There are three tiers on the scale of knowledge and ignorance: the “known knowns,” or the problems you’re aware of and know how to solve; the “known unknowns,” or the problems you’re aware of but the solution may be unclear; and the “unknown unknowns,” or the problems you simply can’t anticipate.
We all know which one is the most dangerous — to both organizations and individuals. Automation approaches and technologies have evolved over the past decade or so to handle these categories of knowledge differently. One basic technology, robotic process automation (RPA), can handle the known knowns quite well, while machine learning can help respond to and analyze the known unknowns more effectively.
But people, with all their intuition for handling complexity, are best at identifying the unknown unknowns — the unexpected outliers. Demands on automation tools and technology are only going to get more complex. It’s going to revolve around solving for the known unknowns and unknown unknowns. Depending too heavily on RPA to respond to unknowns can lead to a vicious cycle: the outlier comes down the chain, the RPA fails to identify it and respond appropriately, and the system breaks.
Here at AKASA, we’ve helped our healthcare customers in revenue cycle management (RCM) identify and manage unknowns through our expert-in-the-loop solution. It’s an antifragile system that leads instead to a virtuous cycle of converting the unknown to known, in which the human expert handles the outlier, the AI learns and adapts from observing the human expert, and then the AI manages the former outlier appropriately in the future.
From a design standpoint, building an expert-in-the-loop approach to automation that is capable of adapting to unknowns presents an enormous engineering challenge. It requires untangling four distinct problems — all of which will likely exist for any automation platform that integrates AI with human experts. Here’s how to solve those problems:
1. Balancing AI Confidence
Engineering the AI to accurately gauge its ability to recognize and handle difficult or unknown tasks is paramount. If the AI doesn’t have enough confidence to manage such tasks, it will bump too many of them to human experts, greatly reducing system efficiency. An overconfident AI, on the other hand, will attempt to tackle unknown outliers it can’t handle, breaking the system or incorrectly handling the task.
2. Triaging Outliers Efficiently
Healthcare is all about the outliers and edge cases that we don’t see coming. The AI must be designed to swiftly triage tasks outside its capability to human experts — without interrupting its regular workflow. Automation systems that rely on basic RPA technology would break in these instances and related work would come to a halt until the bot has been reprogrammed. But AI-based automation systems with the built-in ability to triage outliers to human experts are more flexible. That flexibility is the key to building an autonomous solution that can handle unknowns and work without interruption.
3. Giving Human Experts X-Ray Vision Into The AI’s Brain
Through an intuitive head-up display, human experts in the loop must have direct and ready oversight of the entire AI process, allowing them to observe, assess and either approve or reject the AI’s proposed actions on a very granular level. All in a time frame of minutes or less. Without such oversight capability, resolution of unknown outliers could take hours or days.
4. Enabling Continuous Learning
The AI must be designed to continuously observe, learn and adapt, shadowing human experts in a kind of running apprenticeship. This learning and adaptation must continue indefinitely as the AI becomes more and more capable of handling complex tasks, persistently transforming the outlier into the ordinary — the unknown into the known. Today’s outliers become tomorrow’s built-in solutions.
“You don’t know what you don’t know” is only the starting condition. With close analysis and insight, you can identify what you don’t know — and what it’s costing you and your customers. Then it’s one step closer to becoming the known known — or knowledge you can bank on. Effective, carefully designed automation helps us better find and illuminate the unknown unknowns.