The original AI Society dataset is in English and is composed of 25K conversations between two gpt-3.5-turbo agents. The dataset is obtained by running role-playing for a combination of 50 user roles and 50 assistant roles with each… See the full description on the dataset page:
https://huggingface.co/datasets/camel-ai/ai_society_translated.