AI is the ability of machines or computer systems to perform tasks that require human intelligence, including skills like learning, perception, reasoning, problem-solving and decision-making. Two types of learning exist; Machine learning, and Deep learning. ML learning algorithms have three categories, supervised, unsupervised, and reinforcement learning. AI can be either physical or virtual – depending on where the decisions that AI take are reflected.

AI can aid practitioners significantly in medical diagnostics. Supervised learning excels in tasks like classifying pulmonary nodules on X-rays, detecting imminent strokes, and identifying arrhythmias in ECGs. While unsupervised learning can uncover hidden patterns in unlabelled data for hypothesis generation. An example is Dr. Anunashis Sau’s AIRE model, developed at Imperial College London, which analyses ECGs with great precision and predicting 10-year mortality risk with 78% accuracy. He says, “the AI model detects much more subtle detail [than a cardiologist], so it can spot problems in ECG’s that would appear normal”.  The AI is not going to be used to “replace” cardiologists, rather “to provide doctors with relevant risk information”, ultimately helping improve quality of care in the NHS when it does get implemented.

Deep Learning (DL), specifically convolutional neural networks (CNNs) are already revolutionising medical imaging by hierarchically processing scans. CNNs are the backbone of computer-aided diagnosis (CAD), which speeds up anomaly segmentation, biomarker isolation, and disease prediction. They are trained on vast datasets, and improve speed of CT/X-ray interpretations, which can enable preventative medicine rather than curative medicine. This is an important aspect, as very recently it was reported that “Failing to properly diagnose and treat people with bipolar disorder is wasting billions of pounds a year in the UK”. If predictive medicine can take precedence over curative, then not only do patients feel the positive effects, but also the general taxpayer due to a more efficient NHS.

AI as a diagnostic assistant

Generative AI, for example CDS (Clinical Decision Support Systems), can revolutionise diagnostics departments by analysing X-Rays, MRI’s, Lab results and patient histories at astonishing speeds. It is certainly very advanced in pattern recognition and is able to distil through terabytes of data into summaries or allow doctors to query a certain data set to get answers. These can be used as a tool by doctors, nurses, and specialists – allowing them to cut through the noise and pinpoint diagnoses. The WHO predicts that tools like this will transform outcomes by harmonising fragmented healthcare data. However, this strength is also a weakness; AI’s accuracy depends entirely on standardised, high-quality data. For example, blood reports from a UK hospital and a Nepali clinic might use conflicting formats or labels, which can create errors when AI trained on say UK hospital data is used in a Nepali clinic. Unlike humans, AI cannot use context to group messy inputted data. The issue with traditional big data is the struggle with standardisation and data types – it comes in all forms, shapes and sizes. This makes traditional data analysis tools useless because there is no consistency between the data, even though they may all represent or show the same data.  Until there are international standards in place, the usability of a certain AI model will be restricted greatly, and its integration into clinical practices will throttle AI’s advancements. Thus, for the foreseeable future, it is likely to remain a tool rather than a replacement for physicians who can assimilate many contextual factors when diagnosing a patient.

Should AI handle repetitive tasks?

AI will soon be common tool for medical professionals to use, as it can easily automate many of the repetitive tasks, they carry out everyday. This would greatly free up healthcare professionals up to focus on providing holistic and tailored care. According to a study done on “Changes in Burnout and Satisfaction with Work-Life Integration in Physicians and the General US Working Population between 2011 and 2017”, doctors spend half their working day and an additional 28 hours per month on tasks involving maintaining electronic health records. This study showed that burnout rates were consistently higher in physicians compared to the general US workforce, even after adjusting for factors like age, sex, and hours worked. However, burnout isn’t just about long hours, because even after adjusting for work hours, physicians experience significantly more burnout than other workers. This suggests that AI could play a key role in reducing non-clinical burdens, like updating and maintaining electronic health records that contribute to physician burnout. It is important to remember that AI also has the potential to further increase the burden of physicians if it is not reliable a high percentage, and the output of AI always must be double checked by a human. AI would be able to aid doctors and may even earn a place in the multidisciplinary team, however, not replace them.

Accountability

AI’s role as a clinical aid and not as a replacement is limited by its reliance on high quality training data, which exponentially amplifies diagnostic errors and biases. Accountability remains a critical hurdle: if an AI misdiagnoses a patient, current legal frameworks struggle to assign liability between developers, physicians, or the algorithm itself, particularly in Africa, where reliance on foreign-developed AI risks can mismatch solutions for local contexts. As highlighted in recent analyses, the principal-agent model—where physicians shoulder liability for AI decisions—deters integration of AI by overburdening clinicians, while product liability laws fail to address AI’s evolving “black-box” nature. Proposals like risk-based liability, which prioritises hazard mitigation over fault, or reconciliation models (e.g., regulatory sandboxes), offer routes to balance innovation with patient safety. However, these approaches rely on the government having an open mind and global data standardisation, challenges which are evident in Africa’s fragmented regulatory landscape. As the article concludes, AI’s inability to “bear ethical weight” or “automate accountability” shows us why human oversight remains irreplaceable – physicians can bridge the gap between algorithmic precision and the moral imperatives of care, ensuring trust and responsibility endure beyond machines. This is not to say that physicians do not make mistakes, but simply that someone could be held responsible.

