The Continued Evolution of Artificial Intelligence
The field of artificial intelligence is broad and rapidly evolving by nature — a fact that is, in many ways, the technology’s greatest asset. We rely on artificial intelligence to learn, analyze and inform with greater accuracy and efficiency than a human ever could, which makes keeping pace with the latest breakthroughs, application areas and segments a challenge for even the most seasoned of data scientists.
One thing that is universally clear is that artificial intelligence has enormous potential when it comes to healthcare. And while AI and machine learning have long been part of the healthcare conversation, a newer type of AI implementation is gaining recognition as a means to revolutionize medicine — from the discovery of promising new therapies to the administration of higher quality, more efficient care.
Generative AI approaches present great opportunities to improve clinical workflows, taking traditional pattern recognition a step further to actually generate content that puts those patterns into actionable context. While the full impact of generative AI in healthcare has yet to be seen, we’re taking what we know so far to explore some of the exciting possibilities.
General vs. Generative AI
When people hear terms like artificial intelligence and machine learning, it can conjure images more closely aligned with artificial general intelligence (AGI) — the overarching term essentially describing the ability of a machine to learn and perform any variety of human tasks. In other words, it’s the branch of AI supporting the notion that machines can — and will — one day replace human beings altogether.
While AGI is thought to be a goal of some AI research, its only real application remains in science fiction (think Terminator 3: Skynet Takes Over). In reality, generative AI — and most segments of AI, for that matter — are making advancements meant to augment human capabilities rather than replace them. And with the healthcare industry more focused than ever on the integration and interoperability of people, processes and data, the possibilities of generative models are myriad.
What Is Generative AI?
Introduced in 2014 by notable AI researcher and then-University of Montreal Ph.D. student Ian Goodfellow, generative AI models are systems that produce “representations of the real world,” such as sounds and images. The thinking is that the more a machine can actually learn about the real world, the better it can understand and interpret what it perceives.
To achieve this, Goodfellow developed what are known as generative adversarial networks (GANs) — sets of two neural networks that essentially compete against each other to discern real from fake. One model might be programmed to generate image examples, while another will classify whether the images are real or came from the generator. The goal over time is for the generator to produce more realistic images while the discriminator gets better at distinguishing which images are real and which are fake.
Generative AI in the Healthcare Setting
In the context of image generation and analysis, the potential of generative AI in healthcare is extraordinary. Whereas humans have traditionally generated the neural networks for machines to use as a basis, generative AI enables machines to augment those models by actually generating their own. The ability to create new models — actual content that can be progressively learned from — means everything from better, earlier identification of potential malignancy to more effective treatment plans.
Take, for example, diabetic retinopathy, a condition for which early detection is key to preventing vision loss. Using intelligence from the millions of diabetic and other retinopathy images the machine has scanned in the past, generative AI enables the machine to not only give a pattern-based hypothesis as to the likely diagnosis, but to also interpret the scan and generate some type of content that helps to inform the physician’s next steps. This is where we really start to see how generative AI can heavily augment the clinician workflow.
The Role of Generative AI in Population Health Management
One of the most important applications we see for generative AI is in optimizing clinical workflows to drive value across the healthcare ecosystem. From enabling more precise diagnoses and care plans to informing optimal staffing rotations based on patient load and mix, generative models have the potential to fill significant gaps in the delivery, consumption and overall efficiency of care.
While we are already seeing the adoption of generative AI models in some of these narrow use cases, we expect to see a much more prolific impact in the next 5-10 years. At HMS, we are excited to be on the cutting edge of new AI models and emerging technologies that have the potential to reduce healthcare costs and improve health outcomes.