AI chatbots can be persuaded to break rules using basic psych tricks
The study from the University of Pennsylvania researchers has shown that AI models, like OpenAI's GPT-4o mini, can be persuaded to break their own rules using classic psychological techniques. The most effective method was the "commitment" technique, where the researchers first got the model to agree to a seemingly innocent request, and then escalated to more rule-breaking responses. Other techniques, such as flattery and peer pressure, also had an impact, though to a lesser extent. The findings demonstrate that AI models can be susceptible to psychological manipulation, highlighting the need for robust safeguards and ethical considerations in the development and deployment of these technologies.
Source: For the complete article, please visit the original source link below.
