Azure AI Health Bot helps create copilot experiences with healthcare safeguards

The generative AI era is driving demand for chatbots and copilots for health that assist patients and medical professionals with various administrative and clinical tasks. These chatbots would potentially use large language models (LLMs) to generate conversational AI chat experiences that can provide accurate and reliable information based on large amounts of medical literature and data.

As a result of the growing demand, many healthcare organizations are striving to build their own healthcare copilot experiences that deliver intelligent and engaging chat experiences leveraging LLMs and generative AI.

In the process, healthcare organizations have realized that as part of healthcare’s unique needs, they need a way to combine the benefits of using generative AI for engaging chat experiences with the benefits of protocol-based flows and custom workflows to provide accurate and relevant information. A hybrid approach that combines both would allow them to offer a more personalized and comprehensive service to their customers and end users.

Moreover, healthcare chat experiences need to leverage domain-specialized models and health-specific safeguards to meet the healthcare industry quality bar.

To address these needs, we’re adding new healthcare-specific safeguards for generative AI in private preview within the Azure AI Health Bot services. Preview customers can experience an integration with Microsoft Copilot Studio, allowing healthcare organizations to build their own copilot experiences. Customers can sign up for the private preview here.

• Providing reusable healthcare-specific functionality: providing healthcare-specific, pre-built capabilities, use cases and scenarios—including pre-packaged healthcare intelligence plugins, templates, content, and healthcare-specialized skills and connectors.
• Answering the unique needs of healthcare: enabling customers to build copilots for their patients and doctors, supporting protocol-based workflows side-by-side with generative AI-based answers, and allowing customers to keep alignment with up-to-date industry standards, guidelines, and protocols.
• Applying healthcare-specific safeguards: allowing customers to build copilots responsibly adapted to healthcare needs, apply health-adapted compliance controls, and implement health-specific safeguards and quality measures that are specialized for healthcare.

Generative AI capabilities 

In April 2023, we announced the preview of Azure AI Health Bot with Azure OpenAI Service, enabling fallback answers based on generative AI.

Today, we are expanding those capabilities beyond fallback answers, enabling our healthcare customers to further enrich their copilot experiences with the following capabilities in private preview:  

• Power generative answers that are grounded on customer’s own sources. The sources are incorporated during the copilot experience, alongside authored descriptive scenarios, protocol-based pre-built flows, and skills. Customers are able to bring in their Azure OpenAI Service endpoint and index to enable generative answers grounded on their desired sources.
• Generative answers that are grounded on the customer’s websites. These sources are real-time queried and can include medical guidelines, health articles, patient treatments, frequently asked questions, appointment scheduling information, and many more. This approach ensures that patients receive not only medical guidance but also support for the many aspects of their healthcare journey.
• New healthcare intelligence capabilities to incorporate generative answers grounded on credible healthcare sources. Sources include the National Institutes of Health (NIH), the Food and Drug Administration (FDA), and others.
• Seamlessly use pre-built protocol-based healthcare intelligence capabilities such as symptom checkers and triage, and a rich gallery of pre-built protocol templates side-by-side with generative AI based answers.
• Credible generative AI fallback ensures reliable and accurate responses in healthcare-related scenarios. In cases where answers are not available, this feature leverages credible content to enhance responses, providing users with reliable guidance backed by clinical Retrieval-Augmented Generation (RAG) support. Helping to mitigate potential errors and ensures the delivery of trusted information in healthcare settings.

Built-in safeguards

Azure AI Health Bot with generative AI technology provides built-in healthcare safeguards now in private preview, for building copilot experiences that fits healthcare’s unique requirements and needs. Those include:

◉ Clinical safeguards include healthcare-adapted filters and quality checks to allow verification of clinical evidence associated with answers, identifying hallucinations and omissions in generative answers, credible sources enforcement, and more.

◉ Healthcare chat safeguards include customizable AI-related disclaimers that are incorporated into the chat experience presented to users, enabling the collection of end-user feedback, and analyzing the engagement through built-in dedicated reporting, as well as healthcare-adapted abuse monitoring, among other things.

◉ Healthcare-adapted compliance controls include built-in Data Subject Rights (DSRs), pre-built consent management, out-of-the-box audit trails, and more.

Source: microsoft.com