Healthcare AI — These ARE the Droids You’re Looking For

by Sean Lane, CEO of CrossChx

Artificial intelligence has been a pretty pervasive buzzword in the healthcare arena for the past few years.

I remember going to my first HIMSS in 2012 and seeing “big data,” “advanced analytics,” “predictive analytics” and “actionable intelligence,” throughout the convention center in Las Vegas, NV. By 2016, most of the cool, hip, early adopters had switched to new buzzwords and catchphrases. This time, it was “artificial intelligence” and “machine learning.” You would think someone with my background would be thrilled to see the words of my tradecraft prominently displayed through the expanse of the exhibition halls. Instead, it was frustrating because like “big data” and “advanced analytics,” it was largely misunderstood.

I try to look at the bright side of most things. It’s a task that requires a lot of discipline, but it helps in the endless struggle to stay optimistic, which I think could be the single biggest determinant of success within our control. As I thought about the bright side of this 3 to 4 year cycle of buzzword mania, I reflected on what was actually good about the whole thing.

I realized then that even though people didn’t really know what these buzzwords were, they knew they needed them.

Over the years, demand grew drastically. People wanted these new and buzz-worthy things, and guess what? There weren’t enough companies that manufactured these products; so new companies emerged. It was great to see the vast amount of advancements that were being made. People started to understand these technologies and experts who really knew what they were talking about became kings

It was an industry building demand for a new market.

Now, it’s a whole new year and a whole new set of buzzwords. AI and machine learning are nothing new. In fact, they’re decades old, and any good engineer can “do” machine learning and AI. The secret is not in the ability to set up the open source tools or to understand what a linear regression is. The secret is to figure out how it’s going to fit into life.

AI has entered our lives in many other ways. Siri started it all. Then you had Cortana, Alexa, and Google Assistant. These products and events are key moments along the path of enterprise adoption. Laptops, Blackberries, iPhones, and tablets, all pushed their way into the hands of the enterprise by way of the consumer market.

There will be a few types of AI, and they are going to be distinguished by their user interface, not necessarily by their capabilities. The AI we’re most familiar with, Siri and Alexa, primarily operate through voice UI (although they have some text UI too). I think the most powerful AI in the enterprise will be through the user interfaces it currently uses. Which applications are being used frequently? What about media and communication tools? I think the most relevant AI in the enterprise, especially in healthcare, will use email, chat, SMS, communication apps, and maybe some voice as its UI. I think it is incredibly limiting to only allow your AI to make an impact inside a single application. That seems to be a high friction approach to AI proliferation. You’re basically saying organizations can only get the benefit of AI if and when they use your application. How can you improve and optimize human efficiency if your AI can’t work like humans work?

The definition of AI becomes important in the implementation. Not because there’s really a right or wrong answer, per se, but because it matters how you define it, when you build it, and bring it to life. I don’t necessarily refute any legitimate definition of AI because there are lots of good ways to describe it. I asked my CTO how he defined AI and he called back to the “Turing test of AI.” He defined it as a machine that you can’t distinguish from a human. I like that definition. Again, it doesn’t make other definitions wrong, but I decided that’s how I want to define AI.

Defining AI as a machine that you can’t distinguish from a human puts you in an interesting position when it comes to implementation. It makes you think about it quite differently. For example, many people approach AI through intelligence amplification, or IA. Not necessarily a bad approach, and one we see most often when AI is used inside an application to help make the operator perform at a higher level. But how would you make AI if you wanted it to be as humanlike as possible? I don’t think it means making life-form style robots or even trying to recreate a voice interface to be humanoid. I think it means to have AI work like humans work, only better.

I’ve been kicking around what this could mean for Olive, our AI at CrossChx, for a while now. I think we have figured out a compelling way to bring Olive to life inside the healthcare enterprise by taking this approach. If you were hiring a new human, you’d expect that human to use the tools you have provided for them to accomplish their daily tasks. In healthcare, for example, you’d expect them to use the EMR, email, phone, chat, the patient portal, the revenue cycle tools you’ve purchased, and the list goes on. When you hire someone, you don’t expect them to show up with their own applications to perform their duties. So why expect your AI to do the same?

Why not expect your AI to adopt the tools that are already in place and use them to perform their duties just like any other human?

That’s the approach we’re taking with Olive. We decided that Olive’s UI would be the digital media that the enterprise worker uses. Things like email, chat, EMR’s, patient portals, etc. We decided that if a hospital or other healthcare entity on-boarded Olive like they do a human, gave her an account for their IT systems and created an email address for her, we could program Olive to use those tools to perform tasks that humans normally operate.

For example, in hospitals, humans often have to check the insurance eligibility for the patients that are scheduled to come in for appointments. This task takes time, is incredibly mechanical, and it can — and should — be done by a machine instead. There have certainly been technologies in the form of clearinghouses, software applications, etc. that have made this faster and easier. However, even the best workflow tools are still just a tool for humans to use. They still require a human to interact with them by reading, clicking, typing etc. We propose that Olive, using whatever tools the hospital has, could do this task faster than a human ever could. Olive can log into the EMR, view the schedule, pull up the record of the patients scheduled to come in, log-in to the insurance websites, check their eligibility and copay, and then write that information back in the EMR for the human to see when the patient arrives.

Staff can then be freed up to focus on patient care and harder tasks that require the wonder of the human brain.

Checking insurance eligibility of patients is only one example of the type of tasks Olive could automate. Olive can perform multiple functions including, eligibility, prior authorizations, scheduling and more. Olive is also the perfect employee. She never sleeps, never takes vacation, never celebrates a holiday, never gets sick, never misses a task, and perfectly documents everything she does. Olive sends an email to her manager everyday, at the close of business, describing everything she did.

Olive also has the benefit of machine learning. Machine learning can be defined in several ways, but for the sake of Olive let’s think of it like this: machine learning lets Olive look at large data sets and find correlations that a human couldn’t find alone. Once a single instance of Olive learns it, all instances of Olive learn it, meaning Olive learns globally.

Olive and other AI applications bode well for the adoption we are about to see in the enterprise and I predict a tidal wave. People will take their training and indoctrination from the consumer market and start applying it to everyday enterprise applications. Selling artificial intelligence will no longer be something of science fiction, it will be expected and relatively easy to reason about. It will become a normal occurrence for C-levels and directors in healthcare to see “Artificial Intelligence Meeting” on their calendar in the next 12–24 months.

That’s the world we’re in now. Healthcare has been talking about AI and ML for long enough to make it to HIMSS. Engineers generally know how to start building it. There are great tools available on the market that are free and powerful to get started. Few have figured how to make it useful and deployable to a customer, but most people are really starting to want it.

I believe artificial intelligence will have an overall positive impact on healthcare and that AI will scale humans like never before.

Unlike others, I do not believe the human component will be rendered obsolete. In fact, I believe that in the future AI and the human worker will co-evolve, enabling all of us to work on harder problems.

I believe Olive will be a big part of that future.