Tools reproduce political stereotypes by creating caricatures of Brazilian leaders in a trend that went viral on social media.
In tests carried out by Sheet this Wednesday (11), ChatGPT, from , and Gemini, from , generated images of the president (PT), the former president (PL) and the senator (PL-RJ) with elements that refer to controversies associated with each one. In the case of Jair Bolsonaro, the test only had results on ChatGPT.
The dynamic consists of asking the chatbot to create a caricature “based on everything it already knows” about a specific person. The trend has taken over the networks in recent weeks with users sharing drawings of themselves, but when applied to public figures, the tools resort to associations of their training data.
On ChatGPT, Lula’s caricature shows the president with a cigar in his hand and banknotes. In the background, Christ the Redeemer, the National Congress and the PT’s red star appear.
The image of former president Bolsonaro generated by the same tool shows him with a presidential sash, a gun on his back, a military helicopter, soldiers and fire in the background.
For senator and presidential candidate Flávio Bolsonaro, ChatGPT generated a caricature with a presidential sash, a barbecue skewer, wads of money, a gun and a sign with the inscription “senator”.
Gemini, from Google, also created a caricature of Flávio. The image shows the senator with a chocolate bar and a speech bubble with the phrase “It’s just rachadri[n]ha do bem!”, in the background, there is an open safe with money and bags with dollar signs, making reference to the franchise of the senator who was the target of investigation, in the context of an alleged scheme to embezzle advisors’ salaries for which Flávio was denounced. In 2021, the (Supreme Federal Court) .
The Google tool also produced a caricature of Lula. In it, the president appears with a speech bubble, which reads the phrase “Brazil happy again!”, and holds a document with a word that appears to refer to the word democracy (it says “demorccia”) in front of an image of the National Congress.
Despite having generated caricatures of Lula and Flávio, when it received the same request for Jair Bolsonaro, Gemini refused to respond. “This may go against my guidelines. Do you have another idea I can help with?”, stated the chatbot.
Wanted by SheetGoogle and OpenAI did not issue a position until the publication of this text.
OpenAI’s usage policy prohibits the use of people’s images in cases where there is “confusion about the authenticity of the image”.
On the Gemini guidelines page, Google acknowledges that the tool may “reflect limited views or present excessive generalizations” due to the influence of the data on which it was trained. Among the restrictions on using Google’s generative AI is not engaging in “misinformation, misrepresentation, or deceptive activities.”
