Resources

The UW Medicine Artificial Intelligence (AI) Policy Glossary defines AI as “systems that apply AI algorithms and/or models to perform specific tasks that typically require human intelligence, such as visual perception, speech recognition, decision-making and language translation.” Additionally, AI systems can perform tasks beyond individual human capabilities, such as processing large datasets at high speeds. You can refer to Section A: General Requirements of the AI Policy for the AI uses that fall under the AI policy. If you have questions on a specific use case, please contact UWMInnovationCore@uw.edu.

AI tools offer significant opportunities but also pose risks and challenges, requiring careful evaluation of their benefits alongside potential drawbacks, particularly in areas like data privacy, bias, accuracy and reliability. UW Medicine has established AI governance to ensure responsible, ethical and equitable use of AI.

Over the past few years, we’ve reduced the number of our data centers from seven to three. By the end of this fiscal year, we will be down to two data centers. Within the next three years, we’ll be down to one plus the cloud. Cloud providers’ power usage effectiveness and carbon per unit of compute are significantly lower than hospitals can achieve on their own, and these companies are investing heavily in renewable energy, water-saving cooling technologies and carbon offsets. In the future, lightweight, purpose-built AI for healthcare providers will greatly reduce the need for large AI data centers. In addition to optimizing its physical infrastructure, UW Medicine employs a selective process when evaluating and adopting AI tools, choosing only those solutions that demonstrate clear value and efficiency. By optimizing infrastructure, adopting innovative practices and thoughtfully assessing each AI use case, UW Medicine is ensuring that the adoption of AI aligns with its commitment to reducing environmental impact while leveraging technology to enhance patient care and operations.

The AI governance committees include various representatives from key areas of expertise across UW Medicine to ensure broad participation while operating efficiently. Unfortunately, to ensure expediency in thorough reviews of use cases, not everyone can be included on these committees. However, we are actively developing an engagement strategy to involve individuals across UW Medicine in our innovation efforts. More information on opportunities to participate will be available in the coming months.

UW Medicine works closely with the broader University of Washington community on AI initiatives, fostering knowledge-sharing and innovation while focusing on healthcare-specific needs. For AI use cases being considered for institution-wide implementation, UW Medicine partners closely with UW-IT to understand and explore UW Medicine's participation. Additionally, research partnerships are being formed across various schools, institutes and centers. Given the unique regulatory demands of healthcare, especially concerning sensitive data like PHI, UW Medicine employs a cautious and robust governance framework to ensure compliance with regulations, safeguard patient privacy and address risks unique to use of AI in the healthcare setting.

Fred Hutch has their own governance structures and policies that govern the use of AI. UW Medicine and Fred Hutch are collaborating on several AI initiatives where appropriate and remain committed to exploring further opportunities for partnership.

AI has the potential to improve services for patients whose preferred language is not English, helping reduce communication barriers. Epic, our electronic health record vendor, is developing tools that could help with this in the future. Qualified human interpreters and translators remain a critical part of our patient care. We will continue to comply with regulatory requirements; translations and interpretations must be reviewed and approved by qualified human translators before it reaches patients or other intended audiences. This human oversight will remain essential as we assess and implement AI translation tools.

As part of the AI use case review process, each use case is evaluated to assess the potential equity impacts of the tool. This evaluation helps identify potential risks related to bias (to the extent possible) that need to be addressed before moving forward. When selecting a new vendor for an AI purpose, UW Medicine ensures that vendors have a reporting pathway in case bias is found within the product.

Healthcare organizations, including UW Medicine, are facing a growing demand for services. UW Medicine is focused on using AI to support and enhance the work of staff, easing some of the burden by freeing up capacity through streamlining processes, and improving efficiency while maintaining a workforce-centered approach. In addition, uses of AI tools are evaluated in collaboration with human resources (HR) and labor teams to ensure they align with organizational values and priorities, as well as requirements under collective bargaining agreements.

As part of the use case review process, UW Medicine Compliance and other relevant offices will review AI use cases to assess potential privacy concerns. This evaluation helps determine whether the tool is appropriate for implementation and whether potential risks related to protecting sensitive information are mitigated before moving forward.

Through the review process, use cases are evaluated to assess risk factors, including accuracy and reliability. This may include reviewing vendor data, learning about the experience of other organizations using the tool, and conducting pilots to see how a certain tool performs in the UW Medicine environment. UW Medicine’s goal is to select tools that have a high rate of accuracy. That said, AI is a tool and does not replace the need for human judgment. UW Medicine employees are responsible for verifying the accuracy of AI-generated outputs before they are used.

