From shopping sites that list "Recommended for You" to social media networks suggesting "People You May Know," artificial intelligence (AI) is built into much of what we encounter in everyday life, working to anticipate our needs even before we know we need them.
Now, AI is working in the health and human services space as well.
For many, that prospect may conjure images of robots providing the expertise once given by providers or caseworkers. AI, when done well, doesn't replace human caregivers. It works in the background, helping humans deliver better care.
"AI can be about delivering and coordinating care and providing recommendations of services, through the right channel, to drive engagement with your desired clients," says Don Johnson, Optum Chief Technology Officer and Vice President of Product for State Government.
"That's super impactful."
To make AI work for you, you need to plan and understand where you could encounter pitfalls. You can follow these steps to get the most out of AI:
* Choose a problem AI can solve
* Start with existing data sources
* Insulate your AI system
* Watch for bias
* Actively measure and adjust
Choose a Problem AI Can Solve
A key first step is to understand how best to apply AI.
AI is an especially impactful tool when it solves what Johnson describes as "a Goldilocks type of problem"--issues not so small that they're not worth the time and money necessary for building an AI system, but not so large that they exceed AI's capabilities.
"You're looking for opportunities to streamline processes that are replicable and need to be tackled consistently," explains Johnson. "It's about providing contextualization, especially for large sets of data."
Prior authorization to pay for a medication or treatment is a good example. Typically, a human operator navigates a vast number of policies and procedures before arriving at a decision.
AI can help by aggregating that data, enabling the operator to deliver faster and more accurate authorizations.
Chat logs and other freeform client interactions can be another AI opportunity, specifically natural language processing.
While legacy systems may be trained to understand precise clinical or program terms, clients will often use more descriptive language like "antibiotics" or "my diabetes medication," or "income verification." AI can be trained to associate client terminology with more official terms.
Start with Existing Data Sources
The most efficient way to get an...