Civil Law And UAE: Epistemic Conflict Between Human And Machine Legal Reasoning

1. Introduction

Epistemic conflict between human and machine legal reasoning refers to situations in which:

a human judge reaches a legal conclusion using legal interpretation and evidentiary judgment;

an AI or algorithm produces a different legal conclusion;

an automated system evaluates evidence differently from a human decision-maker; or

lawyers rely upon machine-generated legal analysis that conflicts with authoritative law, evidence, or professional judgment.

The issue is increasingly important in the UAE because legal systems are becoming more digitally integrated through:

artificial intelligence;

automated document analysis;

predictive analytics;

electronic evidence;

digital courts;

machine translation;

legal research systems;

AI-assisted drafting;

algorithmic risk assessment;

automated compliance systems.

The fundamental legal question is:

Can machine-generated legal reasoning replace the legally accountable reasoning of a human judge, lawyer, arbitrator, expert, or decision-maker?

Under the present UAE legal framework, the answer is generally structured around human legal responsibility.

AI may assist the legal process, but the final legal responsibility remains attached to the legally authorized human or institution.

This is particularly important because the UAE's civil-law system requires courts to determine facts, evaluate evidence, interpret legal provisions, and give legally sufficient reasons.

2. Meaning of Epistemic Conflict

“Epistemic” concerns knowledge and justified belief.

Therefore, an epistemic conflict arises when two systems produce competing conclusions about what should be accepted as legally or factually true.

For example:

Human reasoning

A judge examines:

witness testimony;

expert reports;

contracts;

surrounding circumstances;

statutory provisions.

The judge concludes:

Contract A was validly terminated.

Machine reasoning

An AI system examines millions of documents and predicts:

Contract A remained effective.

The conflict is not merely technological.

It creates a legal question:

Which reasoning process has legal authority?

Under current UAE law, the answer is not determined by which system is statistically more accurate.

Legal authority comes from:

constitutionally and statutorily recognized institutions;

applicable legislation;

judicial authority;

procedural rules;

admissible evidence;

legally accountable decision-makers.

3. Machine Accuracy Is Not the Same as Legal Authority

An AI system may be extremely good at:

finding cases;

identifying patterns;

summarizing contracts;

detecting inconsistencies;

predicting outcomes.

But statistical accuracy does not itself confer legal authority.

A machine might predict that:

“80% of similar cases resulted in compensation.”

That prediction does not establish that compensation is legally due in the present case.

The judge must still determine:

applicable law;

relevant facts;

evidentiary reliability;

legal causation;

applicable exceptions;

procedural fairness.

Thus:

Prediction ≠ adjudication.

4. Human Legal Reasoning Under UAE Civil Law

Traditional UAE judicial reasoning involves several stages:

Stage 1 — Identify the applicable law

The court determines which statutory provision governs the dispute.

Stage 2 — Establish facts

The court evaluates documentary, testimonial and expert evidence.

Stage 3 — Interpret the law

The court determines the legal meaning of the applicable provisions.

Stage 4 — Apply law to facts

The court connects established facts with legal rules.

Stage 5 — Give reasons

The judgment should explain the decisive reasoning sufficiently for the parties and appellate court to understand the basis of the result.

This final element becomes especially important when AI systems are used.

5. Why Machine Reasoning Creates an Epistemic Problem

AI systems may produce conclusions through:

statistical patterns;

embeddings;

probabilistic inference;

machine learning;

large-language-model generation;

ranking algorithms.

The system may produce a correct-looking answer without providing the kind of legally sufficient reasoning expected from a judicial decision.

This creates a fundamental distinction:

Machine output

“The probability of liability is 78%.”

Judicial reasoning

“Article X applies because facts A, B and C were established by evidence D; the defendant's argument E is rejected because F; therefore liability follows.”

The second form performs a legal institutional function.

6. Current UAE Legal Position

There is currently no comprehensive UAE civil-law doctrine providing that AI-generated legal reasoning is independently authoritative.

