“Last century machines proved they could replace human muscle. This century, technologies are proving they can outperform human left brains – they can execute sequential, reductive, computational work better, faster and more accurately than even those with the highest IQs.” (Chaffee, 44)
Consider lawyers for example, dozens of inexpensive, if not free, legal services and documents are available at the click of the mouse. Whereas before one would have to pay a lawyer for hours of service to create the same legal document or dispense legal advice. Don’t be fooled lawyers are still in high demand. Technology cannot negotiate a settlement or convince a jury of your innocence. Another example, if your accounting job hasn’t already been outsourced to a $500- a-month accountant in India then surely TurboTax will eventually replace you. Basically, any job that can be reduced to a set of rules may be at risk.
Robot Radiologists may soon analyze your x-rays as Artificial Intelligence (AI) is changing the delivery of healthcare. Perhaps it’s too soon to think about replacement, if ever, but it can dramatically affect the practice of radiology. AI is already helping us to optimize workflow, facilitate quantitative radiology and find genomic markers not seen by radiologists. Our friend IBM’s Watson has been lending computing power to medicine for multiple years being in the forefront of AI. But there is fear and excitement over the AI application in radiology.
Using AI in radiology got its start in CAD, first in mammography and now being applied to other procedures. Clinical radiologists have a wide variety of opinions about how much added value CAD brings to a procedure. Watson demonstrated that machine learning yielded major increases to computer processing power and the application for medicine, but IBM has yet to commercialize the technology. To be clinically relevant, AI must be the result of massive amounts of deep learning. The most common application in clinical medicine currently is in speech recognition and there is much room for improvement in the technology. To apply AI to image evaluation, massive data sets must be available to develop the algorithms. After all, it took Google millions of examples for their deep learning software to recognize the difference between cats and other furry creatures. Facial recognition algorithms are already being used on Facebook to tag friends and cell phones are implementing it as an unlock/sign on feature.
If you mention AI to a hospital executive their eyes light up with the possibilities especially as the database of patient information continues to grow. Clinicians don’t see it in the same perspective. There currently are few practical applications in use aligned with workflow and clinical priorities. The shortage of outcome data continues to delay the application of machine learning to the electronic health record (EHR).
One of the most significant problems facing physicians today is the overload of patient information to sift through. Many physicians compare it to trying to drink from a firehose. As patients begin to upload information into their EHR, the information will increase. Combined with the increasing number of patients to be seen and the data available, medicine only becomes more frustrating. AI could review data and finding anomalies or pertinent data streamline the process. This could be as simple as sorting the data according to the prescriptions, testing, and history relevant to the patient’s diagnosis. In finding the anomalies such as lung nodules or pneumothorax or a malpositioned feeding tube, AI can improve treatment. Who will be the primary winner to commercially make AI relevant and effective? We will continue to watch closely. The AMAZING RACE if on!
Chaffee, J. (2014). Thinking critically. Boston: Wadsworth Publishing.