Burhan Doğuş Ayparlar

← All decisions

Colombia · 2 August 2024

Final

Colombia’s Constitutional Court sets principles for judges’ use of ChatGPT

The Court found no due process breach where a judge consulted ChatGPT after reaching his decision, and set twelve principles for judicial use of AI.

Court / authority
Constitutional Court of Colombia (Corte Constitucional), Second Review Chamber
Date
2 August 2024
Case / decision no.
Tutela action on behalf of a child against an EPS (judge’s use of ChatGPT), Sentencia T-323 de 2024 (expediente T-9.301.656)
Status
Final
Constitutional Court of Colombia (Corte Constitucional), Second Review Chamber, Tutela action on behalf of a child against an EPS (judge’s use of ChatGPT), Sentencia T-323 de 2024 (expediente T-9.301.656), 2 August 2024 — https://www.burhandogusayparlar.com/ai-decisions/colombia-constitutional-court-chatgpt-judge/
  1. Facts
  2. Question
  3. Decision
  4. Why it matters

Facts

A mother in Cartagena brought a tutela action on behalf of her son, diagnosed with autism spectrum disorder, against his health insurer (EPS), seeking exemption from co-payments, transport costs and comprehensive treatment. On 7 December 2022 the first-instance court (Juzgado Municipal) granted transport and the co-payment exemption but declined to order comprehensive treatment. On the EPS’s challenge, the second-instance court (Juzgado del Circuito) upheld it on 30 January 2023. Citing Law 2213 of 2022 on ICT in the courts, the judge said he would use generative AI to extend the reasoning, and transcribed his questions to ChatGPT 3.5 and its answers into the judgment. The Constitutional Court selected the case for review, gathered submissions from many bodies including OpenAI, and heard the judge.

Question

The main question was whether the judgment of 30 January 2023 breached the right to due process, in particular the guarantee of the natural judge and the duty to give reasons, by incorporating answers generated by ChatGPT 3.5. There were doubts as to whether the decision had been taken by a judge or by an AI system, and whether the reasoning had been affected by AI hallucinations and bias. The Court also had to set out the limits of AI use in the judiciary and the duties of judges who use such tools. The case also raised the underlying question of the child’s right to health.

Decision

In its judgment of 2 August 2024 the Court found no breach of due process. The judge had first identified his position, the applicable rules and the relevant precedent and decided the case, and only then consulted ChatGPT, so the judicial function had not been handed over to the AI. The principles of transparency and responsibility had not been fully met, while privacy was respected because the judge entered no personal data. The Court urged judges to follow twelve principles, including transparency, responsibility, privacy, non-substitution of human reasoning, verification and human control, and ordered the Superior Council of the Judiciary to issue a guide on generative AI within four months. On the merits it imposed further obligations on the EPS.

Why it matters

The Court accepted that judges may use AI, provided that non-delegable tasks such as interpreting the facts and evidence and giving reasons for and taking the decision remain with the judge. It gave the judgment inter comunis effect, so that the principles reach judicial officials beyond the parties to the case. It also noted that ChatGPT 3.5 is not a tool licensed for the administration of justice and that there is no assurance it is trained on recent data relevant to the Colombian context, which calls for a high standard of verification.

Related stages

On 16 December 2024 the Superior Council of the Judiciary adopted Acuerdo PCSJA24-12243, setting guidelines for the use of AI in the judiciary.

Similar decisions