Healthcare is an industry that will benefit greatly from Artificial Intelligence (AI), Machine Learning, and Deep Learning technologies. While there have been some early adopters of Artificial Intelligence (AI) in healthcare, it is still an under-utilized technology when it comes to helping deliver better patient care. Despite its futuristic sounding name, we’re at the point where machines can process more information and detect more patterns than humans and the platforms and solutions that deliver this massive compute power are, in many cases, ready for primetime.
Though many people might still think of Skynet and Terminator when they hear this phrase, there is plenty of good that can come from these Artificial Intelligence-based technologies. Examples include the integration of smart devices and personal digital assistants into healthcare, integrating environmental and social data into applications, and to bring efficiencies to other aspects of patient care.
However, it’s not surprising that AI adoption in healthcare has been slow because of one very important factor – timing. We can all agree that timing plays a major factor in all aspects of lives. For example, seven years ago, my now wife had a job lined up in San Francisco that fell through at the last minute and she was transferred to Washington, D.C., where I had just landed a job thanks to a friend. Within two weeks, my future wife and I met and, as they say, the rest is history.
The same sort of timing applies in the business world. It’s clear that multiple factors have to align in order for the “timing to be right”. Fortunately, the stars have aligned for the AI discussion. For the sake of brevity, I’m using AI to describe both Machine & Deep Learning.
The idea of AI has been around since the mid-20th century; however, until recently, it’s been more fiction than reality. So what has changed recently, that’s leading to the conversion from AI’s fiction into reality? The success of AI is reliant upon three major components: the neural network, the processor and the data. As the concept of a neural network dates back to as early as 1943, this concept is by no means new. For clarity: A neural network is a system set up to mimic the operations of neurons in the human brain. The designs and specifics have matured since then, but the overall concept remains. Neural networks needed processing and the data to catch up in order for AI to take off.
The incredible advancements we are witnessing today in healthcare using AI, in everything from verifying insurance to virtual assistants that assist in patient care to improving diagnoses and surgery, are because those two other factors have matured enough to change reality. To begin, in order to properly utilize these neural networks, a large amount of horsepower is required in the processors. Before today’s cloud computing came to fruition, it wasn’t easy to go out and acquire 10,000 cores just to train your neural network. Even with the ability to theoretically rent “unlimited cores,” there is still a potentially prohibitive price tag associated with that strategy. Enter NVIDIA and their GPU’s. By parallelizing the algorithms and leveraging GPU’s, these computationally intensive neural networks are no longer unobtainable, and better yet, they have become affordable. In fact, NVIDIA’s new DGX-1 platform with its new Volta cards packs the same punch as over 1400 x86 servers in a tiny 3U footprint for under $200K.
The final factor in AI’s success is our ability to harness and store a massive amount of data, in a cost-effective way. These neural networks need to be fed copious amounts of data in order to train themselves accurately. This is conceptually similar to training a baby or a puppy: The more repetition of a similar word, object or command, the better the training. In the 1950s, it would have cost over $1M for a single MEGABYTE. My current iPhone has 64,000X that storage—and I’m cheap and wouldn’t buy the 128 GB phone! We’ve all heard the crazy statistics around data production, i.e., in 2020 we will produce 44X the amount we produced in 2009. This proliferation of data is only going to make these neural networks stronger and more accurate, which is great news for the success of neural networks and AI for healthcare providers and payers.
Without the advancements in GPU’s and without the ability to store and manage the amount of data we are generating, the AI craze would never have taken off.
AI is providing every organization – including healthcare organizations – the ability to define their own strategy and execute against it without breaking the budget.
