With the evolution of technology and innovation surrounding artificial intelligence (AI), it is essential to step back and dissect the nuances of how best to implement this emerging frontier in Health care. AI attempts to mimic human intelligence through iterative, complex pattern recognition and matching based on input data at a speed and scale that exceeds the human capability.
Based on this definition, it behooves us to ask the question, what is “the problem or job” we are requesting of this technology. Since the primary output is the recognition of patterns, the information is considered correlative and, not causative. In other words, the information is related and not based on cause and effect. Therefore, hypothesis generation is a major benefit of AI. These identified correlations that drive to hypothesis require testing. The results of such research can be “plugged” back into the machine learning to enhance the capabilities.
As we venture to more complex use cases such as risk prediction, diagnostic choices, and treatment modalities., is crucial that set standard algorithms for treatment be in place. There is no such consensus for many disease identification and treatment. A value of AI may be to help develop a better understanding of these disease states and thus the development of such treatment consensus.
As mentioned, data inputs are foundational to AI. Presently, our data has numerous limitations. Claims data drives payment, not understanding of illness. “Sick” encounters heavily bias present clinical data embedded in electronic medical records. Wearable devices tend to be attractive to healthier individuals, thereby creating a set of information for those that are healthier. Currently, we are neither producing data sets based on economic behaviors nor human nature. And herein lies the challenge – all of these different data sets require integration in order to drive optimal results.
As we progress in the innovation of treatment of patients, it will be imperative for us to measure the outcomes within the context of our present knowledge. Balancing the polarity between innovation, safety, and efficacy will require a scientific and regulatory approach to avoid unintended consequences. Implementing new technology requires one to comprehend that immediate perfection is unrealistic, yet this becomes problematic when discussing the impact on human lives. In order to see the benefits it is imperative to remember that the goal is to improve our present status, which in itself is flawed.
In fact, this problem is evident with AI and self-driving cars. Despite many accidents occurring, resulting from human error, today many remain skittish concerning continued testing when a mishap occurs. Undoubtedly our greatest asset is human life; hence diligence is of utmost importance.
Moreover, the integration of state-of-the-art technology poses additional challenges. Humans have an innate tendency to jump to models of substitution rather than choosing models that enhance a situation. Optimally, AI will augment our clinical care, and not serve as a replacement. Unfortunately, adoption hindrance occurs as more information elevates the present burden of information assimilation. Investigating and understanding our healthcare methods that are not value-added requires us to increase our exploration and opportunities of all options which initially may increase our inputs, and workload, before it can help limit what is less important and optimize our workflows.
Recognizing and acknowledging all these dilemmas is necessary for progress. Similar to the adoption of other innovations it is a step-by-process that will involve trial and error. Let us not throw up our hands in frustration, instead embrace the possibilities of delivering better care to those we serve. Having such a purpose will allow us to grow and learn as we adopt innovative technologies such as AI.