Instead, existing principles concerning:

judicial reasoning;

evidence;

expert evidence;

professional responsibility;

due process;

legal interpretation;

procedural fairness

provide the framework for evaluating AI-assisted legal reasoning.

This means that AI is best understood as a decision-support technology, unless legislation or procedural rules expressly assign a particular automated system a legally binding function.

7. Human Responsibility for Machine-Assisted Decisions

A fundamental principle is:

Delegating analytical work to a machine does not necessarily delegate legal responsibility.

For example, a lawyer who submits an AI-generated legal argument remains professionally responsible for the filing.

Similarly, a judge who uses technology to organize evidence remains responsible for the judicial determination.

An arbitrator cannot necessarily avoid responsibility by stating:

“The algorithm reached this conclusion.”

The legal authority comes from the tribunal, not from the software.

8. Case Law 1 — Arabyads Holding Limited v Gulrez Alam Marghoob Alam [2025] ADGMCFI 0032

This is one of the most directly relevant UAE-region authorities concerning AI and legal reasoning.

Facts

In this ADGM proceeding, lawyers filed a substantial defence containing numerous authorities.

The court found that a significant number of the authorities were fictitious, inaccurate, or miscited and concluded that they appeared to have been generated using AI.

The court imposed substantial wasted-cost consequences.

Principle

The important principle is not that AI itself is impermissible.

The important principle is:

Human legal professionals remain responsible for verifying material generated or assisted by AI.

AI does not eliminate professional duties of:

accuracy;

verification;

competence;

candour;

responsible citation.

Epistemic significance

This case demonstrates a fundamental conflict:

Machine-generated information may appear authoritative without actually being legally authoritative.

A generated case citation can look convincing while having no legal existence.

Therefore:

AI output → verification → human professional judgment → legal submission

rather than:

AI output → automatic acceptance.

9. Case Law 2 — Oheo Bank v Parker [2025] DIFC CA 006

This case is highly relevant to the relationship between machine-like reasoning and human judicial reasoning.

The DIFC Court of Appeal considered:

procedural fairness;

opportunity to present a case;

scope of arbitration;

adequacy of judicial reasons.

The Court emphasized that reasons are an essential part of the judicial process and that parties should understand why they won or lost. It also explained that adequate reasoning facilitates meaningful appellate review.

Epistemic significance

This provides an important benchmark for AI-assisted legal decision-making.

An automated system may say:

“Claim rejected.”

But a legally responsible decision-maker may need to explain:

what issue was decided;

what evidence was accepted;

what evidence was rejected;

what legal rule applied;

why the rule applied;

how the conclusion followed.

Thus, explainability is not merely a technical preference; it can be connected to procedural fairness.

The case also emphasized that serious procedural unfairness can justify judicial intervention in arbitration.

10. Case Law 3 — Khaled Salem Musabeh Humad Al Mheiri v John Cameron [2025] DIFC CA 008

The DIFC Court of Appeal's September 2026 decision is particularly significant for human legal reasoning.

The appeal concerned:

findings of fact;

hearsay evidence;

UAE-law principles;

adequacy of judicial reasons;

the reasoning process supporting legal conclusions.

The Court required clearer articulation of the facts and the legal reasoning connecting those facts to the UAE-law conclusions.

Epistemic significance

The case illustrates that a legal conclusion cannot be separated from its reasoning chain.

This has direct relevance to AI.

An AI system may generate:

“The indemnity should be invalidated.”

But a legally sufficient decision must identify:

Facts → legal rule → interpretation → application → conclusion.

The Court's emphasis on identifying the processes of legal reasoning demonstrates why a black-box output cannot automatically substitute for legally accountable reasoning.

11. Case Law 4 — Taaleem PJSC v National Bonds Corporation PJSC and Deyaar Development PJSC [2010] DIFC CFI 014

This case is important for understanding the treatment of electronic evidence.

The dispute involved extensive commercial documentation and disclosure issues.

The DIFC Court's disclosure jurisprudence emphasizes that a party cannot simply assert that relevant documents do not exist without explaining the search undertaken.

The case became an important authority for electronic-document disclosure and the adequacy of searches.

