Authored By: Keletso Selogate
University of South Africa
Introduction.
Artificial intelligence (AI) is drastically changing how legal systems function globally. AI is becoming a more useful tool in both the public and private sectors, from assessing forensic evidence and examining lengthy legal documents to helping medical professionals and forecasting litigation outcomes.[1] AI is starting to have an impact on judicial procedures as its capabilities grow, assisting with expert analysis, case management, and legal research.[2] A significant legal concern is brought up by these developments: could artificial intelligence systems be accepted as expert witnesses in South African courts?
To help judges resolve disputes involving complex scientific, medical, financial, or technical issues, South African courts have historically relied on expert witnesses to provide specialized knowledge.[3]Expert testimony is presumed by the law to be given by competent natural beings with independent judgement, professional experience, and the ability to be cross-examined in court.[4] But without having legal personhood, professional accountability, or the capacity to articulate their thinking in the same way as human specialists, AI systems can generate extremely accurate evaluations. This raises serious legal questions about the admissibility, dependability, accountability, and transparency of expert testimony produced by AI under the current South African judicial system.[5]
my article argues that artificial intelligence should not currently be accepted as an independent expert witness in South African courts, despite its potential to increase the effectiveness, uniformity, and accuracy of judicial decision-making. Until Parliament passes legislation expressly governing AI-generated expert testimony, AI should instead serve as a decision-support tool under the supervision of trained human experts. To bolster this claim, the article first looks at the legal framework that currently governs expert testimony in South Africa. It then analyses pertinent case law, critically assesses the evidentiary and constitutional issues raised by AI-generated opinions, compares strategies used in a few foreign jurisdictions, and suggests legislative changes to guarantee that the incorporation of AI into the legal system fosters justice, accountability, and the rule of law.
- Legal Framework Governing Expert Evidence and Artificial Intelligence in South African Courts
1.1 The Constitutional Framework
The Republic of South Africa’s 1996 Constitution serves as both the country’s ultimate legislation and the cornerstone of its legal framework.[6] Therefore, any application of artificial intelligence (AI) in legal proceedings must adhere to constitutional values, especially those concerning accountability, equality, justice, dignity, and the rule of law.[7] The Constitution’s provisions provide the legal framework that developing technologies must function within, even if it does not specifically govern AI or digital technology.
Everyone has the right to have disagreements settled in a fair public hearing before an impartial, independent court or tribunal, according to Section 34 of the Constitution.[8] According to this right, evidence must be trustworthy, testable, and fairly assessed when it is presented in court. Courts would have to decide whether litigants might successfully contest an AI system’s conclusions through cross-examination or other procedural protections if it were accepted as an independent expert witness. Concerns about procedural justice and the constitutional right to a fair trial are directly raised by the incapacity to challenge an AI system.
In the same way, accused individuals are guaranteed the opportunity to contest evidence used against them in criminal trials under section 35. Any AI-generated opinion permitted without sufficient safeguards may violate an accused person’s right to a fair trial because expert testimony commonly affects criminal verdicts.
Equally important is Section 9 of the Constitution, which ensures equality before the law. Sets of information that may include societal or historical biases are used to train AI systems. The employment of such systems may violate the constitutional principle of equality if they result in discriminatory outcomes based on socioeconomic position, gender, race, or handicap.[9] Therefore, before admitting AI-generated evidence, courts must examine its validity and fairness in accordance with constitutional principles.
1.2 Expert Evidence under South African Law
South African law has long acknowledged that when disputes demand specific expertise beyond the typical experience of judicial personnel, courts may need the help of expert witnesses. Expert witnesses support the court by providing opinions based on their training, experience, and expertise.[10] Rather than representing the party that appointed them, their main responsibility is to support the court impartially.
The common law, which is strengthened by legislation and court precedent, governs the admissibility of expert testimony. One of the most important cases is Holtzhauzen v. Roodt, in which the court determined that expert testimony is only admissible if it helps the court make a fair decision and offers information beyond the court’s ordinary understanding.[11] The ruling also underlined how crucial factors to take into account when deciding admissibility are the expert’s credentials, expertise, methodology, and objectivity.
AI systems lack professional registration and legal personhood, in contrast to human specialists. They are not allowed to take an oath, take legal responsibility for their beliefs, or face traditional cross-examination. As a result, AI-generated opinions are difficult to accept as independent expert testimony under the current laws governing expert witnesses.
