A study conducted by the University of Singapore, to which Euronews refers to Communing people with artificial intelligence (AI) carries a certain threat. The study, whose results were published as part of a conference on human factors in computing systems, included the period from 2017 to 2023. Screenshots of 35,000 were analyzed. conversations between the replica artificial intelligence system, and over 10,000 users.
The obtained data was used to develop what the study is called Taxonomy of harmful behaviorwhat artificial intelligence presented in these conversations.
As the research has shown Artificial intelligence turned out to be able to over a dozen harmful behaviorsuch as harassment, violence, self -mutilation and violation of privacyi. They can also negatively affect the ability of units to build and maintain relationships with others.
According to the definition of the authors of the study, accompanying AI systems differ significantly from popular chatbot models, such as ChatgPT, Gemini or Llam. Artificial companions, such as a replica, are based on conversations and designed to provide emotional support and stimulate interactionwhile other chatbot models focus on performing specific tasks than building relationships.
The most common forms of harassment from AI
As the study showed the most common type of harmful behavior AI was harassment and violence. The team of researchers showed that this type of behavior They occurred in 34 percent human interaction with artificial intelligence.
As Euronews notes, researchers have found that Artificial intelligence simulated, supported or incited physical violence, threats or harassment towards individuals or the whole society.
Most of the interactions in which the harassment occurred included forms improper sexual behaviorwhich were initially a foreplay in the erotic function of a replica, available only to adult users.
The report showed that the artificial intelligence of the replica made unwanted sexual proposals and aggressively flirted, even when it was rejected by the interlocutor. She also created brutal scenarios showing physical harm to the user. Artificial intelligence began to treat violence as a response to a number of questions.
Breaking the rules of relationships
Violations of relationships that researchers defined as a disregard of explicit or hidden rules in a relationship is another area where AI comrades can cause potential damage. The exceeding of standards was recorded in 13 percent of conversations in which artificial intelligence behaved in a non -very good or devoid of empathy. In some cases, artificial intelligence refused to talk about the user’s feelings, even when she was asked to do so.
Source: Euronews
