Who needs education when we have AI?

Humans, on the other hand, use information to form core conceptual ideas and then make logical inferences based on this. This conceptual understanding is something AI does not possess. Therefore, skills that are predicated upon a conceptual understanding — such as the ability to autonomously exercise moral judgement, the ability to truly innovate outside of known contexts, the ability to tell truth from untruth, and the ability to act with responsibility and agency — remain fundamentally human.

Mapping AI’s strengths and limitations (Figure 1) results in a clearer understanding of the clinical scenarios where AI excels, and the ones that constitute potential blind spots. It performs most reliably in common, protocol-driven scenarios that depend on pattern recognition and probabilistic deduction. However, since AI has no conceptual or pathophysiological understanding, they are less reliable when confronted with rare, atypical or masquerading presentations — a clinical performance profile somewhat akin to those of a novice clinician. More astute practitioners are less likely to be distracted by “surface” clinical features and instead ground their decision making in the pathophysiological mechanisms of disease, which supports more accurate diagnostic reasoning and better-informed management choices, especially when presentations depart from textbook patterns.

In addition, AI lacks one further fundamental skill and hallmark of clinical competence: metacognition, the ability to reflect on one’s own thinking. A person without the insight to appreciate their own flaws and boundaries can be exceptionally dangerous in a professional environment. Collaborative care, multidisciplinary involvement and an openness to complementing one’s weaknesses with the strengths of one’s peers are core components of modern oncological care and remain outside the scope of current AI.

Understanding this, we can now return to our earlier question: In a world with AI systems, why would medical education remain important?

The answer is, because AI cannot provide ethical, accountable or responsible health care and also lacks the conceptual and metacognitive skills required for clinical excellence. Thus, medical education retains a requisite role in nurturing the skills of learners to attain these higher-order skills.

With this clarified, a secondary question becomes apparent: will medical practitioners of the future still need to memorize and operationalize the same immense quantity of medical knowledge as previous, given they will easily be able to ask AI for this information?

To answer this, we need to consider the nature of knowledge and learning. This is constructed in a pyramidal fashion, grounded with a broad base and then progressively built upwards, layer by layer, to attain the “pinnacles” of specialist knowledge. The practice of radiation oncology requires multiple interdigitated layers; anatomy, molecular biology, physics and clinical medicine remain the conceptual foundations of our practice. If we remove, or even thin, these foundations, our metaphorical “knowledge tower” quickly becomes unstable. In effect, we risk becoming like the patient who researches their “red rash” online and jumps to the most improbable of conclusions because they do not have the requisite contextual knowledge to situate their findings. Hence, whether we like it or not, residents and junior staff will still have to memorize cardinal signs and anatomical boundaries for the foreseeable future.

Medical education is thus not in the midst of the existential crisis that many fear. But this does not mean it will have no impact from AI. Many of our teaching practices will need to be reconsidered in this new era. Broadly, these changes will fall into four major categories:

1. Integrating AI into teaching and learning practices

AI provides many new opportunities for learning in an individualized, tailored and responsive way. Already students have begun to use LLMs as AI tutors, asking questions and receiving feedback. Embracing this in a considered, guard-railed fashion will be vital. AI can also analyze individual student data at a speed and scale that would not be feasible for humans, thus enabling personalization of learning. Institutions will need to actively build these capabilities into their digital education infrastructure or risk being left behind in the quickly-evolving educational environment.

2. Reconfiguration of educational assessments

There is a pressing need to reconfigure educational assessments to ensure they are accurately measuring learners’ skills. Currently, many assessment formats are vulnerable to being compromised by AI: essays, short-answer questions and multiple-choice foremost amongst these. Hence, the days of take-home assignments and open-book tests are numbered. Educational organizations will need to consider moving to a greater proportion of real-time assessments (proctored examinations, OSCEs, vivas), or workplace-based assessments, or both.

3. Maintaining learner motivation to learn “lower-order” skills

Educators will need to find new methods to motivate the learning of basic medical facts and diagnostic processes, given that these skills can now be so readily performed by AI. This is a modern-day variant of the “calculator problem” where a five-year-old questions the benefit of learning basic arithmetic when calculators are freely available. In both cases educators need to communicate the principle of pyramidal learning and explain how that learning is necessary to set the platform for future, more advanced processes.

4. Ensuring relevant AI skills are incorporated into educational syllabi

With AI skills becoming an embedded part of clinical practice, radiation oncologists of the future will need to become familiar with basic AI skills to be proficient in their everyday work. This will require a review of educational syllabi to ensure these requisite skills are incorporated appropriately, therefore keeping the educational program relevant to contemporary practice.

In summary, if education is considered purely as information transmission, AI appears highly educated. But if we view education more wholistically as a path that leads us toward the highest forms of innovative and responsible critical thinking then AI is actually not educated at all. As such, the core role of medical education in preparing tomorrow’s health care workforce remains unchanged. However, educators certainly have their work cut out to evolve educational practices in ways that stay current and relevant to modern clinical practice in this increasingly AI augmented world.