Long COVID may be too big a problem for humans to solve—alone, at least.

Increasingly, researchers are turning to artificial intelligence to help them sort through the electronic medical records of millions of long-COVID patients in hopes of better understanding the enigmatic condition with hundreds of potential symptoms.

In some cases, A.I. is helping patients: In a study published last month in The Lancet Digital Health, researchers trained three machine-learning models to identify potential long-COVID patients among hundreds who previously had COVID. Both the models and humans agreed on probable “long haulers” in the vast majority of cases, showing that A.I. can help flag patients who have a high probability of experiencing the chronic condition and get them to care.

Fei Wang, assistant professor of health care policy and research at Weill Cornell Medicine in New York, is coauthor of a recently published study that examined patterns of diagnoses in long-COVID patients.

The researchers used machine learning to examine the electronic health records of thousands of patients and found four patterns among long-COVID patients, he said:

• More severe patients with blood and heart issues, many of whom likely were infected during the initial wave to hit New York City in the spring of 2020. This group had the largest number of patients with preexisting conditions.

• More mild patients with respiratory issues accompanied by sleep problems.

• Patients with new musculoskeletal complaints and neuropsychiatric problems.

• Patients who now suffer from gastrointestinal issues, including abdominal pain.

Long COVID is “so complex because it involves not just an infection” but potential fallout in the lungs and nearly every organ system in the body, in addition to inflammation, immune system issues—“lots of complicated reactions,” Wang told Fortune.

The sooner researchers can categorize patients and ascertain the cause of their disease—perhaps organ damage in some, and out-of-control inflammation in others—the sooner targeted therapies can be developed. It’s possible that some patients complaining of new ailments after COVID have unrelated issues, veritable red herrings—which is why A.I.’s assistance in sussing out patterns among the masses is critical, Wang said.

Later on, when treatments are developed, the patient lists developed by machine learning can be used to recruit patients for trials—a task that can be expensive and logistically tricky.

A.I. can also help researchers further categorize patients by variant and subvariant, enabling them to recognize patterns of long COVID that may correlate with various waves of infection.

For example, the first wave of COVID saw “lots of people being hospitalized, lots in the ICU, lots of mechanical ventilation,” Wang said. “The mortality rate was also the highest then.”

Is the long COVID of such patients caused by the coronavirus or Post-intensive Care Syndrome? The latter is caused by a traumatic and debilitating ICU stay that may have included intubation and prolonged bed confinement. Potential symptoms can include persistent muscle weakness, memory problems, and post-traumatic stress disorder.

It’s a puzzle Wang hopes A.I. can solve, with rapidity.

A recent algorithm-assisted study found a high rate of pre-COVID corticosteroid use in long-COVID patients. Many patients with severe COVID were treated with steroids in the hospital, especially those who were on ventilators. Do steroids cause long COVID, or play a role in causing it? Or are they merely indicative of sicker patients who might be at greater risk of long COVID owing to underlying medical conditions or a more severe course of the virus?

Wang isn’t sure; no one is. But he hopes the use of A.I. in research can lead to more answers, and more treatments, sooner for the millions who survived the virus only to find another—perhaps the biggest—battle lies ahead.

Although many are talking about Omicron as if it’s a more mild strain of COVID, some data suggests that BA.2 is associated with a greater risk of long COVID than BA.1.

“If you look at all this data—you need to be careful about your daily life and protection,” Wang cautioned. “We’ve not ended this pandemic yet.”