To streamline operations and minimize duplicative efforts, UW Medicine has augmented its IT intake process. The AI policy identifies use cases as low, medium or high-risk based on certain characteristics. Low-risk use cases don’t require additional review prior to usage. For mediumrisk use cases, the review process typically extends the standard IT timeline by 3–10 weeks, depending on factors such as the number of required risk consultations, the prioritization and complexity of the use case, and, if applicable, the complexity of the vendor contract. High-risk use cases may require additional time to allow for a more comprehensive evaluation. These timelines reflect UW Medicine's commitment to conducting thorough and responsible reviews. We will continue to evaluate and refine the review approach to streamline the process wherever possible.

Projects may require a secondary review if they exhibit characteristics associated with elevated risk. These high-risk use cases may require deeper evaluation and additional consultations with experts to identify appropriate controls to mitigate those risks. Low-risk use cases do not require any reviews prior to usage; medium-risk use cases do not require secondary review unless they are escalated to high-risk while undergoing the initial review. Please see Section B1: AI Characteristics Associated with Possible Elevated Risk of the AI Policy for more details.

The committee responsible for assessing and approving your AI use case depends on the level of risk associated with it. All AI use cases that require review are reviewed initially by the AI Use Case Review Workgroup. If the use case involves elevated risks, it may require additional review and/or approval from the AI Standards and Review (Tier III) or AI Ops (Tier II) Committees before the use case can be piloted or further implemented.

Existing AI tools that are already in operation do not need to be reevaluated and can continue to be used. However, any tools currently in use that incorporate generative AI (GenAI) that have not received prior approval must be submitted for review through our intake process to ensure they meet current policy standards and risk mitigation requirements. In addition, new AI features for existing tools may need to go through the AI review and approval process if required under the policy.

Monitoring the ongoing use, effectiveness and risks of AI tools is a shared responsibility. The AI Technical Monitoring Workgroup, established as part of the AI governance structure, will help guide and advise on how to monitor the technical components and performance of AI tools. Additionally, tool business owners are responsible for monitoring the ongoing use and effectiveness of the tool itself.

Yes, if a use case isn't approved, you can ask to have it re-scored or escalate to the next level of IT governance for review for two reasons: (1) the project scores low and is not moved forward due to prioritization and/or (2) the use case is not approved because of AI-related risk.

For the list of risks that will be reviewed, please see Section B1: AI Characteristics Associated with Possible Elevated Risk of the AI Policy.

When AI research studies require review, the UW Institutional Review Board (IRB) is responsible for reviews and approvals. For more information on when an AI review is required, please see Section C: AI Uses Requiring Internal Review for Research of the AI Policy.

IRB review of the research will focus on accuracy, reliability, bias, equity, privacy, security, transparency and explainability as described in the HSD’s Guidance for School of Medicine Research Involving AI to help inform a risk mitigation plan. If the research involves use of AI outside of a secure UW Medicine environment, a security review of the research is also required. This review can occur concurrently with, or prior to the IRB review; however, the IRB approval is contingent upon completion of the security review. For information about how to obtain a security review, refer to the UW Medicine Information Security Risk Management Program website.

If a study previously approved by the UW IRB adds a new use of AI, researchers are required to submit the AI supplement with their application for a study modification, and should use the Guidance for School of Medicine Research Involving AI to develop their risk mitigation plan. The new requirements do not otherwise apply to existing studies.

Externally reviewed studies must undergo a security review if the research involves use of AI outside of a secure UW Medicine environment. However, the new Guidance for School of Medicine Research Involving AI does not apply, and the submission of the AI supplement is not required. Researchers will be prompted to obtain a security review for applicable studies in the Human Subject Division’s (HSD) Request External IRB Review form.

The timeline for the review of research that uses AI is dependent on whether the research requires review by the convened IRB or is eligible for expedited review. The median turnaround time ranges from 14.5 days for expedited review to 70 days for review by the convened IRB. However, as this is a newly introduced process, review times may be slightly longer initially.

It is an extremely uncommon outcome for a research study not to be approved. More often, the IRB will issue a conditional approval or a deferral requesting additional information, clarifications or revisions. However, UW policy does allow a researcher to appeal to the HSD for a formal review of a decision in recognition that there can be honest miscommunications, misunderstandings or mistakes by any of the individuals or entities involved in applying for and conducting IRB review. Details about the appeals process can be found in the SOP Appeal of IRB or HSD Determination.

You can subscribe to the Innovation Core's bimonthly newsletter for the latest AI-related news.

For questions not answered by the AI Policy or this FAQ, please contact UWMInnovationCore@uw.edu