At one of my recent talks in New York about AI in the supply chain, one of the key questions that came up was “Are you talking about robots?”
You see, AI has been romanticized into this abstract term that conjures images of walking robots doing your household chores while you just sit back and relax.
But what does it really mean, and where did the term actually come from? Artificial intelligence encompasses the new paradigm of machine learning and big data processes that enable you to get predictive insights from a combination of historical amounts of preexisting data processes and real-time observations. To get to true AI, you need to train large amounts of data sets (both historical and real-time), achieve some baseline, enable deep learning with incremental information, and begin to uncover predictive value.
AI typically works in tandem with the Internet of Things (IOT), which includes devices like wearables and connected home gadgets. Simple put, IoT collects the information, but AI is the engine that will power analytics and decision-making from that information.
IoT connects disparate devices, such as wearables, and can scale to connect a nearly unlimited number of devices, continuously streaming data. AI processes data, makes inferences about this data, and ultimately enables recommendations in real time.
Let’s look at some examples from the insurance industry
When I was at Humana, around 2012, one of the projects we worked on was with seniors (65+) living in their own homes. We wanted to understand how to reduce the incidence of falls and predict the likelihood of a need for emergency services. We needed to do this in real time so we could act beforehand, improving the seniors’ health status and saving costs. Armed with pre-existing claims data, we needed to understand the baseline — e.g., the typical activities that occur in the home. Here IoT devices came into play through the use of mobile sensors.
With the permission of the seniors, we installed multiple mobile sensors in the home, particularly in areas such as the kitchen, bathrooms, and living room. These sensors started by collecting a baseline of biometric data in the rooms over a period of time, then stored the data in the cloud in real time. To facilitate deep learning (which is a form of AI), pre-existing data from previous insurance claims was analyzed side by side with the real-time data. This made it easy to spot exceptions and to act on them and provided the insights to predict the probability of an emergency event before it occurred.
A second example, also from the insurance industry, took place around 2006. We were creating the early warning system for a technology assessment. Medical devices are expensive and are not always effective. So typically, a cost-benefit analysis is necessary to justify their use relative to other options. For example, weight-loss surgery may cost $10,000, but it’s still experimental, especially when you consider the patient and their health status. It has not achieved an efficacy relative to its cost.
Ultimately, we needed to predict that a patient will seek to explore the procedure relative to alternative options, understand how effective such a procedure will be based on health status, and know the provider pricing benchmarks for efficient contracting. Armed with millions of claims data (historical) on such procedures, some of which included the data from wearables (pedometers, fitness trackers data), combined with real-time electronic adjudication (real-time payments), and provider office visits observations, we were able to begin to automate the recommendation process for alternatives to weight loss surgery
Many of these advancements in AI come from an Artificial Neural Network (or ANN). Inspired by the human brain, it loosely models the way a biological brain solves problems, with systems that can self-learn and train themselves, rather than responding to programming. With a neural network, algorithms are trained by humans first. Over time, the algorithms begin to making their own assumptions, relying less on human trainers, and solving complex problems.
In the end, IoT is not enough. There has to be an added intelligence, an AI, that seeks to solve problems and not just process data or provide a dashboard. There has to be data and action.