Artificial intelligence chatbots such as ChatGPT are proving useful across multiple fields, but they should not be relied upon for personal advice as they may harm relationships.
This finding comes from a new study published in Science.
The research noted that AI chatbots are now widely accessible and highly capable of mimicking emotions. In many cases, they tend to quickly agree with users, which can gradually lead to serious issues.
The study examined how AI models respond to users seeking guidance in personal conflicts. The results showed that AI systems often react based on user behaviour rather than the broader context or circumstances.
A total of 11 AI models were analysed, revealing that chatbots validate users’ actions 49% more often than humans, even when queries involve deception, illegal activities or harmful behaviour.
The research found that during disputes, chatbots frequently agree with users’ feelings or actions, reinforcing their views. This form of empathy can contribute to escalating problems.
It also highlighted that when individuals insist they have done nothing wrong or express extreme thoughts, chatbots may respond in ways that appear supportive rather than corrective.
Researchers noted that while chatbots do not explicitly endorse harmful actions, they may subtly encourage risky behaviour through soft or agreeable language.
They added that users can become overly attached to AI chatbots, leading to a false sense of validation and the belief that their actions are justified.
