Anthropic prohibits repeated abuses against the Claude chatbot. How should we talk to artificial intelligence

The American company Anthropic, which develops the chatbot Claude, has stirred controversy after announcing that it will prohibit the repetitive and unjustified abusive behaviors of users towards its artificial intelligence systems. The new policy, applicable only in extreme cases, allows for the interruption of conversations with users who exhibit behavior considered "crude" and reignites the debate about how we should communicate with artificial intelligence, reports the BBC. This type of behavior will be included in the list of prohibited practices, along with harassment of other people, promotion of self-harm, and the creation of intimate images without the consent of the subjects. Although Anthropic specified that the new rule will only apply in "extreme cases" of repeated abuse, the decision has reopened the debate on how we should communicate with artificial intelligence tools. The company explained that the measure does not target "the usual manifestations of user frustration, expressing disagreement, dark creative themes, or testing and exploring models," suggesting that it will only apply in situations where the abusive behavior is clearly deliberate. However, it remains unclear what exactly Anthropic considers "abusive or crude behavior" towards its models. "You cannot be crude with numbers and mathematics" Screenshots of the new policy were distributed and intensely commented on social networks. Some users welcomed the measure, considering that it promotes good manners or contributes to the protection of the well-being of AI models. Others, however, criticized the initiative. One person called it "deeply wrong," while another suggested that it could reduce the severity of abuses directed against people and animals. "You cannot be crude with numbers and mathematics," wrote Dr. Barry Scannell, a technology specialist partner at the Irish law firm William Fry, on LinkedIn. "This level of anthropomorphization of artificial intelligence is harmful. It makes people believe that AI is something it is not, in reality," he added. Mustafa Suleyman, the head of Microsoft's artificial intelligence division, who in September criticized the rival company Anthropic for treating AI as if it were human, reacted to the policy update with an emoji representing a melting face. The changes, initially reported by the technology publication The Verge, were introduced following the company's annual review of its terms of use policy. It was also updated to address new or emerging risks in other areas. Among the changes are the prohibition of using Claude for "deceptive campaigns," as well as clarifying that the tool cannot be used to "mislead voters or disrupt elections," in the context of the approaching elections in the United States. Anthropic specified on Thursday that its policy already prohibited the use of the company's tools for weapons development. However, none of these changes attracted as much attention or provoked as many reactions as the decision to prohibit abusive user behavior towards chatbots. Should we be polite to artificial intelligence? The new rule brings up a hotly debated question: should we address generative artificial intelligence tools politely, or can we adopt a direct, even impolite tone? Some experts argue that using the expressions "please" and "thank you" in conversations with tools like ChatGPT is useless and consumes tokens, the units of text used by large language models to process requests and generate responses. Others believe that politeness can contribute to obtaining better answers. Last year, when a user on the X platform asked about the electricity costs generated by ChatGPT responses to messages containing "please" and "thank you," OpenAI's CEO, Sam Altman, replied that it is about "tens of millions of dollars spent on usage, one never knows." Beyond the debate about the necessity of adopting a polite tone towards artificial intelligence, some experts point out that the way people interact with chatbots could also influence interpersonal relationships. According to them, the habit of using aggressive or abusive language in conversations with AI could, in time, be reflected in how we communicate with each other. Editor: M.I.