Epistemic significance

AI systems increasingly search:

emails;

databases;

contracts;

metadata;

messages;

electronic files.

But the existence of a machine search does not automatically prove that the search was:

complete;

properly designed;

relevant;

proportionate.

Human lawyers and courts must still evaluate:

What was searched?

How was it searched?

What was excluded?

Could relevant information have been missed?

Therefore:

Machine search ≠ complete knowledge.

12. Case Law 5 — Anoop Kumar Lal & Paul Patrick Hennessy v Donna Benton [2021] DIFC CFI 005

This case involved electronic communications and the adequacy of searches for documents including:

emails;

attachments;

private email accounts;

WhatsApp communications.

The court required further search-related information.

Principle

A party's statement that no further documents exist may be insufficient if the party cannot adequately explain the search undertaken.

Epistemic significance

This case illustrates the difference between:

Machine result

“No documents found.”

and

Legally reliable conclusion

“The relevant systems, accounts, devices, date ranges and search parameters were identified and reasonably searched, and no additional responsive documents were found.”

AI can perform the search.

But humans remain responsible for determining whether the search was legally sufficient.

13. Case Law 6 — International Electro-Mechanical Services Co. LLC v Emirates Speciality Hospital FZ-LLC [2020] DIFC CFI 114

This construction dispute involved technical evidence and UAE-law principles concerning engineering certification.

The court considered expert evidence concerning the role of an engineer in certifying works and payment claims.

Principle

Technical expertise may assist the court in determining:

quantity;

quality;

value;

compliance with contractual requirements.

But the ultimate legal determination remains for the court.

Epistemic significance

This is an important analogy for AI.

An AI system may be highly capable of analysing:

construction drawings;

technical specifications;

payment certificates;

project records.

But:

technical analysis ≠ legal judgment.

The machine may assist with factual or technical analysis, while the judge determines the legal consequences.

14. Case Law 7 — Dubai Court of Cassation Commercial No. 767/2021

This authority concerns the proper role of experts.

The UAE judicial approach distinguishes between:

Technical/factual questions

and

Ultimate legal questions.

Experts may assist courts in understanding complex technical facts.

However, an expert cannot ordinarily replace the judge's legal function.

Epistemic significance

This distinction is extremely important for machine reasoning.

AI can increasingly perform expert-like functions:

calculate;

classify;

identify patterns;

compare documents;

predict;

detect anomalies.

But an AI system should not automatically become the legal decision-maker merely because its analytical performance is technically sophisticated.

15. Case Law 8 — GFH Capital Ltd v Haigh [2014] DIFC CFI 020

This commercial dispute involved extensive documentary and evidentiary analysis.

The case illustrates the importance of judicial evaluation of:

documentary evidence;

contractual arrangements;

witness evidence;

surrounding circumstances.

Epistemic significance

Legal truth is not necessarily determined by the largest quantity of data.

A machine may process:

10 million documents.

A court may ultimately rely upon:

10 documents that directly establish the decisive fact.

This illustrates an important difference between:

data volume

and

legal relevance.

16. Case-Law Matrix

The DIFC authorities are persuasive/analogical for UAE civil-law analysis and should not automatically be treated as binding on onshore UAE courts.

17. Human Reasoning Versus Machine Reasoning

Human Legal Reasoning

Human legal reasoning generally involves:

interpretation;

contextual judgment;

normative assessment;

evidentiary evaluation;

procedural fairness;

credibility assessment;

proportionality;

legal precedent;

institutional responsibility.

Machine Legal Reasoning

Machine reasoning may involve:

pattern recognition;

probability;

statistical correlation;

semantic similarity;

document classification;

prediction;

retrieval;

automated comparison.

The two forms of reasoning therefore operate differently.

18. The Problem of Probabilistic Truth

AI frequently operates through probabilities.

For example:

“There is a 92% probability that this document contains a contractual breach.”

But civil adjudication generally requires a legally structured finding.

The court must determine:

Did the defendant actually breach the relevant obligation?

Probability can assist the court.

It does not necessarily replace the legal standard governing proof.

