Our team has covered artificial intelligence (AI) in depth for a few years now, but we've noticed that despite the rapid escalation of research and deployments, we still have a definition problem. Whether in our discussions with members or even in our internal meetings, we always need to make sure everyone is on the same page about terms—and one question in particular comes up frequently: "Are we talking here about AI or robotic process automation (RPA)?"
June 17 webinar: Why so many AI models are biased—how to stop the problem
It's a fair question to ask, but it's the wrong question—and one that can unintentionally obscure your organization's business needs.
Before we explain why, let's (once again) settle on terms:
- Merriam-Webster defines AI as "the capability of a machine to imitate intelligent human behavior"; and
- The Association for Intelligent Information Management (AIIM) defines RPA as "software tools that partially or fully automate human activities that are manual, rule-based, and repetitive."
So why is asking, "Is this AI or RPA?" wrong? It's because the question assumes two things: that AI is a specific technology or function, and that RPA and AI are somehow mutually exclusive. What's important to keep in mind is that AI is really a concept—a way of thinking about computers (software, machines, bots, etc.) performing actions as we would. There is no one AI technology. There is no singular AI product that comes out of a (black) box. AI has various applications and diverse underlying computational methods, some of which we've outlined below.
What this means is that RPA isn't a technology that is distinct from AI, but rather RPA is a form of AI. RPA is an example of how we can program computers to mimic our actions. We already use RPA today to handle repetitive, rules-based, high-volume tasks that have defined inputs and outputs, such as:
- Claims administration and management;
- Appointment scheduling;
- Physician credentialing;
- Insurance eligibility checks or prior authorizations;
- Regulatory compliance and reporting; and
- Accounts receivable, billing, and other aspects of finance and revenue cycle.
RPA's main selling point is that it can enable clinicians and other staff to devote more time to higher-value work and services, while ensuring that automated processes are accurate, efficient, and lower cost—something which can also help combat our industry's burnout problem.
It's not AI or RPA—it's RPA or machine learning
However, RPA is distinct from another specific form of AI that has captured most of our industry's attention: machine learning.
- Merriam-Webster defines machine learning as "the process by which a computer is able to improve its own performance by continuously incorporating new data into an existing statistical model."
Industry stakeholders are fixated on machine learning because it moves the needle closer to our future visions of what a strong AI could become--a machine that can truly replicate the human mind in terms of processing and understanding information, solving problems, and, in some instances, acting with full autonomy. For some, machine learning can seem like something out of science fiction.
We suspect that when people ask us, "Is this AI or RPA?" what they likely mean is, "Is this machine learning or RPA?" This discussion is about differentiating advanced forms of AI from more rudimentary ones. Given its exciting potential, machine learning has so much mindshare that many people now simply equate the term AI with machine learning and use the two terms synonymously.
RPA, on the other hand, is not as sexy. After all, this technology is explicitly programmed to take the load off of us in handling monotonous, low-skill tasks that we don't want to do. RPA solutions won't learn how to play chess or Jeopardy. They do not require emotional intelligence, creativity, understanding, reasoning, or judgment, and we don't need them to. Of course, that's not stopping RPA vendors from getting creative by leveraging computer vision and machine learning capabilities to accelerate the development of RPA workflows. This hybrid approach can assist with the more manual, iterative approach of setting up processes (e.g., interview, script development, trials), and can also improve the long-term value of an RPA solution if it can be repurposed down the line.
The right question to ask about AI and RPA
What's important to understand is that while machine and deep learning capabilities bring a lot of new opportunities for health care organizations, they are not meant to solve all of our problems. It's easy to get caught up in the hype around machine learning, but for many organizations, RPA will be sufficient to achieve their business needs if it is fit for purpose.
So, after clarifying terms, what is the right question? It should be, "How can we use machine learning, RPA, or another form of AI to help achieve our business goals?" The value you get from RPA—just like any technology—depends on the specific use case and enterprise goal. Furthermore, RPA can serve as a steppingstone to more sophisticated AI solutions down the road, allowing organizations to capture "low-hanging fruits" fairly quickly and without some of the risks that comes from deploying more complex, but opaque, AI.
Learn more: Your guide to demystify health care IT jargon
Health care is full of acronyms and jargon—the world of health IT even more so. How does a data mart differ from an enterprise data warehouse? Do you know about FHIR? Can you describe an API?
Here, we have assembled a collection of the most frequently referenced health IT terms, including IT-related professional organizations, regulatory mandates, infrastructure components, concepts, and major IT topics.