Artificial Intelligence is making the leap from much-hyped “trend” for healthcare technology to more widespread adoption. Here’s what A.I. is — and isn’t — and how health plans are proving its value.

Artificial Intelligence for healthcare has come a long way since 2017 when IBM Watson Health, the A.I. supercomputer fell short of its high expectations. A.I. is a powerful force in advanced healthcare technology and is poised to disrupt the industry. Due to several converging factors in 2020, adoption accelerated. And with the release of rapidly-advancing generative A.I. tools in 2023, the hype has reached a fever pitch. Let’s explore what A.I. is — and isn’t — and why it’s here to stay despite current limitations. But first, a few definitions.

Defining Artificial Intelligence

While varied stakeholders have found value in applications of artificial intelligence for healthcare, confusion remains about what it is and isn’t. When the industry uses buzzwords like “A.I.” and “machine-learning” to describe product functionality, know these words are easily misunderstood and thus, often used incorrectly. Let’s level-set with some working definitions of terms you might encounter when evaluating advanced healthcare technology.

Artificial Intelligence: Intelligence applied to a system with the goal of mirroring human logic and decision-making. A.I. is utilized for the purpose of successful knowledge acquisition and application, which it prioritizes over accuracy. A.I. simulates intelligence (the application of knowledge). It includes many subcategories and is often separated into three types: narrow, general and super.

Narrow Artificial Intelligence: Created specifically for a single task or to solve a single problem. Almost all applications of A.I. in use today are of this type.

General Artificial Intelligence: A type of broad and adaptable A.I. that can think and function more like humans. Advancements in neural networks and large language models may offer a path to this reality, though it is not currently possible to achieve. However, A.I.-powered assistants like Siri and NLP-processing tools like ChatGPT are sometimes put into this category.

Super Artificial Intelligence: A theoretical type of A.I. that is imagined to exceed human cognition significantly. It should emerge from the exponential growth of A.I. algorithms self-learning. Does not currently exist.

Machine Learning: An application of A.I. that allows a system to learn on its own. ML learns from data, and it aims to increase accuracy (success is a lesser concern). ML simulates knowledge. Sources include both supervised and unsupervised models.

Supervised Models: Algorithms designed to learn by example, based on labeled datasets that provide an answer key that the algorithm can use to evaluate its accuracy on training data. The term “supervised” learning originates from the idea that training this type of algorithm is like having a teacher supervise the whole process. When training a supervised learning algorithm, the training data will consist of inputs paired with the correct outputs.

Unsupervised Models: Machine learning technique that finds and analyzes hidden patterns in “raw” or unlabeled data. By ignoring labels altogether, a model using unsupervised learning can infer subtle, complex relationships between unsorted data that semi-supervised learning (where some data is labeled as a reference) would miss. And do so without the time and costs needed for supervised learning (where all data is labeled).

Data Analytics: Process of examining data sets in order to draw conclusions about the information they contain, increasingly with the aid of specialized systems and software. Data analytics technologies and techniques are widely used in commercial industries to enable organizations to make more informed business decisions and by scientists and researchers to verify or disprove scientific models, theories and hypotheses.

Predictive Analytics: Historical data that has been collected is utilized to try and predict behavior/outcomes (often called Data Science). To analyze data, it is routed into a report, at which point humans or artificial intelligence apply multiple factors to make predictions about expected outcomes. Though based on decades-old technology, “predictive analytics” often implies that a machine has performed the analysis and offered a prediction (rather than a human).

Neural networks: Seeks to simulate human brain processing, which is facilitated by networks of neurons. At its simplest, a neural network processes information in three layers: 1. Input layer where data enters the system, 2. Hidden layer where data is processed, and 3. Output layer where the system decides what to do with the information.

Deep learning: Allows for increasing numbers of layers through which data passes, where each layer of nodes trains on a distinct set of features based on the previous layer’s output. The further you advance through the layers, the more complex the features your nodes can recognize, because they aggregate and recombine features from the previous layer.

Overcoming Concerns About Artificial Intelligence for Healthcare

Though experts have long predicted that artificial intelligence for healthcare would take hold, adoption has progressed slowly. And no wonder. The industry has been historically hesitant in its pursuit of advanced healthcare technology. And for clinical care applications, particularly, providers express discomfort with “black box” A.I.

Forbes points out, “As humans, we must be able to fully understand how decisions are being made so that we can trust the decisions of AI systems. The lack of explainability and trust hampers our ability to fully trust AI systems.”

In addition to explainability issues, lack of broad access to healthcare data stymies effectiveness of A.I. Data silos and patient data privacy concerns both limit access that could push the technology forward. But federated learning could help overcome this barrier. It’s a privacy-focused approach to machine learning that allows companies to collaboratively form more representative data sets without sharing raw data. Once-generic models get smarter over time through decentralized data and decentralized compute power.

This approach could also help overcome another perennial concern about A.I.: model bias. A.I. systems learn to make decisions based on training data, which can include biased human decisions and amplify historical or social inequities. It’s a complex problem with life-and-death consequences in using A.I. for healthcare. But leveraging larger, more varied data sets, increasing transparency of processes, improving awareness of potential bias in A.I. outputs, and continuing to augment machine decisions with human expertise will mitigate this issue.