19. The Problem of Hallucinated Law

One of the greatest risks is AI hallucination.

A machine can generate:

nonexistent cases;

incorrect article numbers;

imaginary quotations;

outdated legislation;

incorrect jurisdictions;

fabricated legal propositions.

The Arabyads case is particularly significant because it demonstrates that apparently authoritative machine-generated legal material can contain nonexistent or inaccurate authorities.

The lesson is:

Legal research requires source verification, not merely fluent output.

20. The Problem of Machine Bias

AI systems may reproduce biases contained in:

training data;

historical decisions;

human classifications;

incomplete datasets;

institutional practices.

Suppose an algorithm learns:

Previous courts frequently rejected claim type X.

It may recommend rejection of future claims.

But the present case may involve:

new legislation;

different facts;

changed social circumstances;

a new precedent.

Human judicial reasoning must therefore remain capable of recognizing legal change.

21. Precedent Versus Pattern Recognition

This is one of the most important epistemic differences.

Human legal method

“The earlier case established principle X because of reasoning Y.”

Machine pattern recognition

“Cases containing characteristics A, B and C frequently resulted in outcome X.”

These are not identical.

A precedent has legal meaning because of:

authority;

jurisdiction;

reasoning;

factual context;

statutory framework.

Statistical frequency does not itself create precedent.

22. Machine Reasoning and Statutory Interpretation

AI may compare thousands of versions of legislation.

This is useful.

But statutory interpretation requires identifying:

applicable legislation;

commencement date;

transitional provisions;

legislative hierarchy;

statutory definitions;

exceptions;

purpose where legally relevant.

For example, after the UAE Civil Transactions Law changed in 2026, an AI trained primarily on the previous law might incorrectly provide the former article number.

Therefore:

Historical legal accuracy ≠ current legal accuracy.

23. Temporal Epistemic Conflict

Legal systems change.

An AI may answer based upon:

1985 Civil Transactions Law

while the current dispute is governed by:

2025 Civil Transactions Law effective 1 June 2026.

The machine may therefore produce a historically correct but presently incorrect answer.

This is one of the most important risks of AI legal research.

24. Human-in-the-Loop Model

A safer legal architecture is:

AI

↓

Human verification

↓

Legal analysis

↓

Human decision

↓

Reasoned legal output

This model allows AI to perform:

searching;

summarization;

comparison;

document review;

chronology;

anomaly detection.

But the human remains responsible for:

legal interpretation;

factual acceptance;

professional advice;

judicial decision;

final submission.

25. Human-on-the-Loop Model

A more automated model is:

AI makes recommendation

↓

Human reviews recommendation

↓

Human accepts/rejects

This may be appropriate for:

document prioritization;

case classification;

risk alerts;

compliance monitoring.

But the required level of human review should increase with the legal consequences of the decision.

26. Fully Automated Legal Decisions

Fully automated civil adjudication creates much greater legal difficulty.

If:

AI → evidence analysis → legal rule → decision → enforcement

occurs without meaningful human intervention, important questions arise:

Who is legally responsible?

Can the decision be appealed?

How can the decision be explained?

Can the parties challenge the algorithm?

Who audits the training data?

What happens when the system is wrong?

Who bears liability for the error?

The present UAE framework does not generally transform AI into an autonomous judicial person.

27. AI Cannot Automatically Become a Legal Person

An AI system is not automatically:

a judge;

arbitrator;

lawyer;

expert;

legal person;

holder of judicial authority.

Legal authority must come from legislation and institutional authorization.

Therefore:

Machine intelligence does not automatically create legal personality or adjudicatory authority.

28. Epistemic Conflict in Arbitration

The issue is particularly significant in arbitration.

Imagine an arbitral tribunal uses AI to analyse:

thousands of documents;

witness statements;

damages calculations.

The tribunal remains responsible for:

due process;

equal treatment;

opportunity to present the case;

evidentiary evaluation;

legal reasoning;

final award.

The Oheo Bank case demonstrates the importance of meaningful opportunity to present a case and adequate reasons in the DIFC arbitration context.

