Letter Re: Algorithms


I was particularly interested in your 1/9/16 link to an article regarding The risks — and benefits — of letting algorithms judge us. Algorithms are convenient tools and are more ubiquitous in society than you might think. Those maddening automated telephone answering scripts that lead you through a labyrinth of options that do not address what you are calling about are an excellent example. Many businesses and healthcare venues (or providers) regularly use algorithms to help them maintain a minimum standard of service. That is fine, if you are satisfied with a minimum standard of service.

I tried, at the healthcare system I retired from, to allow deviation from algorithms for valid reasons without employees fearing adverse consequences. My reason was my calculation that, even if the algorithms used were “perfect” and were fully complied with, there would be 30 patients per day who would not be best served by them. In health care, this can have adverse, if not catastrophic, consequences.

What constitutes a “perfect” algorithm? Obviously, one hopes to use this tool to address the vast majority of issues/demands/needs of the target population. The algorithm will work best for those who inhabit the fattest part of the bell shaped curve of that population and will still be useful for those to either side, all the way out to the asymptotes (the narrow tails of the graph). What about those IN the asymptotes? By definition this will be 2.5% of the population at either end of the bell shaped curve. That is 5% of the population, or one in twenty.

Returning to my telephone answering example, I must be the mutant who is among that 5% most of the time. There are two other explanations. First, the algorithm they are using is far from perfect. Second, they are limiting options to guide behavior (as noted in the article above which stimulated my discussion). If an algorithm is nothing more than a series of “If _____, then_____” directions for the user, one can easily see that there can be an insufficient number and type of “If, then” branches in the algorithm, or the “thens” recommended can be faulty or manipulated to achieve a desired result. If the desired result is noble, great. If not, then not so great.

Think about this when you use the apps in your smartphone and ask yourself if they know all the facts pertinent to your situation before completely and blindly following them. While most of the apps we use will not lead us astray, I have seen numerous instances in the medical field where blind reliance on algorithms led to an unwanted result. It might be an excuse for some to say that they followed the algorithm to the letter to achieve a minimum standard, but again, should you be satisfied with a minimum standard? If the algorithm said the bridge should be sound if you load it with cars, but you then add a truck and it falls down, is that acceptable?

If I have a point, I guess it is that you should take responsibility for your actions and their consequences rather than blaming an algorithm if things go wrong. They are just tools. My attempts at carpentry are not pitiful because of my tools, they are dismal because I am inept at carpentry and should either avoid it at all cost or learn to do it right. Until I do so, I recommend not sitting on any chairs I craft. Remember this and do not accept excuses from businesses or government who rely completely on algorithms, whether their intentions are good or ill. Remember what the road to hell is paved with. – K