1.3 Legislative Framework
Several statutes are relevant even though South Africa has not passed any laws expressly governing expert testimony produced by AI.
The Electronic Communications and Transactions Act 25 of 2002 makes it easier for digital evidence to be admitted in court by giving electronic data transmissions and electronic records legal recognition.[12] Nevertheless, the Act does not control the trustworthiness or admissibility of views produced by AI systems on their own.
There is also indirect significance to the Protection of Personal Information Act 4 of 2013 (POPIA). During training and operation, AI systems often process large amounts of personal data. POPIA imposes requirements on responsible parties who use AI technologies, requiring that personal information be processed legally, transparently, and securely.[13] Therefore, since AI-generated evidence depends on personal data, compliance with data protection laws is crucial.
1.4 The Regulatory Gap
Despite these legal and constitutional safeguards, South African law lacks a thorough framework controlling the admissibility, dependability, accountability, and transparency of expert testimony produced by artificial intelligence. Current regulations were created under the presumption that expert views come from qualified individuals who can defend their judgements and provide an explanation. This regulatory vacuum widens as AI systems are incorporated more deeply into financial investigations, forensic science, medical diagnosis, and legal practice.
Judges, lawyers, and litigants are unsure about the admissibility of AI-generated opinions due to the lack of legislation. Courts run the risk of taking conflicting stances on the admissibility of AI-assisted evidence in the absence of statutory rules governing algorithmic transparency, validation, and accountability.
II. Analysis of Case Law
2.1 Introduction
The question of whether an artificial intelligence (AI) system qualifies as an expert witness has not before been presented to South African courts. However, the body of existing expert evidence jurisprudence offers important insight on how courts are likely to handle expert opinions produced by artificial intelligence. The courts have often stressed that expert testimony must support the court, be supported by trustworthy methodology, and be provided by an unbiased expert whose logic can be examined during the adversarial process. When determining whether judgements produced by AI should be accepted as independent expert testimony, these guidelines are very important.
2.2 Holtzhauzen v Roodt
Holtzhauzen v. Roodt 1997 (4) SA 766 (W) is the leading South African authority on the admissibility of expert testimony.[14] In this instance, the court considered both the function that expert witnesses serve in supporting the court and the conditions under which expert opinion evidence may be allowed. According to the court, expert testimony is only allowed in cases where the subject matter is outside the presiding judicial officer’s normal area of expertise.[15]Additionally, the court emphasised that an expert’s opinion is not always accepted solely on the witness’s credentials. Instead, the expert must exhibit specialist knowledge, use a trustworthy methodology, and offer unbiased, well-reasoned evidence that helps the court decide the problems at hand.[16]
The subject of whether artificial intelligence (AI) systems should be accepted as expert witnesses is directly related to the concepts outlined in Holtzhauzen. AI systems lack the qualities often associated with expert witnesses, despite their impressive speed and accuracy in evaluating vast amounts of data. AI systems, in contrast to human experts, are unable to take an oath, defend or explain their decisions under cross-examination, or take professional and legal responsibility for their findings. As a result, not all the conditions listed in Holtzhauzen for the admission of expert testimony can now be met by AI-generated opinions. Impartial expert witness.
III. Critical Evaluation and Comparative Perspectives
3.1 Critical Evaluation
The accuracy and effectiveness of South Africa’s legal system could be enhanced by artificial intelligence (AI). It can quickly analyse vast amounts of data, spot trends, and assist professionals in domains like finance, forensic science, and health.[17] Despite these benefits, AI is still unable to fully substitute human expert witnesses due to several crucial legal criteria.
Accountability is one of the primary issues. Human experts may suffer repercussions if they behave carelessly or dishonestly.[18] They are legally and professionally accountable for the opinions they offer in court. However, AI systems cannot be held personally accountable for errors or provide an explanation for who should bear the blame when an incorrect opinion is generated.
Transparency is another issue. Many AI systems function as “black boxes,” which means that they provide results without providing a clear explanation of how those conclusions were arrived at[19]. In order for judges and attorneys to evaluate the credibility of expert witnesses, South African courts need them to provide an explanation for their conclusions. It becomes challenging for the court to assess the reliability of an AI system’s evidence if it is unable to explain its findings.
Constitutional issues are also raised by AI’s incapacity to take part in cross-examination. The Constitution’s Sections 34 and 35 safeguard the right to a fair trial and the ability to contest evidence in court.[20] Recognising AI as an independent expert witness could jeopardise these rights because it is unable to respond to enquiries or defend its conclusions.