Therefore, an AI system should not silently introduce a decisive factual or legal theory without giving affected parties an appropriate opportunity to address it.

29. AI and Procedural Fairness

Suppose an AI system identifies:

“This contract is probably fraudulent.”

If the tribunal relies upon that output without giving the parties an opportunity to challenge:

the dataset;

methodology;

assumptions;

source documents;

error rate,

a procedural fairness problem may arise.

The legal issue is not simply whether the AI is accurate.

It is whether the parties received a fair opportunity to contest the basis of the decision.

30. AI and Judicial Reasons

A machine-generated explanation may say:

“Based on the relevant factors, liability is established.”

That may be insufficient.

A legally meaningful judgment normally requires identification of:

material facts;

applicable legal rules;

decisive evidence;

arguments accepted/rejected;

reasoning connecting law and facts.

The DIFC Court of Appeal's reasoning in Oheo Bank strongly illustrates the institutional importance of adequate reasons.

31. AI and Expert Evidence

AI-generated analysis can resemble expert evidence.

For example:

AI engineering model;

AI valuation;

AI forensic accounting;

AI medical analysis.

But the court may need to know:

who designed the model;

what data it used;

what assumptions it contains;

error rates;

validation methodology;

whether the result can be independently reproduced.

An unexplained algorithm should not automatically receive greater evidentiary weight merely because it is computational.

32. Explainability

Explainability means that a person can understand:

what information was considered;

what reasoning process was used;

what assumptions were made;

why the output was generated.

For legal systems, explainability is particularly important because parties need to challenge decisions.

A black-box result creates an epistemic asymmetry:

Machine knows how it calculated the result → parties do not.

That can make meaningful challenge difficult.

33. Transparency Does Not Mean Full Disclosure of Source Code

A legal system may not always require disclosure of every line of software code.

The relevant question is more practical:

What information is necessary for the parties and court to understand and challenge the basis of the machine-assisted conclusion?

Depending on the case, this might include:

methodology;

data sources;

search terms;

parameters;

confidence levels;

validation procedures;

human review;

limitations.

34. AI and the Standard of Proof

AI may produce numerical confidence:

0.91 probability.

But legal standards are not always reducible to a numerical probability.

The court must apply the applicable evidentiary standard.

A probability score should therefore ordinarily be treated as evidence or analytical assistance, rather than as the legal standard itself.

35. AI and Evidence Authentication

Digital evidence can include:

emails;

metadata;

blockchain transactions;

server logs;

AI-generated records;

system outputs.

The fact that a computer produced a record does not automatically establish:

authenticity;

accuracy;

completeness;

reliability.

Human examination remains necessary.

The electronic-disclosure principles illustrated by Taaleem and Anoop Kumar Lal are therefore relevant to AI-generated or AI-processed evidence.

36. Machine Discovery and Human Interpretation

AI can discover:

“Clause 14 appears in 4,000 contracts.”

But a lawyer must still determine:

“Does Clause 14 create the legal obligation relevant to this dispute?”

Similarly:

AI:

“These documents are semantically similar.”

Lawyer:

“Are they legally relevant?”

Thus:

Discovery ≠ interpretation.

37. Epistemic Conflict in Contract Law

Suppose an AI system interprets a contractual clause as creating an indemnity.

The human lawyer interprets it as a limitation of liability.

The court must determine the legal meaning.

AI's interpretation may be persuasive as an analytical tool, but it does not replace:

contractual interpretation;

governing law;

evidence;

judicial reasoning.

38. Epistemic Conflict in Tort Law

AI may identify statistical relationships between:

conduct;

accidents;

injuries.

But tort liability requires legally relevant analysis of:

duty;

breach;

causation;

damage;

defenses.

Machine correlation is therefore not equivalent to legal causation.

39. Epistemic Conflict in Unjust Enrichment

AI may identify:

AED 10 million moved from A to B.

But it cannot automatically establish:

B was unjustly enriched.

The legal analysis still requires:

enrichment;

corresponding impoverishment;

absence of legal basis;

applicable restitution rules.

