Rise of the Machines? Artificial Intelligence May Represent the Future but It Is Not Ready for Prime Time

Abstract

Commentary Advances in technology and computer engineering have revolutionized many industries and changed the way that we look at and do things today. From finance to travel, in a relatively short period of time, we have learned to tap into the power of machine learning and artificial intelligence to help us find patterns and answers in the chaotic world that revolves around us. Seeing this impact, it was simply a matter a time before the medical community asked the question: What can this technology do for our work and our patients? In their literature review, Polce et al. evaluated the various applications of machine learning and artificial intelligence in joint arthroplasty research. While, in general, the studies reviewed showed that these algorithms worked reasonably well, there was considerable heterogeneity in the sensitivity and specificity of each model depending on the clinical application. The authors found that models dealing with image detection substantially outperformed those analyzing clinical outcomes. They also concluded that a lack of standardization in reporting metrics and rules of algorithm development were barriers to their ability to assess the utility of this technology today. The value and utility of a given technology are highly dependent on its proper application. When it comes to machine learning applications and artificial technology, it is no different. While it seems in vogue to apply such technologies for most things today, these programs are far from perfect. Current algorithms still require considerable human influence behind the scenes to ask the appropriate questions and to train these programs. In orthopaedics, these technologies are still in their relative infancies. Therefore, it is no mystery that the highest performance for these algorithms was achieved in the setting of image recognition, where the input parameters are nearly binary and well defined. It is also no surprise that when the data quality is poor and the associations are complex, as in the setting of clinical outcomes, the performance of these models deteriorated and could vary depending on the degree of training and the quality controls of each computer algorithm. Consequently, standardization of design and input parameters will improve the consistency of reporting and increase the validity of findings using these research methodologies. Consensus and standardization will also speed up and improve the predictive power of these algorithms. The ability of these programs to iterate and improve is largely dependent on the size of the data sample and the reliability and reproducibility of the computer program. When there is no standardization, the rate of development is slowed due to limitations in the rate of “learning” for each particular algorithm. However, when there is consistency, the rate of evolution increases geometrically for both humans and machines alike. Competing programs based on fundamentally agreed-on principles of design allow both the developer and the programs to validate design features, both positive and negative, at a much faster rate. Ideally, this is how machine learning and artificial intelligence should be applied in orthopaedic research. The power of these tools is undeniable, and the expansion in their utilization is also unstoppable. However, it is my hope that the day when medical research becomes fully automated and we make decisions for our patients solely based on recommendations from a computer will never come. As we collect more data from our patients and our procedures, we will need these technologies to help us sift through the granular details and show us the forest from the trees. But until these methods of research and data analysis fully mature, human researchers will still be needed to ask the right questions that impact clinical practice and physicians will need to make the humane decisions that impact the lives of our patients. In arthroplasty, these battle lines are already forming. Questions such as whether a patient is an acceptable candidate for surgery and whether care should be delayed or denied to patients who do not fit certain profiles confront us on a daily basis. While we should definitely be making data-driven medical decisions, our ability to consider nonquantifiable factors such as patient insight into their condition and motivation is what separates us from the machines and is why our patients seek our treatment and comfort. In that way we have to be willing to partner with our patients. Amit Ray, an artificial intelligence scientist, once wrote, “As more and more artificial intelligence is entering into the world, more and more emotional intelligence must enter into leadership.”1 Machine learning and artificial intelligence in orthopaedics should be embraced because they have the potential to improve the quality of our research and thereby eventually improve the care for our patients. However, as with any maturation process, several kinks will need to be worked out. We will need to learn how to integrate this technology into patient care, research, and the education of future surgeons. Consequently, while artificial intelligence may represent the future, it is not quite ready for prime time.

Ähnliche Arbeiten