Santiago Murillo Caballero | Public Law
Since the launch of ChatGPT in November 2022 (OpenAI, 2022) and the subsequent popularisation of other artificial intelligence-powered chatbots based on large language models (LLMs), AI (artificial intelligence) has gained prominance in public discourse. The use of generative artificial intelligence (Gen AI) has doubled over the past five years, and adopting such technologies is expected to deepen across many industries (McKinsey Company, 2024). Consequently, the legal sector has not been exempt from the influence of Gen AI. Legal service providers (Bloomberg, 2024) and judicial authorities worldwide have begun employing Gen AI tools to optimise and automate processes. This essay explores the implications of Ruling T-323/2024, highlighting its potential to shape the integration of Gen AI in Colombia’s legal system.
In 2024, Colombia witnessed its first judgement, where the judicial use of Gen AI became the focal issue. This milestone was marked by Judgment T-323 of 2024 (Blanca v. EPS, 2024), issued by the Constitutional Court as part of its review of a tutela action[1]. Specifically, the Court reviewed a decision rendered by a Labour judge, who, assisted by ChatGPT, resolved a tutela action seeking protection of the claimant’s fundamental right to health. At first glance, the subject matter of the dispute may not appear novel. However, the Court decided to review the appellate ruling to evaluate the implications of Gen AI in the judicial reasoning process.
Judgment T-323 of 2024 examined whether using Gen AI to issue a ruling in a tutela proceeding violated the fundamental right to due process, given the potential influence of its inherent factual inaccuracies[2] and biases, among other risks posed by this technology. The Court concluded that, in this specific case, the fundamental right to due process was not violated. Gen AI did not replace the natural judge but was used to support the reasoning behind the judgment after the human judge had already reached a substantive decision. Nevertheless, the Court seized the opportunity to establish clear duties for judges who employ Gen AI in their decision-making processes.
In its ruling, the Constitutional Court identified several risks posed to essential components of the fundamental right to due process when judges utilise Gen AI tools. These risks may undermine guarantees such as the principle of natural judgment, the reasoning and justification behind judgements, and the evidentiary due process. The Court acknowledged that incorporating AI into the Judiciary requires defining guiding principles and best practices to safeguard the fundamental rights of those engaging with the justice system. Moreover, the ruling articulated specific duties for judges using AI tools, outlining instances of duty breaches: (i) when a judge delegates non-delegable reasoning functions to AI; (ii) when a judge does not disclose the use of AI and its purpose; (iii) when the judge does not verify the reliability of the information provided by AI; and (iv) when the judge does not ensure the protection of the Parties privacy and confidentiality.
While Judgment T-323 of 2024 is far from radical, it constitutes a landmark decision as it is the first time the Constitutional Court has addressed a legal issue centred on using Gen AI. The Court’s approach reveals a somewhat cautious stance towards using Gen AI in judicial administration. It underscores the need for the Superior Council of the Judiciary[3] to regulate such tools in judicial offices. However, it refrains from a definitive position beyond outlining the duties binding on judges adopting these technologies (e. g. fact checking and data protection).
The decision acknowledges and accepts the inevitability of Gen AI’s integration into the Judiciary without stigmatising or seeking to prohibit its use. The Court clarifies this position by distinguishing between two types of Gen AI usage in court: (i) administrative assistance for judicial officials and (ii) decision-making. According to the judgment, there is no objection to judicial officials leveraging Gen AI to optimise administrative or auxiliary processes, as proposed by Buckland (AI, Judges and Judgement: setting the scene, 2023). However, the Court emphasised the serious risks to fundamental rights when Gen AI substitutes judges in substantive decision-making.
Lastly, the ruling reflects an aspiration for a Gen AI system designed for the Colombian Judiciary. According to the judgment, a system specifically trained on Colombian law would mitigate the risks of hallucinated[4] or biased outputs, thereby safeguarding guarantees examined by the Court. However – for now– this remains an aspiration, given the current technological, budgetary, and regulatory limitations preventing the Judiciary from accessing such a tailored tool. While the Constitutional Court has placed Gen AI at the centre of legal debate, Judgment T-323 of 2024 marks the beginning of what will likely be a long and complex discussion about the use of Gen AI in judicial offices, not only in Colombia but worldwide. For now, as the judgment acknowledges, Colombia operates within a regulatory vacuum, with few legislative or regulatory initiatives addressing the use of these technologies in the exercise of public functions.
[1] Tutela action (acción de tutela in Spanish, which literally means protection action) is a mechanism provided by Colombia’s Constitution by which any person can demand from judges the protection of their fundamental rights in a faster and less formal procedure. Spain’s “recurso de amparo” is an equivalent to this institution.
[2] Which are technically called “hallucinations”.
[3] The Superior Council of the Judiciary (Consejo Superior de la Judicatura in Spanish) is the governing body of the Judicial Branch in Colombia.
[4] According to Ji, et. al. (Survey of Hallucination in Natural Language Generation, 2022), “a hallucination is an unreal perception that feels real” thus, hallucinated outputs give the user the impression of being real and accurate but are unfaithful in reality.
References
Blanca v. EPS, T-323/24 (Constitutional Court of Colombia August 2, 2024). Retrieved from https://www.corteconstitucional.gov.co/relatoria/2024/T-323-24.htm
Bloomberg. (2024, May 23). Bloomberg Law. Retrieved from How Is AI Changing the Legal Profession?: https://pro.bloomberglaw.com/insights/technology/how-is-ai-changing-the-legal-profession/#how-technology-is-changing-the-legal-field
Buckland, R. (2023, November). AI, Judges and Judgement: setting the scene. M-RCBG Associate Working Paper Series. United States of America. Retrieved from Harvard Kennedy School: https://www.hks.harvard.edu/sites/default/files/centers/mrcbg/working.papers/Final_AWP_220.pdf
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., . . . Shu Chan, H. (2022). Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 2-59.
McKinsey Company. (2024, April 2). What is Generative AI? Retrieved from https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai OpenAI. (2022, November 30). Introducing ChatGPT. Retrieved from https://openai.com/index/chatgpt/
