As I am Canadian born, I must first apologize before rendering an insult. So please pardon me when I say that anyone who does not appreciate the absolute need to develop a significant familiarity with machine learning, is a fool. [N.B. Machine learning refers to a computer system automatically finding the formula that best fits the existing data, which can then be used to predict future results based on new data].
Of course, once again switching to Canadian mode, it is everyone’s right to live their life as they see to be correct. And this world would be most likely far better off if there was mutual respect for even the most far left and far right points of view. While I of course detest anyone who has Nazi sensibilities, as long as this person does not manifest them in a way that is hurtful to others, I would resist expressing my personal distaste of such a person. I would not have them join me at my Shabbat table. But I also would not call out to their mass extinction. It seems that Canadians are willing to turn both cheeks.
So while I respect the right of an individual to act stupidly and shortsightedly, not having a familiarity with machine learning leaves a person vulnerable to those who do. More so, as I have stated in previous blogs, physicians have absolutely no excuse for not becoming, at the very least, low-level experts in machine learning. A doctor should not need to go to a separate specialist to run a machine learning algorithm on the data that he or she has collected, as part of any type of research study.
After initial analysis and conclusions, it is then reasonable to consult with a true expert in the field. It could be that the true expert decides that the algorithm used to achieve machine learning on the specific data set was not the best choice. During a lecture from Stanford University specifically on machine learning, the professor noted that on brief perusal of a colleague’s work, he immediately recognized that the algorithm chosen was incorrect and had led to a waste of six months of work.
Machine learning is by no means a technical process. It is not a question of “choose A&B, AB and C, or D only”. To effectively implement machine learning, one has to make certain assumptions and has to develop an overall understanding of the message within the data. Doctors, as of today, are not at all trained to do such things. In fact, in many research situations, doctors pass their data onto a statistician [who charges a very significant sum for his skills] in order to derive the necessary conclusions for publication. This is nothing short of ridiculous. The researchers would argue that their forte is actually doing the research, and it would be a waste of time to focus their efforts on understanding such concepts as chi-squared. But the researcher who doesn’t understand the fundamental concepts of what is presented in the paper he or she wrote, is unable to validate what is being attributed to his or her name.
I remember doing a course in epidemiology as a medical student. I found it minimally useful. I remember doing three courses in statistics over the course of my junior college years and university years. Once again, I found them effectively useless. Now it is true that all of this could be a reflection of my repeated brain trauma from being used as a substitute for a football by my older brother and his friends. But to be fair to myself, I found the courses poorly taught and leaving me with no practical skills outside of deciding when the train will arrive in Detroit if it leaves Toronto at the speed of 80 km an hour.
I have presently taken upon myself the task of spending an hour a day studying from various online free resources that altogether will re-teach me algebra, probability, differential calculus and other necessary topics. All of these are required if I am to understand the theory behind machine learning, rather than just learn how to plug numbers into an automated interface. I will already tell you that it is a painful process to have to hand calculate the standard deviation of a series of numbers. But I hope and pray that by the end of the year, I will have remastered the math that I need to truly understand the theory behind machine learning.
I am doing all of this for two reasons. First of all, despite my background in both computer science and medicine, and years of experience in developing a very successful electronic health record, I can’t get a job. I am 54 years old and really cannot compete with kids who already have not only mastered big data analytics and machine learning but far more. The second reason, which is by no means less important to me, is that machine learning will be the lingua franca of the upcoming 1 to 2 decades. Anybody who is not versed in big data analytics and machine learning will simply have a blank look on his face whenever these topics are described.
When a paper is presented to a medical audience and the author explains how a deep learning algorithm was able to determine that parameter X predicted Y, a good part of the audience will stare blankly at the screen. This is not a successful formula for passing on new information to the medical crowd. In fact, the medical audience will wait for the conclusion slide that provides a simple formula that from this day on, a 10% rise in blood pressure is equivalent to a 20% rise in cardiovascular risk [these are made up statistics].
The fact that few doctors in the audience will be able to validate this data is pathetic. The fact that even fewer doctors in the audience will be able to take this data, which will reside in the open cloud, and combine it with other data from other research to find other correlations, is even more sad. Doctors claim that there will always be a need for physicians even as artificial intelligence continues to develop. The question is how you define the term doctor.
Unless there is a fundamental and massive change in medical school curricula, most doctors will truly be reduced to technicians who simply act on the conclusions of those doing actual research. Those doctors who embrace the whole new field of data analytics will be in a class of their own, and will financially be far more successful than their colleagues. I specifically include this last sentence, as financial interests tend to be of great import to most doctors. So at the very least, don’t study calculus for the knowledge, don’t study physics for its help in managing patients, but study math for the promised increase in income that it will provide.
PS. Canadians are polite. I never said we are not cynical.
Thanks for listening