Can AI problem-solve?

Qualities of a great diagnostician is creativity and intuition – which AI often struggles to replicate. An example of these qualities as a House M.D. episode: A patient’s mysterious poisoning had stumped clinicians who tested multiple treatments for drug toxicity, food-borne illness, and pesticide exposure, which only neglected his condition. The eureka moment came by accident: phosmet insecticide from a market was absorbed through unwashed jeans. Context-dependent and non-linear thinking and reasoning are properties that AI cannot replicate yet.

While AI can excel in data crunching tasks like analysing scans or lab results, in practicality real-world medicine relies on nuanced, and unpredictable variables. A rare complication, which the model may not have been trained on, will certainly leave the AI system unable to provide a conclusive answer. Human physicians however can adapt dynamically. AI can provide the gift of time, time to focus on irreplaceable human elements of care – like empathy.

The recent introduction of DeepSeek’s R1 model has significantly impacted the AI industry, had a notable impact causing a sharp decline in NVIDIA’s stock value. On January 27^th, NVIDIA’s share price plummeted 17%, erasing over $589 billion in market capitalisation. This was due to Deepseek’s release, a Chinese trained AI model that rivals OpenAI’s o1 model in performance, while being trained at a fraction of the cost and time. Deepseek was trained with a budget of $6 million, whereas OpenAI has invested nearly $1 billion. However, the nuanced understanding and empathetic patient care provided by human physicians involve complex reasoning and emotional engagement, which AI is unlikely to master in the next 50 years.

Does using AI have a detrimental impact on the brain?

A study done by Microsoft on the use of Generative AI (GenAI) in workflows, which found that if a user trusts AI models over themselves, they were found to have lower levels of critical thinking, while higher self-confidence over an AI model meant higher levels of critical thinking. The study mentioned that although AI is excellent at doing repetitive tasks, and exception handling is often done by humans, the user is deprived of the routine opportunities to practise their judgement and strengthen their cognitive musculature, often leaving them atrophied and unprepared when these exceptions do arise and they are expected to handle them. This suggests that over-reliance on AI could diminish the analytical mindset that is essential in real life medicine practice. Additionally, the study showed that use of GenAI allowed users to focus more on information verification, and response integration.

Furthermore a closed loop system may lead to diagnostic monoculture, where clinical decisions in certain subspecialties reflect only the AI/ML recommendations, and physicians lose the ability to discover newer or better treatments. Overreliance on AI/ML may also result in physician deskilling, the loss of critical medical knowledge and skills, and decrease physicians’ ability to identify errors.

AI is no longer something distant – unlike flying cars – it is a present force, weaving itself into every where we can set our eyes. Its potential to streamline diagnostics, reduce physician burnout, and increase access to healthcare is undeniably a positive effect. However AI’s boasts are not without its price. Under the surface lie crucial questions about trust, equity, accountability, and the preservation of human touch in medicine.

AI may be able to crunch numbers at insane speeds, but it cannot actually care. It can recognise patterns, but it cannot recognise pain.

If medicine an art as much as it is a science, then AI must be the brush – not the artist.

Bibliography

Boudernhem, R. (2024, March 15). Shaping the future of AI in healthcare through ethics and governance. Retrieved from nature: https://www.nature.com/articles/s41599-024-02894-w

Drabiak, K. (2022, September 27). Leveraging law and ethics to promote safe and reliable AI/ML in healthcare. Retrieved from fronteirs: https://www.frontiersin.org/journals/nuclear-medicine/articles/10.3389/fnume.2022.983340/full

Hayward, C., & Pym, H. (2025, April 1). NHS billions wasted as bipolar patients left ‘forgotten and failed’. Retrieved from BBC: https://www.bbc.co.uk/news/articles/c045pp740vro

ICL. (2025, February 20). www.private.imperial.nhs.uk. Retrieved from https://www.private.imperial.nhs.uk/news-and-blogs/ai-model-can-predict-health-risks-including-early-death-from-ecgs

Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. Retrieved from Microsoft: https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf#page3

MJMR. (2024, May 13). In the future, will artificial intelligence be able to replace doctors? -narrative review En el futuro ¿la inteligencia artificial será capaz de reemplazar a los médicos? -revisión narrativa Sergio David Pintado Brito a. Retrieved from Research Gate: https://www.researchgate.net/publication/382041668_In_the_future_will_artificial_intelligence_be_able_to_replace_doctors_-narrative_review/fulltext/668960e3f3b61c4e2cb73db1/In-the-future-will-artificial-intelligence-be-able-to-replace-doctors-narrative-re

ScienceOpen. (2021, September 25). Applications of Artificial Intelligence (AI) in healthcare: A review. Retrieved from Science Open: https://www.scienceopen.com/hosted-document?doi=10.14293/S2199-1006.1.SOR-.PPVRY8K.v1

Shanafelt, T. D., West, C., Sinsky, C., Tockel, M., Tutty, M., Satele, D., . . . Dyrbe, L. (2019, February 22). Changes in Burnout and Satisfaction With Work-Life Integration in Physicians and the General US Working Population Between 2011 and 2017. Retrieved from ScienceDirect: https://www.sciencedirect.com/science/article/pii/S0025619618309388?via=ihub