AI can play a significant supportive role even though it shouldn’t take the position of human expertise. AI can lower expenses, increase productivity, and help professionals analyse difficult data. Because of this, South Africa should permit the use of AI as a tool for decision-making while maintaining the responsibility of certified human experts for the evidence that is submitted in court.
3.2 Comparative Perspectives
Other jurisdictions have also adopted a cautious approach to AI in legal proceedings. The European Union’s Artificial Intelligence Act requires human oversight, transparency, and accountability when AI is used in high-risk sectors, including the justice system.[21] Similarly, courts in the United States continue to require expert evidence to be presented and explained by qualified human experts, even when AI tools are used during investigations.[22]
These international developments show that AI is viewed as a tool to assist experts rather than replace them. South Africa can learn from these approaches by developing legislation that regulates AI-generated expert evidence while protecting constitutional rights and maintaining public confidence in the legal system
Conclusion
By complementing expert analysis and helping with intricate legal and forensic investigations, artificial intelligence has the potential to increase the effectiveness and precision of South Africa’s legal system. AI cannot, however, be recognised as an independent expert witness under the current judicial system. Because AI systems cannot explain their reasoning, take an oath, or assume legal responsibility for their opinions, they are currently unable to meet South African law’s requirements for expert witnesses to provide trustworthy, objective, and accountable evidence that can be tested through cross-examination.[23] Consequently, existing constitutional principles, legislation, and case law support the use of AI as an assistive tool rather than a replacement for human experts.[24] Legislation governing the admissibility, transparency, dependability, and accountability of AI-generated evidence should be introduced by Parliament to guarantee the responsible application of AI in legal proceedings. With these changes, South Africa would be able to take advantage of technology advancements while safeguarding the right to a fair trial, defending the rule of law, and preserving judicial integrity.
Bibliography
Table of Cases
Daubert v Merrell Dow Pharmaceuticals Inc 509 US 579 (1993).
Holtzhauzen v Roodt 1997 (4) SA 766 (W).
Table of Legislation
Constitution of the Republic of South Africa, 1996.
Criminal Procedure Act 51 of 1977.
Electronic Communications and Transactions Act 25 of 2002.
Law of Evidence Amendment Act 45 of 1988.
Protection of Personal Information Act 4 of 2013.
Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act).
Books
Abbott R, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020).
Schwikkard PJ and Van der Merwe SE, Principles of Evidence (5th edn, Juta 2021).
[1] Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020)
1-15.
[2] European Commission, Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence
(Aerificial Intelligence Act) COM 2021.
[3] PJ Schwikkard and SE Van der Merwe, Principles of Evidence (5th edn, Juta 2021) 489–506.
[4] Holtzhauzen v Roodt 1997 (4) SA 766 (W) 772–773.
[5] Constitution of the Republic of South Africa, 1996 ss 34 and 35; Electronic Communications and Transactions Act 25 of 2002.
[6] Constitution of the Republic of South Africa, 1996, s 2.
[7] Constitution of the Republic of South Africa, 1996, ss 1, 9, 10, 34 and 35.
[8] Constitution of the Republic of South Africa, 1996, s 34.
[9] Constitution of the Republic of South Africa, 1996, s 35(3).
[10] Constitution of the Republic of South Africa, 1996, s 9.
[11] PJ Schwikkard and SE Van der Merwe, Principles of Evidence (5th edn, Juta 2021) 489–506.
[12] Holtzhauzen v Roodt 1997 (4) SA 766 (W) 772–773.
[13] Electronic Communications and Transactions Act 25 of 2002, ss 11–15.
[14] Holtzhauzen v Roodt 1997 (4) SA 766 (W)
[15] Holtzhauzen (n 1) 772–773.
[16] Holtzhauzen (n 1) 773.
[17] Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020) 1–15.
[18] PJ Schwikkard and SE Van der Merwe, Principles of Evidence (5th edn, Juta 2021) 489–506
[19] Ryan Abbott (n 1) 45–60.
[20] Holtzhauzen (n 3) 772–773.
[21] Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act).
[22] Daubert v Merrell Dow Pharmaceuticals Inc 509 US 579 (1993).
[23] Holtzhauzen v Roodt 1997 (4) SA 766 (W) 772–773; PJ Schwikkard and SE Van der Merwe, Principles of Evidence (5th edn, Juta 2021) 489–506.
[24] Constitution of the Republic of South Africa, 1996, ss 34 and 35(3).
