Who is Flying the Plane?

Have you ever peeked into the cockpit while boarding a plane? Maybe you wondered how the pilots manage to operate the controls while keeping track of all the gauges. Getting a plane safely from one city to another is not a one-person job. While the pilot remains in control, others help guide the journey in several ways. Advisors work from the ground, studying weather patterns and air traffic and offering suggestions about the safest route. A copilot sits in the cockpit, working closely with the pilot—handling communications, adjusting controls, and checking that everything is running smoothly. Meanwhile, planning and guiding the flight is the job of the navigator. In the past, navigators were people. Today, that role is mostly handled by computerized systems that calculate and adjust the flight path automatically.

Airplanes are not the only places where roles like this matter. In many complex jobs, responsibility is shared: one person might be in charge while someone gives advice, someone works alongside the leader as a teammate, and a computer system or specially trained expert steers the bigger plan. In medicine, a doctor is a bit like an airplane pilot, but other medical specialists play critical supporting roles in caring for patients. These days, artificial intelligence (AI) is beginning to assist medical professionals in healthcare too. Some AI tools simply offer suggestions. Others act more like copilots, working side by side with human experts to assist them with parts of the job. Researchers are even building AI systems that could one day guide specific medical tasks more directly—not just advising or assisting, but even making certain decisions on their own.

From Advisor to Navigator

Here is a real medical example. People with diabetes must get eye exams often, during which detailed photographs are taken of the back of the eye. Tiny changes in blood vessels can be an early warning sign that diabetes is damaging the eyes (Figure 1). If these changes are detected early, treatment can prevent vision loss. How might a medical AI system assist doctors in this situation?

Before exploring the roles AI can play in medicine, it helps to understand how medical AI systems learn (Figure 2). AI does not think or reason like a human doctor. Instead, it is trained using huge collections of medical data, such as thousands or even millions of images (e.g., x-rays, CT scans, photos), lab results, medical records, or other health-related information. Each teaching example can be labeled to show whether a disease or health problem was present or not. Over time, the AI system learns to recognize patterns that are often seen in certain medical conditions—patterns that might be difficult for a human to see quickly.

AI as Advisor

At the advisory level, AI works a bit like the person studying the weather and warning the pilot about a storm. The AI system examines patients’ eye photographs, looking for patterns in color, shape, and texture, and compares them to patterns it learned in training (Figure 3A).

An AI advisor might highlight a small area where blood vessels look unusual or estimate the chance that early eye damage is developing. AI is great in this type of advisory role because it can scan images quickly and consistently. However, it does not make a diagnosis or decide on treatment—it simply offers information. The doctor reviews the image, considers the patient’s medical history, symptoms, and other test results, and decides what the AI’s findings mean and what to do next. A pattern that looks worrisome to a computer may turn out to be harmless, or it may confirm what the doctor already suspected.

AI acting as an advisor is already common in many hospitals, especially for estimating the risk of breast cancer from medical images or spotting unusual heart rhythms or signs that a patient might later develop heart trouble [1].

AI as Copilot

Beyond just pointing out possible problems, AI can work at the copilot level to actively perform part of the work (Figure 3B). In our eye exam example, AI might automatically measure changes in the blood vessels, compare the new photograph to images from previous visits, or organize findings into a report for the doctor to review. It may also sort patient images by how serious the disease is, helping doctors focus first on patients that need treatment quickly. Basically, a copilot AI works alongside the doctor, like a teammate.

The difference between AI as an advisor vs. a copilot might seem small, but it is important. As an advisor, AI says, “You may want to look at this”. As a copilot, AI says, “I will help you do this”. Even at the copilot level, the doctor remains in charge and makes the final decisions. In our example, copilot AI still does not decide whether a patient has diabetic eye disease or what treatment is needed.

Copilot AI systems are less common than AI advisors, but new types are being built and tested [2]. Some copilot AIs are already working with surgeons by tracking instruments during operations, and in clinics by organizing patient records and prioritizing urgent cases.

AI as Navigator

At the navigator level, AI takes on an even more independent role. Instead of only assisting the doctor, these systems can carry out a specific medical task on their own, within carefully defined limits (Figure 3C). In eye care, a system called LumineticsCore has been approved to analyze photographs of the eye. If signs of diabetic eye disease are present, the system determines whether the patient should be referred to an eye specialist—without the need for a human doctor to analyze the image first [3].

