Bridging the Chasm: How AI in Healthcare Is Closing the Gap Between Potential and Practice

Finding ways to effectively scale doctor-patient interactions might be one of the most pressing challenges facing humanity today. Can AI be the solution?

Contemporary medicine is one of humanity’s greatest triumphs. Most of us live not just longer but healthier and more fulfilling lives than our ancestors thanks to centuries of carefully accumulated knowledge embodied in our everyday infrastructure and skillfully applied by trained experts. Some of our largest institutions, legal frameworks, and cultural practices have evolved over time to make this possible.

Technology (including the development of drugs and devices) and contextually applied expertise are the twin foundations of this achievement. But there’s a fundamental difference between them. Economies of scale have made the marginal cost of most drugs and procedures very low: in many cases, and especially so in drugs, their affordability is driven by R&D costs and IP considerations rather than their manufacture or scarcity.

Human expertise is an entirely different matter. finding ways to effectively scale doctor-patient interactions is one of the most pressing challenges facing humanity today.The doctor-to-population ratio in low-income contexts can be 1:10,000 or even lower, and although the ratio in high-income contexts can be 1:300 of higher, and therefore above the WHO-recommended threshold of 1:1,000, most people’s experience with the healthcare system is one where interactions with doctors are short and infrequent, either because the patient lacks consistency in their follow-up or because resource limitations make a more generous allocation of time unfeasible.

“Short and infrequent interactions with doctors” is probably not in most people’s list of top complaints about modern medicine, but in a world where chronic diseases are the most important driver in the loss of healthy years of life it might be one of the most serious of silent killers.

Chronic diseases often resemble the slow and silent spread of rust or mold, where the deterioration is gradual and may not trigger an immediate alarm. These conditions can be insidious, developing quietly over time, and the feedback or symptoms they provide can be easy to dismiss or overlook. Diseases like diabetes, hypertension, and many forms of cancer can progress silently until they become significantly severe.

The stage at which a chronic disease is diagnosed can greatly affect the management and prognosis of the condition. Early detection through regular screening, increased awareness, and better access to healthcare can lead to earlier diagnosis and more effective management.  Prevention and early detection, especially for conditions that are subtle in their early stages, becomes, then, crucial and the current design of our healthcare system cannot provide it. Most people simply lack the knowledge, resources, and trained attention to monitor themselves for the earliest development stages of chronic diseases. We depend on doctors to guide and understand an appropriate, adaptive pattern of diagnostic procedures that can detect chronic diseases at the stage in which they can be managed easily and cheaply before they turn into conditions serious enough to force patients to seek medical attention – at which point many of the most important tools of medical science have lost much of their ability to help.

This is a widely known issue. There is a second, and less widely known issue: patients deviate from the prescribed course.

Noncompliance can be a significant barrier to the effective management of chronic diseases.  The extent to which individuals follow their healthcare providers' recommendations, is crucial for the effective management of chronic diseases. However, adherence can be undermined by a range of factors. Complex treatment regimens can overwhelm patients, while a lack of understanding about the treatment can result in a failure to appreciate its importance. The side effects of medications can deter patients from continuing with their treatment, and the ongoing nature of long-term therapy can lead to what is often termed 'treatment fatigue.' Financial constraints, especially with high-cost medications, present another significant hurdle. Cultural and personal beliefs may also impede compliance, as can psychological challenges such as depression or anxiety. Poor communication or strained relationships with healthcare providers can further erode a patient's commitment to their treatment plan. Additionally, lifestyle changes that a treatment might necessitate can prove too demanding, and low health literacy can leave patients confused and less likely to adhere to their prescribed course of action. Addressing these issues often requires a multifaceted approach, including patient education, streamlined treatments, enhanced provider-patient communication, and supportive interventions. and can lead to various complications, including the progression of the disease, reduced quality of life, and increased healthcare costs. Doing all these things effectively in the span of an equivalent of 10 seconds per day – the average length of a medical consult distributed over the time span between them – emphasizes the enormity of the challenge.

Enter AI.

Not as a plot twist. This is a article on the application of AI in healthcare, in a blog about the application of AI in healthcare, at a time when the application of AI in healthcare is a topic of febrile interest. The analysis made so far isn't meant as an introduction to presenting AI as an innovative move, but instead to clarify the problem beyond the generic perception that there is one. The limiting factor isn't the availability of everything doctors do. It’s the narrower capability of paying continuous attention to a single individual and consistently applying protocols to monitor, assist, and, when necessary, refer them to specialists.

Whatever their limitations in other areas, paying continuous attention and consistently applying protocols is something that AIs – not hypothetical human-equivalent AIs but the sort of AIs we can build right now – can do exceedingly well. Just as importantly, they can do it at a nearly infinite scale. Instead of having to explicitly or implicitly ration access to medical expertise and therefore the most effective care, once expertise lives in software we find ourselves in a situation where we might as well not just give everybody a dedicated personal physician-AI but have it run with whatever frequency we want. This is significantly less ambitious than it might sound: even when we aren't interacting with them, social networks are constantly gathering information about us and figuring out how best to capture and monetize our attention. Building systems with a different goal but the same focus, scope, and ambition requires not a technological but a conceptual revolution. It's not harder to guide somebody through controlling their blood pressure according to standard medical protocols than it is to keep them engaged in a well-designed microtransactions game. In many ways it's much easier: you aren't attempting to capture their attention – the most scarce and overexploited resource – but to pay attention for them. Over the long term a person with a personal physician or an AI playing the same role will need to think less about their health than one without.

Explicitly and by design, we do not propose to replace doctors. As burglar alarms or fire alarms are not designed to replace firemen or policemen. They are designed to cope with the resource allocation problem.  Similarly, the systems we need to build don't do everything doctors do: they do the routine subset that AIs can do equally well and they do it all the time for everybody, meaning that when doctors do see patients they see them when their full expertise and empathetic capabilities are most useful, not when they happens to remember to make an appointment or simply to verify a routine check. We don't expect doctors will be less busy than they are now; we do believe they will be more effective and enjoy their work more. As AI capabilities improve the threshold of the routine will rise. More of what medical research has laboriously discovered will be applied more often to more people, and every patient-doctor interaction will be more meaningful to both.

Saying "AI" though, is not enough. As we have discussed in another article (link), it is less a specific technology than a marker of where the frontier of computing lies today. What forms of AI should be used, and how is not a trivial question.  But regardless of the specific application, the scarcity of healthcare professionals makes leveraging artificial intelligence (AI) for remote monitoring of patients, especially those with chronic conditions mandatory. AI systems can continuously analyze health data, providing a kind of vigilance that can complement the periodic check-ups by human doctors. This technology can alert healthcare providers to potential issues before they become acute, enhancing patient care and allowing doctors to manage their workload more effectively. Remote AI monitoring acts as a force multiplier in healthcare, extending the reach and efficiency of the medical workforce where human resources are limited.

As we stand on the cusp of a technological renaissance in healthcare, the integration of remote AI monitoring is not just an innovation but a beacon of hope. It serves as a force multiplier, a testament to our collective ingenuity, empowering a finite medical workforce to transcend traditional boundaries. In areas where the scarcity of human resources is the tightest knot, AI extends the healing touch of healthcare, ensuring that distance is no longer a barrier to care. This fusion of artificial intelligence with medical expertise heralds a new era where every heartbeat matters and no patient is beyond reach. With each step forward, we reaffirm our commitment to a future where technology and humanity converge for the greater good of global health and well-being.