This illustrates the difference between transaction detection and legal characterization.

40. Epistemic Conflict in Environmental Liability

AI may process:

satellite images;

pollution readings;

weather data;

emission records.

It may conclude:

“Factory A probably contributed to the pollution.”

The court must still determine:

legal duty;

scientific causation;

legal causation;

contribution of other actors;

statutory responsibility.

Therefore:

Environmental prediction ≠ environmental legal attribution.

41. Machine Error and Civil Responsibility

If a lawyer relies on incorrect AI output and suffers loss, possible legal issues may include:

professional negligence;

breach of contract;

confidentiality;

data protection;

misrepresentation;

failure of professional standards.

If an enterprise deploys AI and causes damage, the court may examine:

who selected the system;

who configured it;

who supervised it;

whether risks were foreseeable;

whether safeguards existed.

The machine itself does not necessarily become the legal defendant.

42. AI Vendor Versus User

Consider:

AI developer → AI platform → law firm → client

If the AI produces incorrect legal analysis, several legal relationships may exist.

Developer

May have contractual obligations concerning software.

Platform provider

May have service obligations.

Law firm

May owe professional duties to the client.

Client

May suffer the ultimate loss.

The court must determine the applicable contracts and duties rather than simply assigning responsibility to “AI.”

43. Professional Responsibility

The Arabyads decision provides a particularly important lesson.

A lawyer cannot necessarily defend inaccurate legal submissions by saying:

“The AI generated them.”

The professional remains responsible for:

checking authorities;

verifying quotations;

confirming legislation;

checking jurisdiction;

confirming whether a case actually exists.

Thus:

AI assistance does not erase professional responsibility.

44. AI and Legal Research

AI legal research should ideally follow:

Step 1

Identify the jurisdiction.

Step 2

Identify the current legislation.

Step 3

Check commencement dates.

Step 4

Identify relevant cases.

Step 5

Verify case existence.

Step 6

Read the actual judgment.

Step 7

Identify ratio and factual context.

Step 8

Check whether the case remains good law.

Step 9

Apply the principle to the facts.

Step 10

Have a human lawyer verify the final proposition.

45. AI and the 2026 UAE Civil Transactions Law

The change from the former 1985 Civil Transactions Law to the 2025 Civil Transactions Law effective 1 June 2026 creates a major temporal-risk problem.

An AI system may retrieve:

Article 318 of the former law.

But a current dispute may require:

Article 274 of the new law.

Similarly, decennial liability provisions moved from the former Articles 880–883 to the new Articles 821–824.

This illustrates a fundamental epistemic danger:

A machine can retrieve legally authentic historical information that is nevertheless wrong for the present dispute.

Human legal verification must therefore include temporal validation.

46. Epistemic Hierarchy in UAE Legal Reasoning

A useful hierarchy is:

Level 1 — Constitution and legislation

Binding legal authority.

Level 2 — Applicable judicial authority

Relevant judicial interpretation.

Level 3 — Admissible evidence

Facts established in the proceedings.

Level 4 — Expert analysis

Technical assistance.

Level 5 — AI analysis

Research and analytical assistance unless otherwise legally authorized.

Level 6 — Statistical prediction

Supporting information, not automatically legal authority.

This hierarchy explains why an AI's confidence score cannot simply override legislation or a judicial finding.

47. Human-Machine Complementarity

The issue should not be framed only as:

Human versus machine.

A more useful model is:

Human legal reasoning + machine analytical capacity.

AI is particularly useful for:

large-scale document review;

chronology;

citation discovery;

contract comparison;

anomaly detection;

translation;

data extraction.

Humans remain particularly important for:

interpretation;

accountability;

procedural fairness;

credibility;

normative judgment;

legal responsibility.

48. The Principle of Human Legal Accountability

A useful emerging principle is:

The entity legally authorized to decide remains responsible for the decision even when computational systems assist the decision-making process.

This principle preserves:

accountability;

appeal;

procedural fairness;

professional responsibility;

institutional legitimacy.

