MORGANTOWN — For some people, the term “artificial intelligence” conjures up images of science fiction war games or assassin droids in Star Wars.
But those images don’t tell us much about what AI really is, and they definitely don’t help when it comes to the public’s misconception of AI.
“The concept of artificial intelligence often raises concerns because there is a fear that it’s fake or will replace human thinking,” said Sally Embrey, vice president of public health and medical technologies at DataRobot, which recently opened an office in Morgantown. “But the truth is that AI can’t ever fully replace human decision-making.”
Clearly, it’s the word “artificial” that throws people. It seems to say that results or predictions are somehow out of our hands as humans.
But, it’s humans who guide the intelligence by articulating how the data is used. In fact, we’re now seeing more of the term “augmented intelligence” rather than artificial intelligence. After all, there’s nothing artificial about the programmers, or coders, who write the models. They are brilliant people who set up the simulations and define the parameters. They get the system going and essentially say, “Now run with it.”
“What AI can do is allow people to do superhuman things by increasing an individual’s capabilities to process data and build predictive models,” Embrey said. “This is done by augmenting artificial intelligence and machine learning with human expertise. When done properly, the results are unique, data-driven insights. This is why we call it augmented intelligence. It makes humans greater and better at making data-driven decisions.”
Since DataRobot’s recent opening, and with its resulting partnership with West Virginia University, North Central West Virginia is poised to become a national research center that utilizes the power of AI to tackle the challenges of rural health care.
But bringing AI into health care has its own set of challenges. Scientists are dealing with human behavior, and there is nothing more unpredictable than humans. Data fed into a program that’s seeking to find, say, triggers for migraine headaches, is often “fuzzy.” A computer simulation works with the data it’s given, so if the data isn’t clear, results may also be a bit fuzzy.
“AI is already being used effectively to help treat many diseases, but there are still challenges,” Embrey said. “A key factor moving forward is not so much having enough data as it is connecting the data we have in a way that we can make sense of it. We live in a world of data overload, so it’s hard to make insights without technology to help spot patterns and insights.”
More than 20 years ago, IBM introduced Watson, the AI supercomputer that was going to change the world, including the field of health care. After Watson beat Jeopardy champion Ken Jennings, the future seemed clear. But things turned out differently.
“Watson was going to revolutionize medicine and everything else, but it really didn’t do it very well,” said Dr. Clay Marsh, vice president and executive dean for health sciences at WVU. “And the reason why is because some of the information you need, and some of the way you categorize it, and some of the outcomes that you measure, are fuzzier and therefore the models have a lot more trouble.”
In other words, humans. We’re just so fuzzy. And again, unpredictable. But, the good news is AI is finding its legs in health care.
“Advanced computers are already performing monotonous, repetitive tasks by scanning volumes of data to find patterns and anomalies,” said Jonathan Manis, senior vice president and chief information officer of Christus Health, which is headquartered in San Antonio, Texas. “Artificial intelligence assists in the education, screening, triage and treatment of patients. We are only just now beginning to understand the potential of this powerful tool for the health and wellness industry.”
As it stands, AI is useful for detecting anomalies that may indicate cancer or other diseases, but it’s not at the point where it can develop patient treatment plans. Scientists and clinicians will still be on the front lines of determining how to treat different illnesses and diseases. But patients will benefit from AI in other ways as well.
“Computers capable of learning and adapting to a particular situation or circumstance will help lower health care costs, enhance the service experience and greatly improve clinical outcomes,” Manis said.
Because AI can sift through data faster than humans, a computer model, or simulation, can provide clues to health problems in seconds, rather than months or years.
“AI seeks to learn what are the most important parts [of the data], which may give you some idea how to feed new information,” Marsh said. “That’s how you help the model learn, and then eventually the model starts to learn, and it teaches you which of those different types of information might be the most important. Then you can set up that information as a kind of sensor system – like a canary in the mine.
“We’re working with one [computer simulation] that’s run 3 billion different models in the last 18 months,” Marsh continued. “And those models are fitting the data differently. When you get a model that says yes, that’s more of what we’re seeing, then we start with that model and work from there.”
Regardless of the speed of AI, or its accuracy in organizing data, people are still needed to interpret the results. People have to guide the system, pay attention to patterns, adjust the input, and be ready to take steps when answers are clear.
“So we’re making progress that way,” Marsh said. “We want to incorporate information in a way where we are not only using that logic and data, but we’re also using intuition, branching information and visual cues.
“Sometimes you find that you get a bunch of data, and it’s 100 percent this is the answer, it’s a straight line, and usually you can see that,” Marsh continued. “But sometimes these models are saying, hey, idiot, look, the house is on fire — get out of the house.”
From the patient’s perspective, AI can make medical jargon easier to understand, and therefore help patients understand their medical treatments.
“What I think is probably most important is that AI can be used to help patients understand the care they are receiving,” Embrey said. “There is a field of AI known as Explainable AI, which can, for example, be used to simplify medical text to easily readable information. This ‘explainability’ increases the likelihood of a patient being able to be proactive about seeking treatment or taking their medication.”
To Marsh, it’s critical to keep AI available to everyone. “We can democratize what we’re doing, share it, and continue to make it easier and better and smarter,” said Marsh. “AI can give people better service, more precise approaches, and [it helps to] avoid errors and complications. That’s continuous improvement.”
West Virginians use AI in their health care when they track their diets with smartphones or if they use an app that instantly tells them their heart rate. AI is used throughout doctors’ offices and hospitals.
“To me, the way I think about AI is not like separate research,” Marsh said. “It is part of our world — it is what drives everything. Our whole world is informed by AI.”