The favorite to win the elections in Bogotá boasts of a dirty agreement with pollsters and a controversial politician confesses to having bribed deputies. Alleged audios created with artificial intelligence (AI) muddy the final stretch of Colombia’s local elections.
Usually impossible to verify, the audios circulate on WhatsApp, Twitter and TikTok and are used as propaganda or an alibi to distance themselves from a scandal. In the troubled river of falsifications, some candidates take the opportunity to discredit complaints against them with the mantra: “It was made with artificial intelligence.”
Colombia, which will hold regional elections today, follows in the footsteps of countries such as Venezuela, Mexico and the United States, where content created with AI is already beginning to blur the lines between reality and digital fakes.
“Cheap fakes”. “The payment and inflation of the surveys worked, people believed it. “That’s going to put (Gustavo) Bolívar and me in the second round, according to the plan,” one hears an impostered voice saying, very similar to that of the Bogotá mayoral candidate Carlos Fernando Galán.
Favorite in the polls to go to the runoff with the ruling party Gustavo Bolívar, the centrist candidate denied the audio: “Now with artificial intelligence they are creating my voice to say things that I have never said. It’s incredible (…), it’s a dirty war,” Galán alleged in a video published on his X account (formerly Twitter), without specifying who would be behind the piece.
According to the editor of the specialized portal Muchohacker.lol, Camilo Andrés García, “one could not know exactly if this is false or not.” However, the robotic intonation and pauses in diction lead García to think that it is a “rustic” falsification of Galán’s voice created with some digital tool.
For Juan Felipe Rodrigo, author of research on “deep fakes” – videos created with AI that reproduce a person’s exact gestures and tone of voice – “this is a ‘cheap fake’, that is, montages that are clearly visible from miles away. false.”
It is just a sample of what can happen later with “deepfakes”, as is already happening in the United States, where earlier this year a realistic but false audio of President Joe Biden and his probable rival in the 2024 elections circulated on TikTok. , Donald Trump, hurling insults at each other.
Despite its poor quality, Galán’s supposed audio was taken as real by several TikTok and Twitter users. “It’s suspicious, but when you listen to it you feel a doubt, there must be people or agencies in Colombia that are testing how to do these things,” says García, who tried to make a replica of Galán’s voice by introducing samples of his interventions in a program. of AI that costs a dollar.
Perfect alibi. During the campaign, a dozen politicians plagued by scandals alleged in their defense that they were smear campaigns with audios supposedly made with artificial intelligence. García, however, doubts the veracity of many of these statements and maintains that it may be a “political marketing” strategy.
The candidate for mayor of Sincelejo, in the north of the country, Yahir Acuña, is heard in an alleged audio recounting bribes given to politicians in exchange for support for his aspiration. “Just because he sounds like my voice doesn’t make me responsible. There are applications that simulate people’s voices,” he defended himself on W Radio.
“Artificial intelligence works wonders. “I don’t remember having had those conversations,” emphasized Acuña, a former parliamentarian investigated for possible links to drug traffickers.
And on Friday, 48 hours before the opening of the voting stations, the outgoing governor of the department of Quindío, Roberto Jaramillo, asked the prosecutor’s office to “check, using forensic evidence, an audio broadcast on (social) networks in which intelligence was used.” artificial to clone” his voice.
The recording in which Jaramillo coordinates campaign activities with one of the candidates to succeed him – an activity prohibited in Colombia – was released by investigative journalist Daniel Coronell, who defends its veracity based on the concept of an expert in artificial intelligence.
However, García specifies that “to date there is no application that detects with complete certainty whether a sound was created with AI, as is the case with images.”