49. Can a Judge Blindly Adopt AI Reasoning?

The existing legal principles concerning judicial reasoning indicate significant difficulty with such an approach.

A judgment should not merely say:

“AI determined that the claimant was liable.”

The court must independently establish the legal basis of its decision.

AI may help the judge:

organize evidence;

identify authorities;

detect inconsistencies.

But the judgment must remain a judicial act.

50. Can a Lawyer Blindly Adopt AI Research?

The Arabyads authority provides a strong warning against this.

A lawyer should verify:

cases;

quotations;

statutory provisions;

citations;

legal propositions.

AI-generated legal research is therefore a drafting/research input, not a substitute for professional verification.

51. Can an Arbitrator Use AI?

There is no general principle that an arbitrator can never use technology.

But significant concerns arise where AI affects:

evidence evaluation;

credibility;

legal interpretation;

outcome;

procedural directions.

The parties may need appropriate transparency concerning material use of AI, especially where the use could affect their ability to present or challenge their case.

The principles emphasized in Oheo Bank v Parker concerning opportunity to present a case and adequate reasons are highly relevant by analogy.

52. AI and the Right to Challenge

An epistemically fair system should allow a party to challenge:

data;

assumptions;

methodology;

legal sources;

machine-generated conclusions.

Without meaningful challenge, the machine becomes an opaque authority.

The judicial system, however, is structured around contestability.

53. AI and Judicial Independence

Another concern is whether judges could become overly dependent on:

predictive systems;

automated recommendations;

algorithmic rankings.

Judicial independence requires that the decision remain the judge's own legally accountable determination.

AI may inform.

It should not silently determine.

54. AI and Institutional Legitimacy

Courts derive legitimacy from:

law;

jurisdiction;

procedural fairness;

reasoned decisions;

appeal;

accountability.

An AI system does not automatically possess these institutional characteristics.

Consequently:

computational intelligence ≠ judicial legitimacy.

55. AI and the Doctrine of Reasoned Decision-Making

The recent DIFC appellate cases are particularly valuable.

In Oheo Bank, the Court stressed that adequate reasons are an essential component of due process and meaningful appellate review.

In Al Mheiri v Cameron, the Court similarly focused upon whether the factual findings and UAE-law reasoning sufficiently supported the conclusions reached.

Together, these principles provide an important framework:

AI output → human evaluation → reasoned legal decision → reviewable judgment.

56. The “Black Box” Problem

A black-box AI system may produce:

“High risk.”

But the user may not know:

why;

based upon what data;

which factors mattered;

whether historical bias influenced the result;

whether the result is reproducible.

In legal proceedings, such opacity can create difficulties concerning:

evidence;

fairness;

reasons;

appeal;

accountability.

57. AI and Legal Epistemology

Traditional legal epistemology asks:

“How does the court know that a fact is true?”

AI adds another question:

“How does the machine know?”

The machine's answer may be statistical rather than legal.

For example:

Human: witness credibility + documentary corroboration.

Machine: similarity with prior cases = 0.87.

The two forms of knowledge should not automatically be treated as equivalent.

58. The Problem of Explainability Versus Accuracy

An AI system may be:

Highly accurate but opaque

or

Less accurate but explainable.

Legal institutions cannot necessarily select systems based solely on predictive accuracy.

A legally useful system must also permit sufficient understanding and challenge where its output materially affects rights.

59. Human Oversight Model

For UAE civil-law systems, a practical model is:

Tier 1 — Low-risk AI

Examples:

spelling;

translation;

formatting.

Minimal legal concern.

Tier 2 — Analytical AI

Examples:

document classification;

contract comparison;

chronology.

Human verification required.

Tier 3 — Substantive legal AI

Examples:

legal research;

liability analysis;

damages calculation.

Strong human verification required.

Tier 4 — Decision-influencing AI

Examples:

case outcome prediction;

credibility assessment;

automated legal recommendations.

Enhanced transparency and human oversight become particularly important.

Tier 5 — Autonomous adjudication

A substantially different legal framework would be required if machines were to exercise independent adjudicatory authority.

