The misuse of AI has made them a guirigay

With the ‘Abalos Case’ we are attending one of the most relevant corruption cases of recent years, but in the Supreme Court there has been a rocambolesque situation. The transcripts of the statements of Judge Leopoldo Puente or the former Minister José Luis Ábalos, among others, have become a division of the statement in a gallimatisms.

Incomprehensible statements. He Diari de Tarragona or the Diario de Navarra They have had access to documents with the transcripts of the statements of those involved in the ‘Abalos Case’. These transcripts seem to demonstrate an “uncontrolled and without supervision” use, they say in the Diario de Navarra, the AI ​​tools to transcribe these texts. The lawyer of the Supreme Administration of Justice has sent those written copies, but in them there are incomprehensible paragraphs.

Terrible transcripts. These documents show several examples of disastrous transcripts. When answering Ábalos about his alleged presence in flats related to prostitution, it seems to say “it really is that in Madrid not from Madrid.” In another question of the judge, he answers according to the transcription “I do not remember right now, because on that date he already has but regardless of this you do not remember that it is compressible that he cannot remember the dates.” To the question about the illicit use of a villa in the line of conception, the transcription indicates that the answer was “occasionally because a time of the time that we were not, but it was caused, it was fixed if that 2 with that I have completed my income did not have any other income.”

AI or correctly identifies witnesses. This disaster extends to the identification of the witnesses. In the Diario de Navarras they comment on what happened with Claudio Rivas, “Víctor de Aldama’s greatest partner in the Macrofude de los Hydrocarbos.” The transcripts documents show that the AI ​​called this witness “Claudia Arribas”, “Claudia up” or “Claudio Arribas”. Also in one of his answers about his activity is equally absurd according to the transcription of AI: “A group of companies that is dedicated to the hydrocarbons of hydrocarbons, for the sale of hydrocarbons.”

But the remedy that the disease. As indicated in the Diari de Tarragona, it was decided to use AI for transcripts when to avoid leaks, access to the original videos of those interrogations was limited. The misuse of AI has caused the only way in which parties can now analyze the statements is to request the court to vision of those original recordings.

Misuse of AI. Here we have a worrying example of what can happen to a misuse of AI in official organizations. The problem is not known, and the AI ​​system may not be thought for transcripts in Spanish, for example, but what is clear is that no one reviewed the transcription made by AI, something that must be an essential part of the management of these systems. The propensity of the chatbots to invent data and hallucinar is already known, and here those limitations were clear.

It cannot be left alone. A El País article From November it reminds us how these types of situations are common when using translation tools that often apply literal translations that make no sense. It happened with the embarrassous translation of the Official Tourism Website of Santander, which in 2018 translated the name of the Botín Center as “Loot Center” (center of the looting “or the historic center as” Historic Helmet “(Helmet It is a helmet in English, but referring to the protective helmet of the motorists, for example).

Be careful with whisper and transcripts by AI. Researchers at the University of Michigan They analyzed Whisper transcriptsa popular and remarkable OpenAI application for audio transcripts. Hallucinations (errors and invented phrases) were detected between 38% and 80% of the cases analyzed. As they point out in the country, already an OpenAI spokesman recognized the limitations and stressed that “the tool should not be used in high -risk environments.”

CLATIGPT CLAKES. A few months ago we attended another unique case of a misuse of AI. A lawyer He filed a complaintbut to save time and resources he turned to the use of chatgpt for his initial brief. The problem is that AI had included erroneous references to Colombian legislation instead of applicable Spanish regulations. The lawyer realized the error and quickly presented a letter assuming responsibility and excusing himself for that improper use. The Superior Court of Justice of Navarra decided not to impose a sanctionbut again that risk of using these tools in inappropriate forms and without reviewing the texts or content they generate was evident.

Spain has Aesia precisely for this. Last year we talked about the Spanish Artificial Intelligence Supervision Agency (Aesia). This agency has precisely as a mission to investigate and fine those who misuse AI from 2025. Research began theoretically on February 2, and on August 2 this agency will assume full sanctioning power. We will see if they take letters in this misuse of AI in the ‘case Ábalos’, but of course it would be a clear example of a inappropriate and even dangerous use of AI.

Image | Ministry of Transport

In Xataka | The companies of AI have been jumping the copyright for years. They have just suffered a disturbing legal defeat