Artificial IntelligenceThe Promise of Assistive Intelligence, The Only Kind of AI Healthcare Needs

By Rohan D’Souza, Head of Product, KenSci

Introduction

Artificial Intelligence and Machine Learning have taken a traditionally conservative industry by storm. Every day, you hear about another exciting news story being discussed in and around the hallways of the world’s leading healthcare institutions.

AI has moved from buzzword to well-recognized industry term, one that executives believe will be a catalyst to propel their organizations into the age of digital transformation.

Why do so many believe that AI is going to dramatically change the way we deliver and manage care over the next decade?

1. Right Place

Healthcare is ripe for AI, thanks in part to the billions of dollars in investments made over the last two decades in systems of records (EMRs, CRMs, ERPs, etc.). With the implementation of these transactional systems, most healthcare organizations are realizing that they are data rich but often information poor. Normalizing, sifting, cleansing and presenting insights from these assets is a dual challenge.

• First, it requires both a deep understanding of the nuances of data
• Second, it demands a technical artistry that can set aside the complexities of data in ways that enable business owners to perceive the underlying structure and make sound, data-driven decisions.

2. Right Time

Enough has been said about how the U.S. spends nearly 18% of its GDP on healthcare — with a projected trajectory that is clearly unsustainable. But not enough has been said about how scary the future looks when compared to inflation, reimbursement rates, and drug prices. With the push to value-based care, health systems continue to be pressured to provide services to their communities at sustainable margins. While most of the “sexy” AI stories have revolved around predictions, we see tremendous value in applying cutting-edge ML techniques (such as unsupervised learning) to identify outlier patterns to data. Our goal is to find the signals amid the noise in variation across healthcare settings — thus providing business owners with visibility and insight into their organization that they have never enjoyed before.

While it’s easy to get caught up in the chest-pounding act of model metrics like AUCs, p-value, c-stat, or any of the other statistics data scientists like to brag about, we’ve come to learn that sticking to a few core principles is a winning recipe for a successful AI-powered enterprise.

Lovingly, we refer to this as CUTA (pronounced cute-ah)

 Consumable

We believe that any insight generated by an ML system should be easily consumable. These insights should be easy to integrate, minimally invasive, and delivered to end-users in whatever system is most comfortable for them — whether that’s their EMR or a fancy dashboard.

 Understandable

We believe that any insight generated by an ML system should be understandable (transparent) during the process of data input, throughout its interpretability, and through output to the user. Our passion for explainable AI has made us one of the leading experts in the field of localized interpretable AI and its importance for life-and-death decisions.

 Trustable

We believe that any insight delivered to our customers should be trustable and aligned to our mission of ensuring that our ML-generated insights are accurate, consistent, and useful.

 Actionable

We believe that any insights generated by an ML system can be mapped to an action that is concise yet sufficiently comprehensive, and with clear and measurable impact.

In the end, if you’re looking for Assistive Intelligence for your AI solutions, make sure you are doubling down on what we have learned during the last two years: That delivering on ROI requires a joint agreement on the baselines, the suggested intervention or process improvement, and, finally, a constant need to measure the output of our variation analysis, and to do so as soon as the problem has been defined.

This post has been sponsored by House Communications