60. Remedies for Erroneous Machine Reasoning

Where AI-assisted legal reasoning produces harm, possible legal responses may include:

appeal;

setting aside;

procedural challenge;

professional disciplinary action;

contractual claim;

negligence claim;

correction of evidence;

exclusion or reduced weight of unreliable material.

The appropriate remedy depends on who used the AI and the legal context.

61. Future UAE Legal Development

Future UAE legislation and judicial practice may need to address:

disclosure of material AI use;

human oversight;

auditability;

algorithmic transparency;

legal-research verification;

AI-generated evidence;

confidentiality;

privilege;

data protection;

cybersecurity;

responsibility for AI errors;

automated judicial assistance;

AI use in arbitration;

machine-generated expert evidence.

62. Core Legal Tests for AI-Assisted Reasoning

A court or tribunal could conceptually ask:

Test 1 — Authority

Who legally made the decision?

Test 2 — Verification

Was machine output independently verified?

Test 3 — Evidence

What evidence supported the conclusion?

Test 4 — Reasoning

Can the legal reasoning be articulated?

Test 5 — Fairness

Did the parties have an opportunity to challenge the material?

Test 6 — Accountability

Who bears responsibility for the result?

Test 7 — Reliability

Was the system appropriate for the task?

Test 8 — Temporal validity

Was the legal information current and applicable to the relevant date?

63. Central Distinction: Machine Knowledge vs Legal Knowledge

The most important conceptual distinction is:

Machine knowledge

“Based upon available data, this result is statistically likely.”

Legal knowledge

“Under the applicable law, based upon established evidence and legally sufficient reasoning, this consequence follows.”

The second is not reducible to the first.

64. Overall Case-Law Synthesis

The cases establish a developing but coherent framework.

Arabyads demonstrates that humans remain responsible for verifying AI-assisted legal research.

Oheo Bank demonstrates the importance of procedural fairness and adequate reasons.

Al Mheiri v Cameron demonstrates the importance of articulating the factual and legal reasoning supporting a conclusion.

Taaleem and Anoop Kumar Lal demonstrate that electronic searches and document discovery require meaningful human assessment of search methodology.

International Electro-Mechanical Services and Dubai Cassation Commercial No. 767/2021 demonstrate the distinction between technical assistance and ultimate legal determination.

These authorities do not establish a general UAE doctrine prohibiting AI.

Instead, they provide existing legal principles through which courts can evaluate AI-assisted legal processes.

65. Case-Law Summary Table

66. Conclusion

Epistemic conflict between human and machine legal reasoning is an emerging issue in UAE civil law rather than an established independent cause of action.

The existing legal framework nevertheless provides important principles.

First, legal authority belongs to legally authorized institutions and decision-makers, not automatically to computational systems.

Second, AI-generated information requires verification. The Arabyads case demonstrates the professional risks of relying upon apparently authoritative but inaccurate AI-generated legal authorities.

Third, reasoned decision-making remains fundamental. Oheo Bank and Al Mheiri v Cameron demonstrate the importance of explaining the factual and legal pathway leading to a conclusion.

Fourth, electronic evidence and machine searches require methodological scrutiny. Taaleem and Anoop Kumar Lal demonstrate that the existence of an electronic search does not eliminate the need to establish that the search was reasonable and properly conducted.

Fifth, technical expertise does not automatically become legal judgment. The distinction between expert assistance and judicial determination provides a useful model for AI-assisted adjudication.

The emerging UAE principle can therefore be expressed as:

Machine intelligence may assist the discovery, organization, analysis and prediction of legal information, but legal responsibility remains with the human or institution legally authorized to interpret, decide, advise or adjudicate.

Accordingly, the central challenge is not simply whether machines can “reason like lawyers.” The deeper legal question is whether their reasoning can be incorporated into a system that preserves authority, evidence, transparency, procedural fairness, human accountability and reviewable legal reasoning.

For UAE civil law, the likely future model is therefore not simply human versus machine, but:

Machine analysis → human verification → legally authorized judgment → reasoned and reviewable decision.