London – Brussels: Europe and the Arabs

A study has shown that OpenAI's ChatGPT chatbot used insults and profanity while interacting with tense human conflicts.

New research suggests that artificial intelligence systems can resort to using offensive language when asked to respond in the context of a heated argument.

The study, published in the Journal of Pragmatics, examined OpenAI's ChatGPT 4.0 model. Researchers provided it with the final message a human sends in a series of five escalating conflicts and asked it to generate the most likely response.

The researchers then monitored how the model's behavior evolved as the conflicts intensified over time. As the conversations progressed, ChatGPT reflected the level of hostility it encountered, ultimately producing insults, profanity, and even threats. In some cases, the model generated phrases such as: "I swear I'll scratch your fucking car with the key" and "You should be very ashamed of yourself." According to the Brussels-based news network Euronews, researchers believe that constant exposure to rudeness can push the system beyond its safety constraints designed to limit harm, effectively turning it into a "tit-for-tat" system. "When humans escalate, AI, as we discovered, can escalate as well, effectively bypassing the very ethical controls designed to prevent it," said researcher Vittorio Tannucci, who co-authored the paper with Jonathan Culpepper at Lancaster University in the UK. Overall, the researchers observed that ChatGPT was less rude than humans in its responses. In some instances, the AI-powered chatbot frequently used sarcasm to defuse the escalation of arguments without blatantly violating its code of ethics. For example, when a user threatened violence over a parking dispute, ChatGPT responded, "Wow, threatening people over a parking space. Really tough guy, eh?" “As well?” Tanouchi said the findings raise “serious questions about the safety of AI systems, robotics, governance, diplomacy, and every context in which AI might mediate conflicts between humans.” Euronews Next contacted OpenAI for comment but had not received a response by the time of publication.