The adoption of clear consensus standards for medical device technology is essential to the development, management, and use of safe and effective health technology. The incorporation of artificial intelligence (AI) and machine learning (ML) applications into medical devices holds tremendous promise to bring a new avenue for innovation and functionality to these tools.
Ultimately, AI/ML-enabled medical applications could provide new scalable solutions to the industry that may enhance the quality of care available to patients. However, as one thinks about the role of AI/ML in medical applications, several key questions arise: What is the foundational purpose of these algorithms, what additional risk do they present, and how should they change the expectations of the role of the end user?
AI/ML-enabled medical applications have the potential to change health technology and medicine in profound ways, but in this time of rapid and exciting change, it is important that digital solutions and AI/ML-enabled medical applications be evidence based, validated, actionable, and connected. As AAMI works to develop consensus standards for AI/ML-enabled medical applications, important foundational work has been done to define terms, management approaches, and strategies for risk mitigation.
One important term that warrants discussion is the use of the term “artificial” intelligence. Although almost universally used across the industry, I would propose that this word is probably neither appropriate nor truly representative of the technologies being developed or in use today. Artificial intelligence is defined by having the ability of an intelligent agent or machine to understand and/or learn any single intellectual task that a human being is capable of learning and executing. Our machines and devices today simply cannot achieve this. Instead, I would propose that the term “augmented” intelligence is more fitting and descriptive. The term “augmented” intelligence is an important conceptualization of AI that focuses on AI's assistive role, emphasizing that its design enhances human intelligence rather than replaces it.
One way to think about it is that artificial intelligence is a tool that produces an output, while augmented intelligence integrates human intelligence and machine-derived outputs to improve health. Sometimes a distinction is made between “general” artificial intelligence and “narrow” artificial intelligence, with the latter representing any algorithm designed to perform a single specified task. Almost every AI/ML-enabled medical application in development (of which I am aware) looks much more like a “narrow” AI tool than a “general” AI tool. Further, all of the AI/ML-enabled medical applications that have been deployed also fall into the category of narrow AI. That being said, until general artificial intelligence is within reach, narrow AI and the varied uses in healthcare and health technology can serve as an important means of scaling human capacity.
Despite being used near universally across the medical device industry, use of the word “artificial” in the term “artificial intelligence” is probably neither appropriate nor truly representative of the technologies being developed or in use today. Use of the word “augmented” is more fitting and descriptive; it is an important conceptualization of AI that focuses on its assistive role, emphasizing that AI's design enhances human intelligence rather than replaces it.
Given the limitations of currently available technologies, AI/ML-enabled medical applications are best optimized when they are designed to leverage human intelligence once deployed. It's worth pointing out that the rate-limiting factor to innovation and transformation is often not the medical device technology itself but the interface between the human and machine.
To leverage these emerging technologies to their fullest potential, we must ensure that both the performance of the AI systems and the actual deployment transform care so that we are able to provide equitable, inclusive, human and humane care. Machines and algorithms are not able today to do this without human input, design, validation, and deployment; therefore, the term “augmented” intelligence is likely a better fit than “artificial.”