AI simulates children to test chatbot risks
A report on the development by the American company 'Circuit Breaker Labs' of software agents that simulate users of different ages and backgrounds to test protection systems in chatbots, with warnings from the World Health Organization and UNICEF regarding psychological risks to children and adolescents.
A technical report reviewing the efforts of the American company 'Circuit Breaker Labs' in developing AI agents that simulate children and adolescents from different backgrounds, languages, and cultures, aiming to test protection systems in chatbots before they reach real users. The company relies on the 'red teaming' method to subject models to difficult and unexpected scenarios, and says its agents can conduct over 100,000 simulated interactions. This coincides with warnings from the World Health Organization against using generative AI for emotional support among youth, and data from 'Common Sense Media' that 72% of adolescents in the United States have used AI companions. UNICEF data also indicates that 20 million children have used AI in 10 countries. The report raises a question about the sufficiency of simulation alone to protect children, pointing to the need for continuous testing, independent oversight, and age-appropriate design.