Each week we find a new topic for our readers to learn about in our AI Education column.
It’s said that an ounce of prevention is worth a pound of cure. It might also be worth a whole lot of money.
Welcome to AI Education, where, this week, our topic will be preventive AI, which is less a technology in and of itself, but instead more of a paradigm shift in technology with provocative implications for humanity, particularly in realms like finance and health.
Let’s stick with the health metaphor for a moment, because we’ll place particular emphasis on preventive artificial intelligence’s health care applications in this piece. Think of some of the basic diagnostic and monitoring technology in a hospital setting like an intensive care unit or an emergency department. Patients whose illness or condition is of a certain acuity or severity are often placed on continuous monitors that keep track of blood pressure, pulse and blood oxygen saturation. When someone’s pulse drops or rises past certain thresholds, or their heart rhythm becomes dangerous and unstable, or their blood oxygen begins to drop, alarms alert medical staff—who sometimes come running with a defibrillator and a battery of cardiac drugs.
This technology, as with most of the technology we’ve used up until now, is reactive. It reads current conditions, and responds to changes in those conditions. Even our automated technologies tend to be reactive, like our thermostat that engages or disengages our heat or air conditioning when certain temperature thresholds are reached. They detect that something has happened or is already happening, and then respond.
Moving Beyond Reactive Technologies
Artificial intelligence has the power to be more than a reactive technology, because we’ve given AI the ability to ingest large volumes of data, understand the patterns within that data, and then make predictions and insights based on those patterns. With that power, we can design applications that don’t merely react and respond to what is already happening, but anticipate what might or what is likely to happen in the future, and then take proactive steps to manifest or prevent that result.
So instead of the alarms ringing when our heart goes into an unsafe rhythm, sending our care team running, AI might be able to tell a cardiologist that within the next few years, our heart has a high probability of ending up in that unsafe rhythm, enabling a life-saving proactive intervention like maintenance drugs or an implanted defibrillator and altogether avoiding a pricey emergency department visit or lengthy intensive care stay.
A proactive intervention is less expensive than a reactive intervention, and not just in terms of monetary cost, and not just in healthcare. This editor has worked in healthcare settings and has seen first-hand the morbid side-effects of emergent, life-saving care. CPR often causes broken ribs when done correctly. Improperly diagnosed and managed diseases—like diabetes—can lead to the loss of life and limbs. But the stakes are high also in finance, where a little fraud prevention is potentially a huge savings versus the cost of making depositors or investors whole. The whole idea is to stop the bad things before they even happen.
How Preventive AI Works
We’ve discussed machine learning, AI training and AI models in depth in AI Education, but the topic most closely related to preventive AI is predictive AI. In fact, one can think of preventive AI as predictive AI attached to an action or intervention mechanism, which may itself be governed by artificial intelligence technology. Predictive AI uses machine learning to ingest large volumes of information using various algorithms. Neural networks may be employed to help identify patterns within the data. Different techniques are used to classify and organize information and understand the relationships between different pieces of information. For example, linear regression algorithms help find correlations between different pieces of data.
By training on data, algorithms learn historical correlations between different variables that precede certain outcomes, and then are able identify those correlations as they occur in real time, which enables them to make predictions. As more data is collected over time, and more real-time predictions are made, a predictive model can “learn” from its own activity and refine itself, becoming more accurate and successful. The prediction is then used to inform automated decision-making, which may carry over to response actions.
Where Preventive AI Is Working
We’ve already talked a bit about healthcare, where predictive AI is being embraced to help predict diseases before symptoms appear. Let’s consider some of our more recent mass experience with disease, waves of coronavirus and influenza infections. These viruses were often active and communicable in individuals days, sometimes several days, before symptoms would appear. AI can also be used to help prevent, monitor and manage chronic health conditions, as well as to help provide aftercare and monitoring for patients returning home after a treatment or a hospital stay.
Preventive AI is also finding industrial applications in the realm of predictive maintenance. Whereas in the past maintenance either occurred on a fixed schedule (actually known as preventive maintenance) or in response to breakdowns and failures, preventive AI can be used to help anticipate failures and breakdowns so that maintenance can occur proactively before a failure occurs, minimizing down time. AI can use things like maintenance, operation and performance history, and real-time data from sensors on and within equipment, to help maintenance teams intervene only when necessary, but still theoretically keep ahead of equipment failures. Similarly, preventive AI is being deployed as part of autonomous vehicles and smart factories to help anticipate and avoid accidents and mechanical failures.
We’re seeing similar use cases emerge in financial services. Of course, we’ve already mentioned fraud prevention—there’s potential value to be found in technology that can detect and prevent fraud before financial transactions even settle. Preventive AI is also being used to help prevent other potentially negative events like loan defaults and compliance violations, where lenders and other financial businesses want to flag issues before they result in losses or regulatory action. Preventive AI is potentially a risk mitigator for these businesses, but it’s also potentially a timesaver for individuals and professionals managing finances, automating monitoring responsibilities and, if not always successfully predicting and moving proactively to deal with events, at least greatly reducing the response time to them.