Making the Case for A.I. in Healthcare

Early adopters in the healthcare payer sector understand the benefits and risks associated with A.I. all too well. And skepticism of vendor claims of A.I. is high (and rightly so). However, the value of artificial intelligence in healthcare has increased significantly due to three primary drivers.

First, A.I. compute has been doubling every three and a half months making applications faster and cheaper. Now, large image classification systems are training much faster, from three hours down to 88 seconds. Progress on natural language processing (NLP) classification tasks is also “remarkably rapid.”

Second, the novel coronavirus pandemic accentuated the foundational weaknesses in the healthcare system. All at once, various stakeholders around the industry sought A.I. applications to address the resulting challenges. They also found value in collaborating to achieve the data access, sharing and quality needed to power these tools. Solutions for drug discovery and disease prediction emerged at a record pace. And the data and research breakthroughs promise to continue to push the industry forward.

“Generative A.I. is that breakthrough technology. This is not something that will just disappear, make a lot of promises and not meet expectations. This will go bigger and bigger.”

Finally, the recent advancements in generative A.I., or the use of text or images to generate a human-like interaction, have opened more possibilities. Though currently too riddled with inaccuracies for real applications in clinical care, its potential in reducing administrative burden is heady. And the large language models these tools are based upon work best when built upon large amounts of structured and unstructured data. Healthcare is, of course, ripe with both.

Applying A.I. to Advanced Healthcare Technology

A.I. is expected to permeate every facet of the industry, with annual spending on artificial intelligence in healthcare estimated to reach more than $34 billion in 2025, up from $2.1 billion in 2018. A recent survey of healthcare organizations found 98% have implemented an A.I. strategy or plan to develop one.

The current applications of A.I. in healthcare are narrow and highly functional. But as adoption accelerates, it generates momentum which perpetuates the benefits of A.I. In fact, 59% of healthcare leaders expect to achieve a full return on their investment within three years.

Some current applications of A.I. for health plans include:

• Detecting and preventing claims overpayments including instances of fraud, waste and abuse
• Value-based care initiatives
• Claims management
• Supporting coordination of benefits
• Surfacing business intelligence insights
• Increasing effectiveness of clinical audits
• Member outreach and engagement
• Automating administrative processes
• Predicting healthcare needs

“With A.I., the more quickly organizations in early or middle stages of AI deployment move forward, the sooner they will overcome uncertainty and unlock the rewards of this powerful business tool.”

So far, no single A.I. vendor has emerged as the leader in the industry. In the coming years, healthcare payers will continue to see technology disruptors enter the market. For many plans, a selection of effectively managed vendors will be the most effective strategy to drive ROI. Though health payers will have to be careful of hype, particularly from tech vendors of artificial intelligence solutions for healthcare who lack industry experience. What works for one sector — say, finance — does not easily translate into healthcare, which is often more complex, more heavily regulated, and more data sensitive.

How Your Health Plan Can Find Value in A.I.

You might think that being a fax/email/spreadsheet organization means your health plan is woefully out of date, but you might not be as behind as you fear. Even as the early adopter stage of A.I. comes to an end, it still confers a competitive advantage for those health plans that can scale its use. But now is the time to move. Envision the role A.I. will play in your organization in the years to come and develop a strategy roadmap to invest in and leverage A.I. assets.

It pays to keep in mind that, while A.I. can offer much needed technology advantages to health plans, it can’t solve all payment integrity challenges on its own. Ensure your health plan has the best chance to realize these benefits now and into the future. You can and should deeply question your technology vendor and demand specifics around their solution’s A.I. capabilities.

Health payers will need to see all the moving parts of their tech ecosystem, including real-time metrics on advanced healthcare technology performance, in order to prove its value even if A.I. capabilities are touted. Increasing visibility across disparate departments, deriving insights from siloed data, realizing cost savings, and augmenting human expertise and productivity are exactly the type of improvements that show A.I. at its best.

Pareo® – Powered by A.I.

At this stage, A.I.- powered solutions may be more commonplace than you realize, but the true value offered by artificial intelligence for healthcare varies from vendors to vendor. The most powerful way to harness A.I. capabilities is when they are applied as part of a broader solution, an advantage provided by an integrative platform like Pareo®.

Pareo offers multiple applications for of A.I. as part of a broader “one-source” system insight platform for health plans and payers. Pareo leverages artificial intelligence to increase expert auditor efficiency and effectiveness with clinical audits. It integrates OCR technology to make unstructured data searchable, filterable and sortable. Then, NLP and machine learning applications of A.I. help auditors prioritize cases for review and automatically tag relevant documentation. The solution also generates confidence scores for recoverability and denials so cost containment leaders can better trust results.

A.I. is a crucial technology for your health plans to adopt or expand upon throughout your organization. Set yourself up for success and start using artificial intelligence now to address the challenges in healthcare brought on by increasingly complex data.