AI and healthcare: Creating an “economy class” or democratising medical expertise?

Artificial intelligence is moving from an experimental technology to an active participant in healthcare. So far, AI tools are helping interpret medical images, summarise clinical records, identify drug candidates, and analyse genomic data. Yet as these systems become more capable, a larger philosophical and practical question is emerging: will AI democratise healthcare access, or create a two-tier system in which some patients receive traditional physician-led care while others are directed toward AI-mediated alternatives?

That debate recently gained renewed attention following discussion in the Journal of the American Medical Association (JAMA) about a possible future in which autonomous AI systems provide care with limited direct physician involvement. Critics have described such a future as an “economy class” version of medicine, raising concerns about quality, safety, and equity. Supporters argue that this characterisation misunderstands the reality facing millions of patients who already struggle to access timely healthcare.

It is worth pointing out, in my interpretation, that the JAMA authors are taking a deliberately provocative position. They argue that AI-alone care may eventually outperform both physician-only and physician-AI hybrid care for many cognitive medical tasks

The issue is not whether AI will influence medicine. That is already happening. The real question is whether expanded reliance on AI improves healthcare access without compromising care quality.

The argument for democratisation

Supporters of AI-enabled healthcare point to an uncomfortable reality: healthcare access remains highly uneven. Patients living in rural or remote regions often face lengthy waits for specialist consultations. In Canada, many communities continue to experience physician shortages, particularly in rural and northern regions. Access to genetics specialists, neurologists, and other highly specialized professionals can be particularly limited.

From this perspective, AI is not replacing a physician who is readily available. Instead, it may provide support where expert care is difficult or impossible to access.

Advocates argue that technology can make expertise more scalable. If an AI system can reliably interpret information, identify relevant research, or flag potential risks, it could extend the reach of scarce medical specialists and shorten the time required to obtain useful information.

One field often highlighted as a potential success story is genomics. The interpretation of genomic data generates enormous volumes of information that can be difficult even for trained specialists to analyse efficiently. AI-powered systems are increasingly being developed to allow researchers and clinicians to interrogate genetic datasets using plain language queries rather than highly specialized bioinformatics tools.

Companies such as Boston-based Bystro AI are pursuing this vision by applying large language models and computational analysis to genomic interpretation. The goal is not simply automation but making complex genetic information more accessible to researchers and potentially, in the future, to patients.

Supporters view this as democratisation rather than degradation. If technology allows a patient in a small community to access sophisticated genomic insights that would otherwise be unavailable, that access may represent a meaningful improvement in care.

Whether AI can drive ‘democratisation’ is contentious. As AI lowers the barriers to using advanced tools while simultaneously concentrating power and resources among major tech corporations.

The concerns about a two-tier system

Critics, often from the scientific and social policy arenas, argue that the benefits of expanded access should not obscure significant risks. Healthcare differs from many industries because decisions directly affect patient safety. Misdiagnoses, inappropriate treatment recommendations, or failures to recognise complex clinical situations can have serious consequences.

Large language models remain susceptible to factual errors, incomplete reasoning, and so-called “hallucinations,” in which systems generate plausible but incorrect information. While AI performance continues to improve, reliability remains a central concern. Many scientists worry that healthcare organizations may adopt AI not primarily to improve access but to reduce costs. In such a scenario, wealthier patients might continue receiving physician-led consultations while other patients interact primarily with automated systems.

This possibility underpins the “economy class” analogy. The concern is not merely that AI exists, but that it could become a lower-cost substitute for human expertise rather than a supplement to it.

There are also concerns about accountability. If an AI system makes an erroneous recommendation, who bears responsibility? The physician? The healthcare institution? The software developer? Regulatory frameworks continue to evolve, but questions about liability and oversight remain largely unresolved.

The middle ground: augmentation rather than replacement

Other experts believe the future will not involve a binary choice between human physicians and autonomous AI. Instead, AI may function most effectively as an augmentation tool. Under this model, clinicians remain responsible for diagnosis and treatment decisions, while AI performs tasks such as reviewing medical literature or identifying potential drug interactions.

This approach offers potential efficiency gains while preserving human oversight. In genomics, for example, AI may help identify variants of interest and summarize relevant scientific publications. However, the interpretation of clinical significance, communication of risk, and patient counselling would remain under the supervision of trained healthcare professionals.

Supporters of this model argue that it captures the strengths of both humans and machines. AI contributes speed and scalability. Clinicians contribute judgment, contextual understanding, and ethical decision-making.

As an example, genomics may become one of the most important testing grounds for healthcare AI. Genetic testing has become increasingly affordable, generating vast datasets that challenge traditional interpretation methods. A single whole genome sequence can contain millions of genetic variants, only a small proportion of which may have clinical relevance.

The shortage of specialized geneticists and bioinformaticians has created bottlenecks in many healthcare systems. AI tools capable of rapidly searching genomic databases, scientific publications, and variant repositories could dramatically accelerate analysis. Yet genomics also illustrates why human involvement remains important. Many genetic findings involve uncertainty. Variants may be classified as having unknown significance. Risk estimates may depend on family history, environmental exposures, and emerging research findings. Patients often require counselling to understand the meaning and limitations of results.

Returning to the central part of this article. The debate raises a broader ethical question. Is some access to expertise better than no access at all? Consider a patient who waits nine months for a specialist appointment versus a patient who can immediately access an AI-assisted preliminary assessment. For many individuals, the comparison is not between AI and a physician. It is between AI and no meaningful guidance at all.

The most successful implementation of AI may therefore depend on ensuring that automated systems are used to expand access while maintaining consistent quality standards across patient populations.