Healthcare organizations embracing generative AI must heed Rich Birhanzel’s counsel. To responsibly implement AI, they must prepare proprietary data, establish robust controls, and align human involvement. Customized models using organizational data hold value, while careful risk assessment and governance are essential. People’s role in guiding AI and training for complex tasks are key. Generative AI’s potential to transform healthcare’s accessibility and experience is evident, requiring a proactive approach to harness its power.
In the realm of rapidly evolving healthcare, the incorporation of generative AI has gained tremendous momentum. This form of AI, underpinning innovations like the widely popular ChatGPT application, is regarded as a remarkable technological leap. Nevertheless, it is important to acknowledge that generative AI is far from flawless and can exhibit errors, akin to human mistakes.
To effectively harness generative AI within healthcare, a responsible approach is imperative. Rich Birhanzel, the healthcare lead at the esteemed research and consulting entity Accenture, offers valuable insights into achieving responsible integration. He underscores three pivotal actions: preparing proprietary data, establishing appropriate controls, and aligning human elements with the technology.
Q: One of your core recommendations for initiating responsible generative AI implementation involves preparing proprietary data. Could you elaborate on this aspect?
A: The vast capacity of large language models behind generative AI enables the processing of extensive datasets, potentially granting them a grasp of an organization’s entire knowledge repository. Language-based content, encompassing applications, documents, chats, emails, videos, and audio, can fuel groundbreaking innovations, optimization, and reinvention.
Presently, most organizations embark on preliminary endeavors by utilizing pre-existing foundation models. The true value emerges when entities personalize or fine-tune models with their proprietary data, thus catering to their distinct requirements.
However, fine-tuning foundation models necessitates access to domain-specific organizational data, semantics, knowledge, and methodologies. The quality of underlying data has always been entwined with the success of machine learning and AI deployments. The significant data intake by large language models accentuates the demand for a robust data foundation.
Establishing such a foundation demands meticulous data curation efforts, further compounded by the sensitivity of personally identifiable information (PII) and protected health information (PHI) potentially present in training data. Moreover, addressing bias in curated datasets used for fine-tuning is crucial. Although healthcare data standards have progressed, the sector still lags in structured, high-quality data availability.
Addressing these challenges mandates a strategic approach to data acquisition, growth, refinement, protection, and deployment. Building a contemporary enterprise data platform on the cloud, supported by a dependable set of data products, can pave the way for success.
Q: Your second recommendation centers on implementing appropriate controls. Could you elaborate on this principle?
A: The swift adoption of generative AI accentuates the need for healthcare entities to define a responsible AI mission and principles. This endeavor must be accompanied by the establishment of transparent governance structures to instill confidence in AI technologies.
It’s vital to incorporate responsible AI methodologies throughout, commencing with risk assessment controls during the design phase. These principles should emanate from leadership and translate into effective governance, encompassing risk management, and compliance with organizational standards, policies, applicable laws, and regulations.
Enhancing compliance with existing and forthcoming regulations, formulating risk mitigation policies, and operationalizing these through a risk management framework with regular monitoring are crucial elements. A proactive shift from reactive compliance to mature responsible AI capabilities is pivotal. These encompass governance, risk management, technology enablement, and training.
Prioritizing training and awareness and subsequently expanding to execution and compliance are integral steps in ensuring a responsible AI ecosystem.
Q: Your third piece of advice pertains to aligning people with the technology. Could you elaborate on this concept’s significance in the context of generative AI?
A: Generative AI applications in healthcare necessitate human guidance based on experience, perception, and expertise. Processes need refinement to accommodate generative AI capabilities while elevating the role of human workers.
Healthcare organizations should implement training programs to empower personnel, spanning clinicians to administrative staff, to adapt to AI advancements. As AI tasks become more complex and judgment-oriented, training programs are essential. For instance, physicians interpreting health data require deeper technical insights into AI models to confidently utilize them as “copilots.”
In healthcare domains with significant generative AI potential, an effective strategy involves deconstructing existing roles into core task components. Subsequently, assessing the influence of generative AI on each task, be it full automation, augmentation, or minimal impact, is essential.
Notably, generative AI can alleviate the documentation burden on healthcare workers. Rethinking work processes and facilitating technological adaptability will be instrumental in realizing generative AI’s full potential.
Q: Envisioning the role of generative AI in healthcare over the next five years, what are your thoughts?
A: We stand at a pivotal juncture. For years, AI and foundation models have quietly revolutionized our perception of machine intelligence. We now embark on an exciting era that will reshape information accessibility, clinician-patient interactions, and healthcare management.
Accenture’s research reveals that a significant majority of healthcare provider executives recognize the transformative potential of generative AI. This technology promises to reshape healthcare delivery, enhancing access, experiences, and outcomes. Indeed, generative AI holds the power to usher in an era of enterprise intelligence that extends far beyond what we have witnessed thus far.
