Steve Mutkoski, Microsoft Corporation, February, 2024
Health systems around the world are facing challenges, compounded by the multi-year experience of the Covid-19 pandemic. Almost universally, countries are struggling to make gains in areas such as overall health of patient populations; per capita cost of care; patient experience and engagement, and; clinician and health worker burnout (as well as associated problems with new staff recruitment). These challenges require innovative solutions and a coordinated effort from all stakeholders to improve the quality, accessibility, and affordability of healthcare.
Artificial Intelligence (including newly arrived “Generative AI”) is already proving to be a transformative tool in healthcare, offering remarkable potential to revolutionize patient care, diagnostics, and treatments, as well as to relieve healthcare provider burnout and ease administrative burdens. The term “Generative AI” is used here to refer to algorithms that can generate new content or data that bears striking resemblance to human generated content or data, including the well-known ChatGPT from OpenAI and the underlying GPT-3 and GPT-4 Large Language Models (“LLMs”).
We have witnessed massive growth in efforts to leverage AI and machine learning (ML) tools in the healthcare sector over the past few years. From ML models that help to analyze claims for reimbursement, triage cases, authorize treatments, manage clinic supply chains, or schedule frontline workers, a whole host of operational and administrative AI tools have been introduced to help healthcare entities leverage data to manage financial and workflow processes.
Microsoft has been at the forefront of not only the rapid advances in underlying AI technologies but also in committing to develop AI responsibly and actively participate in building appropriate legal and regulatory blueprints for AI. Starting with important direction from CEO Satya Nadella in 2016 who announced that, “We want to build intelligence that augments human abilities and experiences,” to direction from Vice Chair and President Brad Smith in his 2018 book The Future Computed about the ethical and societal implications of AI, that has evolved into Microsoft’s Responsible AI Principles and a companywide approach to developing, assessing, and deploying AI systems in a safe, trustworthy, and ethical way. The Microsoft Responsible AI Standard v2 sets out the Responsible AI Principles of Accountability, Transparency, Fairness, Reliability & Safety, Privacy & Security and Inclusiveness. These principles are universal and as relevant to the Health and Life Sciences verticals as they are to other fields.
More recently, Microsoft has made significant commitments and proposals in Governing AI: A Blueprint for the Future, where the company proposes a tiered regulatory approach that includes licensing of datacenter infrastructure for training and development of powerful foundation models, regulation of deployment of those models and compliance with existing and evolving laws for applications that use those models. The company made additional commitments to advance safe, secure, and trustworthy AI, including implementing the U.S.’ National Institute of Standards and Technology (NIST) AI Risk Management Framework and attest to alignment with it in contracts with customers.
Principles similar to those listed have been widely used by developers in bringing the existing generation of healthcare AI to market, resulting in hundreds of AI/ML-enabled medical devices that have received regulatory clearance and are currently in use in healthcare settings. Microsoft is optimistic that as it and others in the industry continue to advance Responsible AI practices, collective learnings will be useful as the industry explores how LLM-based tools can be safely designed for healthcare settings.
Considering the risk concepts embedded in the NIST AI Risk Management Framework, it is not surprising that Microsoft’s early conversations with health solution developers, healthcare providers and policymakers have tended to focus on a relative modest set of near term use cases for Generative AI. Those use cases involve creating draft summaries of existing documentation, extracting key information from records or documentation, summarizing medical journal articles or other information sources and assisting with other similar repetitive administrative tasks. We have also taken note of a growing body of research highlighting that, while Generative AI tools can be powerful and often generate compelling looking answers to complex questions, today’s LLMs may not meet clinicians’ real-world information needs “out of the box”, and so should not be used for more complex tasks that relate to recommendations for diagnosis or treatment of patients without further research and incorporation of mitigations and safety measures.
Like any new and powerful technology, education about the technology and teaching of new skills needed to use the technology safely and effectively will be critical. While it will be important for organizations developing AI to have comprehensive training about responsible AI, it will also be vital that organizations deploying health AI also invest in education and skills development for their employees who will often be the end users of these AI-enabled tools.
