Kenyan Content Moderators Traumatized While Training OpenAI's ChatGPT

The event involves the development and use of an AI system (OpenAI's GPT) where human annotators were exposed to harmful explicit content to train the AI's content moderation capabilities. This exposure caused real psychological harm (insomnia, anxiety, depression, panic attacks) to the workers, which qualifies as injury or harm to the health of persons (harm category a). The AI system's development and use directly led to this harm, making this an AI Incident under the provided definitions.

The event involves the development and use of an AI system (OpenAI's GPT) where human annotators labeled explicit content to train the model. The process caused direct psychological harm to the workers involved, including trauma, anxiety, and depression. This harm is a direct consequence of the AI system's development and use, meeting the criteria for an AI Incident. The article does not describe potential or future harm but actual realized harm to people involved in the AI system's training.

The event involves an AI system (ChatGPT) and its development process, specifically the human-in-the-loop training method (RLHF) that requires trainers to label explicit and harmful content. The trainers have suffered direct psychological harm (trauma, panic attacks, anxiety, depression) as a result of their work, which is a direct injury to health caused by the AI system's development. Therefore, this qualifies as an AI Incident under the definition of injury or harm to health caused by the development of an AI system.

The event involves human contractors performing content moderation and labeling to train AI systems such as ChatGPT, which is an AI system by definition. The work exposed them to harmful explicit content, causing direct psychological harm (trauma, anxiety, depression). This harm is directly linked to the AI system's development and use, as the labeling was essential for the AI's alignment and safety features. The article documents realized harm to health, fulfilling the criteria for an AI Incident under harm to persons. Although the harm is to contractors rather than end users, it is a direct consequence of the AI system's development process.

The event involves human contractors exposed to highly disturbing content as part of the Reinforcement Learning from Human Feedback (RLHF) process used to train ChatGPT. This exposure caused psychological trauma, including anxiety, depression, and panic attacks, which are harms to health. The AI system (ChatGPT) is explicitly involved, and the harm is directly linked to the AI system's development and use. The article details realized harm, not just potential harm, and thus this is an AI Incident rather than a hazard or complementary information.

The event involves human labelers exposed to harmful content as part of training an AI system (OpenAI's ChatGPT) to avoid explicit content generation. The psychological harm to these workers is a direct consequence of the AI system's development process. The harm is significant and clearly articulated, including trauma, anxiety, and depression. Therefore, this meets the definition of an AI Incident under harm to health of groups of people caused by the AI system's development.

The event involves the development and use of an AI system (OpenAI's GPT) that required human labeling of explicit content to train the model for content moderation. The workers experienced direct psychological harm (insomnia, anxiety, depression, panic attacks) as a result of their work with the AI system's training data. This constitutes an AI Incident because the AI system's development and use directly led to harm to people. The article does not describe a potential or future harm but actual realized harm. Therefore, the classification is AI Incident.