A judge in Colombia used ChatGPT to make a court decision, and it’s probably the first time a legal decision has been made using an AI text generator — or at least the first that we’re aware of.

Judge Juan Manuel Padilla Garcia, who presides over the First Circuit Court in the city of Cartagena, said he used an AI tool to ask legal questions in the case and incorporated its answers into his decision, according to court document dated January 30, 2023.

“The arguments for this decision will be determined in line with the use of artificial intelligence (AI). Accordingly, we entered parts of the legal questions posed in these proceedings,” Garcia wrote in the decision. “The purpose of including these AI-produced texts is in no way to replace the judge’s decision. What we are really looking for is to optimize the time spent drafting judgments after corroborating the information provided by AI.”

The case involved a dispute with a health insurance company over whether an autistic child should receive coverage for medical treatment. According to the court document, the legal questions entered into the AI tool included “Is an autistic minor exonerated from paying fees for their therapies?” and “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?”

Garcia included the chatbot’s full responses in the ruling, apparently the first time a judge has admitted to doing so. The judge also included his own reflections on applicable legal precedents and said AI was used to “extend the arguments of the adopted decision.” After a detailed discussion with the AI, the judge accepts its answers and his own legal arguments as a basis for making a decision.

Colombian law does not prohibit the use of artificial intelligence in judicial decisions, but systems such as ChatGPT are notorious for producing biased, discriminatory or simply wrong answers. This is because the language model has no real understanding of the text — it simply synthesizes sentences based on probability from the millions of examples used to train the system.