support socially and environmentally beneficial outcomes, for example in healthcare, agriculture, food safety, education and training, media, sports, culture, infrastructure management, energy, transport and logistics, public services, security, justice, resource and energy efficiency, environmental monitoring, the conservation and restoration of biodiversity and ecosystems and climate change mitigation and adaptation.
5)At the same time, depending on the circumstances regarding its specific application, use, and level of technological development, AI may generate risks and cause harm to public interests and fundamental rights that are protected by Union law. Such harm might be material or immaterial, including physical, psychological, societal or economic harm.”
The landmark case of Ishfaq Ahmed vs. Mushtaq Ahmed (PLD 2025 SC 582) marks the first instance where the Supreme Court of Pakistan established a “constitutional threshold” for the use of Artificial Intelligence (AI) in the judiciary.
Facts and Issues of the Case
Facts: The case began as an ordinary civil dispute between two brothers over tenancy rights and property ownership in Lahore that continued for seven years. Ishfaq Ahmed (petitioner) evicted his brother, Mushtaq Ahmed (respondent), for not paying the rent. Despite clear evidence of ownership and non-payment, the case faced systemic delays across multiple court tiers.
Issues:
The civil issue was whether the respondent was a defaulting tenant liable for eviction. Whereas, the systematic issue was how to address the chronic backlogs in Pakistan’s judiciary that infringe upon the constitutional right to a fair and expeditious trial (Articles 10A and 37(d)) of The Constitution of Pakistan 1973. And lastly, the digital issue was, to what extent can AI be integrated into judicial decision-making without compromising judicial independence and human empathy?
Conclusion: Balancing Legal Validity, Recognition, and Morality
The Hierarchical Chain: In Kelsenian terms, the Supreme Court’s decision validates AI as a “subordinate tool” within the legal hierarchy. For AI to be legally valid, its use must be authorized by a higher norm—in this case, the Constitution of Pakistan (the Grundnorm) and subsequent judicial guidelines.
Imputation vs. Causality: Kelsen distinguishes between causality (natural science) and imputation (law, where “if A happens, B ought to follow”). The judgment aligns with Kelsen by ruling that AI operates on “causality” (data processing), whereas judicial acts are acts of “imputation”. Because only a human official can exercise the will to link a fact to a legal consequence, machine autonomy lacks the capacity to create a legal norm. AI’s role is therefore limited to the “cognitive” stage of legal research, while the “volitional” stage (the decision) must remain with the judge.
2. H.L.A. Hart’s Concept of Law
Hart defines in his book, “The Concept of Law”, a legal system as a union of Primary Rules (obligations) and Secondary Rules (rules about rules).
Rule of Recognition: Hart’s “Rule of Recognition” is the social fact that identifies which rules are valid. The Ishfaq Ahmed case serves as a new Secondary Rule of Adjudication. It modifies the Rule of Recognition in Pakistan by establishing that for a judgment to be recognized as legally valid, it must be the product of human reasoning, even if assisted by technology.
Open Texture of Law: Hart famously argued that laws have an “open texture,” requiring judges to exercise discretion in “hard cases” where rules are vague. The Supreme Court used this logic to argue that AI, which relies on past data, cannot navigate the “open texture” of future social and moral complexities. While AI can handle the “core” of settled rules, the discretionary power to resolve penumbral (uncertain) cases is a human-centric function that defines the legal system’s “internal aspect”; the shared acceptance of law as a standard for conduct.
The court explicitly held that AI must “support, not supplant” human judges. It ruled that judicial power is a “sacred trust” that requires a human conscience, which a machine cannot possess.
It linked human oversight to Article 10A (Right to Fair Trial) of the Constitution of Pakistan. The court reasoned that a “fair trial” implies a process where a human mind considers the unique, subjective nuances of a case—something an autonomous algorithm cannot do
The judgment warned against “machine autonomy” because algorithms often operate as a “black box” without transparent reasoning. For a legal decision to be valid in Pakistan, it must be reasoned and public, which requires a human to validate the AI’s output.
Thus, it wouldn’t be wrong to quote here that “excess of everything is bad,” as it was proved in the case of Mata v. Avianca, Inc. (S.D.N.Y. 2023), that too much reliance on machines subsequent to harmful results indicates the non-reliance on machines.
Facts & Issue
In Mata v. Avianca, a plaintiff sued an airline for personal injuries, and the airline moved to dismiss the case as time-barred. The plaintiff’s attorneys filed a brief in opposition that cited several “bogus” court cases, such as Varghese v. China Southern Airlines, which had been fabricated by the AI chatbot ChatGPT. Even after the court and opposing counsel pointed out that these cases did not exist, the attorneys continued to insist they were real. The core legal issue was whether the attorneys’ reliance on AI-generated hallucinations and their subsequent failure to correct the record violated Rule 11 of the FRCP (Federal Rules of Civil Procedure) USA, warranting judicial sanctions.
Relevance to H.L.A. Hart’s Theory
Hart’s theory of legal positivism is grounded in the Rule of Recognition, which identifies valid law through social facts and established institutional criteria. In Mata, the “fake” cases failed Hart’s criteria because they lacked any institutional pedigree—they were not products of a recognized law-making body or court system. From a Hartian perspective, the attorneys’ actions constituted a breakdown of the secondary rules governing how law is identified and applied. Hart would see the sanctions as the legal system’s way of reasserting its boundaries, ensuring that only rules derived from the authoritative Rule of Recognition are treated as “law” within the community.
Relevance to Hans Kelsen’s Theory
Under Kelsen’s Pure Theory of Law, the legal system is a hierarchy of norms in which every law derives its validity from a higher norm, ultimately reaching the Grundnorm (Basic Norm). The attorneys’ submission of fake cases introduced “non-norms” into this hierarchy. According to Kelsen, a court’s decision is only valid if it is based on a pre-existing norm within the legal order. By attempting to base a judicial decision on fictitious precedents, the attorneys were asking the judge to step outside the “pure” legal science. The Rule 11 sanctions represent a “coercive act” (a sanction) prescribed by a valid legal norm to protect the hierarchy from the pollution of non-legal, fabricated “facts” masquerading as norms.
Hybrid models: intelligent escalation and collaboration
Hybrid models leverage the strengths of artificial intelligence and human decision-making to form the optimal strategy. In this “golden mean,” artificial intelligence can operate autonomously in simple activities, but also allows for the seamless inclusion of human intervention when needed
Intelligent escalation helps AI recognize when to hand off tasks to humans. Two main triggers drive this: high-value or high-risk scenarios (transactions exceeding specific thresholds) and low confidence scores (AI lacking sufficient data to make reliable decisions). Organizations must create clear escalation paths for this system to work. AI should give human agents full context, including previous interactions and analysis. The most successful implementations progress gradually instead of jumping to full autonomy. They start by increasing AI assistance before moving toward greater automation.