This does not mean that AI working in a navigator role replaces the doctor completely. Navigator-type systems are designed for one clearly defined purpose—like screening for diabetic eye disease. They cannot diagnose every eye condition or help with other health problems. Human doctors still explain the results to the patient, provide treatment, and manage follow-up care. But, in the specific task they are trained for, a navigator AI does more than advise or assist. It makes a specific medical decision on its own.

Navigator-level AI systems are still rare. One example is automated insulin delivery systems, which can automatically adjust insulin doses for people with diabetes based on real-time blood sugar readings from small wearable devices [4]. Several others are still being developed and may assist doctors in the near future.

What Still Needs Improvement?

The more independent an AI system becomes, the more carefully it must be designed and tested. It must work reliably, be checked regularly to make sure it stays accurate, and have clear limits so medical professionals understand exactly what it can and cannot do. Before AI navigators can become common, there are still several big things scientists and doctors need to improve upon.

First, if AI is going to take on the responsibilities of a navigator, it must be able to combine many types of medical information easily. Medical decisions rarely depend on just one piece of information, like an eye image. Doctors must often consider blood test results, heart rate measurements, and other scans, along with a patient’s symptoms and medical history. The challenge is not just teaching AI to connect all these types of clues, but also making sure the information can actually be shared with the AI system in the first place. Patients’ medical records may be stored in different systems in different hospitals, imaging devices might come from different companies, and patients might collect data at home using different kinds of wearable devices. If these sources cannot share information smoothly, AI cannot see the full picture well enough to guide care safely.

Speed is another challenge. In hospitals, information changes quickly. A patient’s heart rate can change, lab results can arrive, or new symptoms can appear at any moment. For an AI system to truly act like a copilot or navigator, it must be able to update its predictions in real time to safely assist in urgent situations. That requires powerful computing systems that can process streams of information continuously and accurately. Not all hospitals have the advanced computer systems needed to support this level of technology.

Designing AI that fits into how doctors and nurses actually work is also essential. If an AI sends too many alerts, interrupts at the wrong time, or presents information in a confusing way, it will not be helpful and may even increase the risk of mistakes. If health professionals rely on AI too much, they might stop questioning its suggestions. If they do not trust it enough, they may ignore important warnings. For AI to work as a true copilot, it must build the right level of human trust. This means carefully designing how information is displayed, when alerts are given, and how responsibilities are shared between people and machines.

The Future of AI in Healthcare

As more AI systems move from advisor to copilot to navigator, they could change how doctors care for patients. When information from different doctors and devices is connected more easily, diseases might be found earlier and treatment could begin sooner. In busy hospitals, AI might even catch problems that humans would miss.

But even the smartest AI cannot replace a human doctor or nurse. Computer systems are great at finding patterns in data, but they cannot understand fear, hope, or pain. They do not sit beside a patient, explain what is happening, and comfort them. When a serious medical decision must be made, a person, not a machine, is responsible. Those parts of medicine still belong to people. Future hospitals may use smarter AI systems, but helping patients heal will always require thoughtful humans who can listen, and care, and understand what matters to each patient.

Glossary

Advisor: ↑ An AI role in which the system provides information or highlights possible problems, while a doctor reviews the results and makes all decisions.

Copilot: ↑ An AI role in which the system carries out parts of a task, such as analyzing/organizing data. The doctor reviews the info, decides what it means, and chooses the next steps.

Navigator: ↑ An AI role in which the system carries out a specific, limited medical task on its own. Doctors still explain the results, provide treatment, and manage the patient’s overall care.

Artificial Intelligence: ↑ Computer systems trained to find patterns in data and use them to make predictions or suggestions, often helping people solve problems or make decisions.

Diabetes: ↑ A condition in which the body has trouble controlling blood sugar levels, which can damage organs like the eyes, heart, and kidneys over time.

Diagnosis: ↑ The identification of a disease or condition based on symptoms, medical history, and test results; a diagnosis gives a name to the health problem and helps guide treatment.

Conflict of Interest

HA is Chief Medical Officer of Harbinger Health. This company based in Cambridge, MA, United States was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

The remaining author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Acknowledgments

We wish to thank Dr. Susan Debad for providing us with a first draft and for her continued collaborative input as co-author. We would also like to thank the coauthors of the original manuscript: Wanheng Hu, Jessica Morley, I. Glenn Cohen, and Payam Barnaghi. Infrastructure support for this research was provided by the National Institute for Health and Care Research (NIHR) Imperial Biomedical Research Center (BRC). The funders were not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

AI Tool Statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.