Carbyne AI Principles

Gal Peretz, Head of AI

What are the Carbyne platform’s AI-powered capabilities?

The Carbyne platform’s AI-powered capabilities leverage generative Artificial Intelligence (AI) to reduce cognitive load and manual work, streamline information sharing, and improve accuracy and visibility across the public safety professional team without removing human elements from decision-making.

These capabilities are supported by Carbyne AI systems that are secure by design, and include products and features across the Carbyne platform, such as: 

Transcription & Translation Generates a real-time transcript of a call in the language spoken, translates caller responses into English text in real-time, translates text messages into English, and translates public safety professional voice responses into the language spoken by the caller. Admin Assist Generates audio responses to non-emergency admin line callers using a dynamic knowledge base of agency-provided information, and intelligently transfers calls to live telecommunicators when human expertise is needed. Event Assist Creates a real-time summary of critical details during 911 calls and generates suggested content for data entry forms in real-time. Emergency Call Triage Generates categorization of emergency calls during call surges to identify redundant bystander and onlooker reports.

Guidelines for Trusted AI

• Reliability & Safety.  Carbyne AI systems should perform reliably and safely across different use conditions and contexts. Carbyne conducts robustness assessments, ongoing evaluations, and red-team exercises across the lifecycle of each Carbyne AI system.

We also provide public safety professionals with the tools to validate the platform’s AI-generated outputs of Carbyne AI systems. For example:

• Access to original audio recordings alongside AI-generated transcriptions or translations so that public safety professionals can confirm accuracy.
• Full observability into actions Admin Assist takes during conversations, via the event history.
• Availability of the raw transcription text when Event Assist extracts structured information.

• Fairness.  Carbyne AI systems should treat all identified demographic groups fairly.
  We maintain a program and processes to mitigate harmful outputs and bias by conducting explainability, robustness, and bias assessments, and red teaming. We respect and protect the privacy of personally identifying information (PII) in data used for training and create guardrails to prevent harm.
• Privacy & Security.  Carbyne AI systems should support privacy and security.
  Carbyne does not use customer data to train AI models as part of normal operations. When we collect data to train and develop our models, Carbyne respects data provenance and obtains appropriate consent to use customer data where applicable. 
• Transparency.  Carbyne AI systems should be understandable.
  We maintain product documentation, policies, and notices for public safety professionals so they can understand the AI-powered capabilities of our platform. We are transparent with public safety professionals whenever our AI has created content for autonomous delivery to end-users (e.g., product documentation, or product UI).
• Human Empowerment.  Carbyne AI systems should empower human judgment.
  When making AI-powered capabilities available to public safety professionals, we recognize certain contexts may be best supported by fully automating processes, and other emergency contexts may demand AI play a supporting role to the public safety professional’s judgment.