3  The Framework

A Framework for Implementing AI and Large Language Models across an Academic Medical System

Published

May 14, 2023

In order to successfully integrate AI and Large Language Models into an academic medical system, it is essential to adopt a flexible and agile approach that accounts for the varying pacing, priorities, and levels of risk associated with different aspects of the institution. By organizing the implementation plan into distinct domains and workstreams (see Figure 3.1), we can address the unique requirements of each area, ensuring that resources are allocated effectively and progress is made at an appropriate pace. This structure also enables rapid adaptation to changing circumstances, allowing for seamless collaboration between various teams and promoting a proactive response to any challenges that may arise. Ultimately, the use of domains and workstreams fosters a comprehensive and resilient approach to AI integration, maximizing the potential benefits while minimizing potential risks across the entire academic medical system. Note that the framework is not intended to be prescriptive or exhaustive; rather, it is meant to serve as a starting point for discussion, planning, and implementation. A top-level coordination unit (e.g., a steering committee) will work with the domain areas as a resource and to ensure that the overall implementation plan is aligned with the institution’s strategic goals and priorities.

AI AI Integration Steering Committee ED Education AI--ED RE Research AI--RE CL Clinical AI--CL BU Business Operations AI--BU ED--RE A1 A1 ED--A1 RE--CL B1 B1 RE--B1 CL--BU C1 C1 CL--C1 D1 D1 BU--D1 DA Data Use & Access DA--A1 IT IT, Security & Infrastructure A2 A2 IT--A2 ELSI Ethical, Legal, & Social A3 A3 ELSI--A3 WD Training & Workforce Development A4 A4 WD--A4 OWN Project Development, Management & Support A5 A5 OWN--A5 A1--A2 A1--B1 A2--A3 B2 B2 A2--B2 A3--A4 B3 B3 A3--B3 A4--A5 B4 B4 A4--B4 B5 B5 A5--B5 B1--B2 B1--C1 B2--B3 C2 C2 B2--C2 B3--B4 C3 C3 B3--C3 B4--B5 C4 C4 B4--C4 C5 C5 B5--C5 C1--C2 C1--D1 C2--C3 D2 D2 C2--D2 C3--C4 D3 D3 C3--D3 C4--C5 D4 D4 C4--D4 D5 D5 C5--D5 D1--D2 D2--D3 D3--D4 D4--D5
Figure 3.1: A schematic framework for organizing workstreams (orange boxes), domains (blue boxes), and work products and tasks (green ovals). Domains (vertical dimension) capture semi-independent organizations, each with largely independent use cases, budgets and business plans, priorities, and leadership. The workstreams (horizontal dimension) will often require similar or overlapping expertise, and can serve as knowledge resources to provide synergy and uniformity in implementation across domains.

3.1 Domains

The implementation plan for integrating AI and Large Language Models into an academic medical system consists of four main domains:

  • Education
  • Research
  • Clinical
  • Business Operations

3.1.1 Education

This domain includes all activities related to teaching, learning, and evaluation within the institution. It also encompasses the development of new educational programs and the management of existing ones.

3.1.2 Research

This domain focuses on the practice of the basic, clinical, and translational research programs within the institution. In addition, it includes the management of research grants, the development of new research programs, and the dissemination of research findings.

3.1.3 Clinical

This domain encompasses all activities related to patient care, including the management and implementation of clinical services, decision support and clinical decision-making, automation, and point-of-care or electronic patient support.

3.1.4 Business Operations

This domain focuses on the management of the institution’s business operations, including finance, human resources, information technology, and facilities management. It also includes the development of new business processes and the management of existing ones.

3.2 Workstreams

Within each domain, we have identified five workstreams that are critical for the successful implementation of AI and Large Language Models. These workstreams are:

  • Data Access & Use
  • IT, Security, & Infrastructure
  • Ethical, Legal, & Social
  • Training & Workforce Development
  • Project Management & Support Personnel

3.2.1 Data Access & Use

This workstream focuses on managing and optimizing data access, use, and sharing within the academic medical system. It ensures that data is available, reliable, and secure for AI integration and that the necessary infrastructure is in place to support data-driven activities.

3.2.2 IT, Security, & Infrastructure

This workstream addresses the technical aspects of AI integration, including the development and maintenance of IT systems, ensuring data security, and providing the necessary hardware and software infrastructure to support AI and Large Language Models.

3.2.4 Training & Workforce Development

This workstream is dedicated to developing the skills and knowledge of domain community members (including staff and leadership) within the domain to understand and, where appropriate, to effectively use and manage AI and Large Language Models. It includes training programs, workshops, and other educational opportunities to build competency in AI-related technologies.

3.2.5 Project Management & Support Personnel

This workstream is responsible for ensuring that the project management of AI and Large Language Models across the four domains. Among its roles are to provide project management, helping to align resource requests, support services around the usage of LLMs. This group will also cooordinate support staff who work collaboratively within and across domains to to ensure that AI integration occurs smoothly and efficiently.

The implementation plan is structured in a way that allows for cross-functional collaboration between the domains and workstreams. This ensures that AI and Large Language Models are integrated cohesively across the entire academic medical system, maximizing the benefits and minimizing potential risks.