The field of artificial intelligence (AI) has been around since the mid-20th century, but in the last two years, it has gone from science fiction to science fact at an incredible pace. Large language models (LLMs) like ChatGPT have dominated headlines in this space. LLMs are essentially AI systems that can understand, generate, and manipulate human language in various domains. LLMs represent a once-in-a-generation technological advance that holds incredible promise to change the nature of work for many of us. As physicians, it’s crucial that we engage with this transformative technology now or risk being left behind as it reshapes our world and our profession.
The LLM revolution has been brewing for years, fueled by a perfect storm of technological advancements:
Although neural networks have been around for a long time, there hasn’t been all that much data, relatively speaking, for them to train on. But now, the internet is about 30 years old, and a shocking amount of data is available. Though difficult to estimate, the internet contains over 50 trillion gigabytes of data; to put this in perspective, it would take over 120 million years to download all that data on a single machine (assuming 100 Mbps)! So, algorithms now have a lot of data to train on.
The rapid advancement of graphics processing units (GPUs) has also been a game-changer for AI over the last two decades. NVIDIA, a company that’s been making headlines lately, has developed GPUs that can perform over 100 teraflops (trillion floating-point operations per second). To put that in perspective, a single teraflop is equivalent to doing one trillion calculations per second! With this kind of computing power, training massive AI models that would have taken months can now be done in a matter of days to weeks.
In 2017, Google researchers introduced a new AI architecture called the transformer in their paper “Attention Is All You Need.” This new structure for artificial neural networks revolutionized the field of natural language processing by allowing AI models to weigh the importance of different words in a long sequence, enabling them to understand context and meaning far better than before. The transformer architecture now forms the foundation for most top-performing LLMs.
These advances laid the groundwork for a watershed moment in AI: the public release of ChatGPT in November 2022. OpenAI democratized access to LLMs such that now anyone could go online and interact with one of the most incredible tools ever conceived. The public’s response was overwhelming, with 100 million users flocking to the platform within two months. ChatGPT’s launch represented a major milestone in the evolution of AI, marking the beginning of a new era in which advanced language models are not just the domain of researchers and tech giants but a technology that anyone can access and engage with.
However, the gold rush mentality surrounding AI raises concerns that its development in health care could be driven by misaligned incentives. The painful lessons of the electronic health record (EHR) era, where systems were designed for billing efficiency rather than clinical care and worsened physician burnout, must be avoided. If AI assistants are rushed to market to maximize profits rather than improve patient outcomes, they risk causing more harm than good.
To prevent history from repeating itself, the medical community needs to proactively shape the development of health care AI. By participating in research, we can help create AI that augments clinical decision-making without trying to replace human judgment. We can advocate for transparency to mitigate bias and ensure AI is fair across patient populations. We can also contribute our hard-earned clinical wisdom to complement AI’s raw intelligence, providing empathy and a nuanced understanding of illness to maximize the efficacy of these tools.
Here are some key steps physicians should take to engage with AI:
1. Educate yourself about AI and stay current on the latest developments, especially regarding health care applications.
2. Advocate within your organization for physicians to sit at the table in discussions and decisions around procuring or developing AI systems. Systems are looking for ROI on AI investments, which is fine, but we need to ensure that physicians and patients also see the benefits.
3. Participate in research studies to test medical AI tools and provide feedback to help refine them.
4. Collaborate with data scientists, ethicists, patient advocates, and other stakeholders to develop guidelines for the responsible design and deployment of medical AI.
5. Share your experiences and perspectives on AI, both the challenges and the opportunities, to foster greater dialogue in the medical community.
I’m both incredibly excited and a little apprehensive about how AI will be implemented in health care. As the historian Melvin Kranzberg said, “Technology is neither good nor bad; nor is it neutral” – its impact will depend on how we choose to develop and apply it. I firmly believe that if physicians proactively engage in shaping AI, we can steer it in a positive direction that empowers us to deliver better care while easing the burdens of modern practice.
Brandon Hunter is a pediatric critical care physician.
