14 August 2025
Opinion as featured in NZ Doctor - 14 August 2025
The genie is out of the bottle. After three years of swimming in the world of artificial intelligence as both a clinical leader and the clinical prompt engineer for our organisation's AI tools, I’ve become convinced that the arrival of AI in general practice promises something different than that which many fear.
Rather than replacing us, AI clinical decision support tools will reveal what is most valuable about general practice: the profoundly human aspects of caring, and helping patients navigate the complex and messy realities of health and wellness.
Let's be clear about where we are right now. Patients are already using ChatGPT (or other AI chatbots) as the new "Dr Google." Recent Australian research estimates that 10% of adults (about 1.9 million people) asked ChatGPT health-related questions in the six months preceding June 2024. As of today, this number is almost certainly a vast underestimate. Our patients are arriving at consultations armed with high-quality information, differential diagnoses, and specific questions about their treatment.
Meanwhile, GPs are already using these tools for clinical decision support, whether officially sanctioned or not. Resisting this trend is not just futile, it's potentially dangerous. We must embrace this technology to ensure it is integrated into clinical practice safely and equitably, to the benefit of both patients and clinicians.
AI is a better diagnostician than you.
Well, maybe not. But the evidence for AI's diagnostic capabilities is now overwhelming. ChatGPT alone achieved a median diagnostic accuracy score of about 92% in a recent Stanford study, significantly outperforming physicians who scored 74-76% whether using conventional resources or AI assistance. AI systems have demonstrated 94% accuracy in detecting lung nodules, substantially outperforming human radiologists' accuracy on the same task. This isn't a future possibility. It’s here now.
Even working in AI every day I find it difficult to stay abreast of developments. New tools and models are announced almost daily. One study estimates the pace of AI algorithmic improvement at roughly 400% per year, while the cost of querying an AI model like ChatGPT dropped from $20 USD per million tokens in November 2022 to just 7 cents per million tokens by October 2024, a more than 280-fold reduction in approximately 18 months.
In 2022, the smallest model achieving over 60% on a key AI benchmark had 540 billion parameters. By 2024, a model achieved the same performance with just 3.8 billion parameters, a 142-fold reduction in model size to performance over two years.
The trajectory is clear: these tools are becoming exponentially more capable while becoming dramatically more accessible.
Traditionally, medical training and practice have relied heavily on doctors functioning as human encyclopedias of medical knowledge. Much of the emphasis in generalist medical training has been on breadth of technical knowledge across diseases and their treatments, and on becoming expert in pattern recognition.
In reality, we know that the depth and breadth of technical medical knowledge varies significantly across GPs. Some are exceptional diagnosticians; some are stronger in particular areas; others have knowledge that has atrophied or have not kept abreast of the vast and ever changing body of medical knowledge. We are, after all, only human. AI tools will level this playing field, providing every GP access to diagnostic capabilities that rival or exceed the best specialists.
This doesn’t spell the end for general practitioners. While some specialty areas of medicine may face disruption (diagnostic radiologists might have legitimate concerns), AI will actually reveal and amplify the unique value of general practice.
Experienced GPs know that the art of good general practice is about relationships and building rapport, caring and empathy, health navigation and education, and drawing together the complex strands of physical, mental, social, and spiritual wellbeing. These are skills that define excellent general practice, yet they are often overshadowed by the technical demands of diagnosis and treatment planning within the disease focussed model of healthcare. And these are skills that AI is simply not good at.
AI will liberate us from the role of medical encyclopedia, allowing us to become what patients really want and need: sage carers who understand the complexity of their lives, establish trust, and focus on helping them implement meaningful change to their health.
This transformation isn't limited to GPs. AI will serve as a technical knowledge equaliser which, given the current GP workforce crisis, can only benefit patients. Nurse practitioners, prescribing nurses, physician associates, clinical pharmacists and others will all have these tools at their disposal.
In this new landscape, GPs with specialised technical skills (dermatology, procedures, complex case management) will become increasingly prominent. The multidisciplinary primary care team will thrive, with AI providing the technical foundation that allows each team member to focus on their unique human contributions.
Importantly, AI tools also have the potential to address some of healthcare's persistent inequities. The Australian research also found that AI health tool usage was higher among groups who traditionally face barriers to healthcare access. AI, carefully implemented, will help democratise high-quality, objective diagnostic support for Māori, rural and other underserved communities that face significant access barriers and unconscious systemic bias.
As Lucy O'Hagan recently noted in these pages, the identity of the GP has been in flux for decades. She's right that we're no longer the traditional family doctor who handled everything from delivering babies to end-of-life care. Rather than lamenting this change we should see AI as clarifying and amplifying what remains most valuable about our role.
O'Hagan asked whether GPs are ‘grey area specialists.’ I agree. With AI handling the technical diagnostic heavy lifting, what’s left is being specialists in the most complex grey area of all: human health in all its messy, multifaceted complexity.
We face an unprecedented opportunity to engage with this technology, ensure it serves patients safely and effectively, and shape its development to enhance rather than diminish the doctor-patient relationship. The positive impact of AI Scribing tools on face to face engagement during consultations is an obvious example.
The future of general practice isn't about competing with machines, it's about partnering with them to provide care that is both technically excellent and profoundly human. AI will handle the pattern recognition and data processing; we'll handle the wisdom, empathy, and therapeutic relationships that heal.
Dr Graham Denyer is Chief Medical Officer at Tend Health, where he is a practising GP. Part of the leadership team since before launch, he developed Tend’s clinical governance framework and now leads its implementation. Graham has also played a central role in designing Tend’s AI-powered clinical tools. He will be speaking on AI at the GPCME South Plenary Session on AI at 8:30am Saturday 16 August 2025. He holds fellowships with both the Australian College of Rural and Remote Medicine and the Royal New Zealand College of General Practitioners.