Knowledge is telling the past. Wisdom is predicting the future.”
Timothy Garvey, American endocrinologis
The new year is always a good time to pause and for perhaps a few fools to make some predictions in every sector, including ours of artificial intelligence in medicine and healthcare. Given that we are witnessing several fast-moving developments in artificial intelligence, here are my predictions for this coming year and beyond (in no particular order) in artificial intelligence in medicine and healthcare:
- Increasing awareness that federated/swarm learning can obviate the need for raw data sharing
No matter how much institutions wish to be collaborative, data sharing is often the biggest obstacle for institutions to work together on a data science project. And yet, most healthcare centers are not yet indoctrinated on the concepts of federated or swarm learning. Along with these strategies that can obviate the need for data sharing, the use of synthetic data will also be more accepted by the medical community as a strategy to not solely rely on data sharing.
- The transformer models making impact in domains other than just natural language processing
While almost all of the work thus far using transformer models has focused on natural language processing, it is likely that we will see the powerful “attention” mechanism being deployed for other domains such as image interpretation. The concept of converging current convolution neural networks with these attention mechanisms may be promulgated this coming year as we explore more sophisticated AI technologies for medical images (including moving images).
- Escalating use of large transformer-type language models (LLM) in healthcare projects
Language models such as GPT-3 and upcoming GPT-4 (to be released this year) as well as ChatGPT will find uses in healthcare. Early use of these language tools to synthesize a medical review has been acceptable (but not impressive). There is already discussion of using a version of MLOps for these LLMs (aptly called LLMOps). These language tools are even more essential as we lose senior clinicians in healthcare, but one issue may be insufficient healthcare data to support these LLMs.
- Exponential appreciation and deployment of robotic process automation tools in healthcare
There are huge inefficiencies that result in revenue loss in healthcare, ranging from high no-show rates to failures to obtain authorizations for procedures. Some of these deficiencies in operational aspects of health systems can be ameliorated by appropriate use of robotic process automation (RPA). Health systems will be pushed to use automated tools to decrease cost as well as increase revenue especially now with the COVID pandemic exaggerating these inadequacies.
- The educational curriculum in a very few medical schools will include artificial intelligence
While a very few medical or osteopathic schools from around the world are actively pursuing inclusion of data science and artificial intelligence content in the curriculum, even fewer healthcare professional schools (pharmacy, nursing, dental) are working towards this future-thinking educational strategy. With the exponential rise in artificial intelligence capabilities, the gap between medical school education and real world knowledge is becoming even larger.
- Clinical training programs will continue to be relatively slow in adoption of artificial intelligence
With the exception of a few AI-mature subspecialties (like radiology, cardiology, pathology, critical care medicine, etc), most subspecialty clinical training programs will not be aggressive enough in educating and training the trainees in concepts and applications of artificial intelligence in their subspecialty. The crux of the slower adoption is probably the lack of sufficient numbers of senior clinicians and leaders who have had experience with AI applications in their subspecialties.
Next week, I will list another list of predictions about AI in clinical medicine and healthcare in 2023 and beyond.
