Skip to content

How RAI Works

The Responsible AI (RAI) platform is seamlessly integrated into AI services and Copilot solutions to provide real-time protection during user interactions. Regardless of the type of conversation—whether single-turn or multi-turn—or the content format—text or multimedia—RAI ensures continuous safety and compliance.

RAI Platform Architecture:

image-20241230093536210

Highlights of RAI Platform:

Real-time Annotation and Safeguarding: RAI processes all communication in real time, maintaining a complete conversational context within the scope of a single request connection. This ensures that the model receives the most comprehensive context possible during detection, enhancing semantic-level protection by leveraging the full conversational history.

image-20241230093522595

Flexible and Contextual Conversation Support:

  • Single-turn conversations: RAI can immediately evaluate and safeguard content within a single interaction.
  • Multi-turn conversations: RAI tracks context across multiple exchanges to provide ongoing protection.
  • Multimedia content: RAI extends its capabilities to analyze and protect conversations that include multimedia elements such as images or videos.

Decision Signals for AI Services:

RAI provides precise feedback signals to integrated services, enabling them to:

  • Continue Conversations: If no harmful content is detected, the AI service can proceed with the interaction seamlessly.
  • Interrupt Conversations: When harmful or unsafe content is identified, the service can pause or terminate the conversation to prevent further production of harmful content.

By integrating RAI, AI services and Copilot solutions can ensure a safe and compliant user experience, dynamically adapting to a variety of conversational scenarios and content types.