Algorithms Are Not Magic

We hear a lot about algorithms when we are poking around on social media. Someone, somewhere has programmed things in a way that we are more likely to see things that interest us, and of course, more likely to see ads that lead us to buying stuff.

In the course of several conversations over the last few days with respect to my work in the world of technology and oncology, algorithms have been coming up again, too. One would think that the algorithms necessary for cancer research would be more “special” than the ones for social media.

To think this through, I wanted to find the simplest definition of an algorithm. So I Googled, and Google’s algorithm brought up this definition that I clicked through to from a kids’ site.

“An algorithm is a set of guidelines that describes how to perform a task. Think of an algorithm as step-by-step instructions that create a predictable pattern in a set of numbers or in lines of code. Mathematicians, engineers, and computer scientists develop and implement these “instructions” to provide real-world solutions.” (https://www.idtech.com/blog/algorithms-for-kids)

Step-by-step instructions. Guidelines. Performing a task. That is all pretty cool. Especially when it comes to oncology. Algorithms that map the latest in cures to the exact contours of a cancer patients’ latest biomarkers. Pretty cool, right?! So, it would seem that all we need to is create the right algorithm, and voilá!! Well, there is one more piece to the puzzle. “Computer algorithms work via input and output. They take the input and apply each step of the algorithm to that information to generate an output.”

So, for as powerful as the algorithms may be, it still depends on inputs. And, that is where the problem is with oncology. The inputs are everywhere. Literally, everywhere. Databases at a cancer center. Databases at pharmaceutical companies. Databases at testing centers. The inputs are scattered, and so, the algorithm’s outputs are only as good as the inputs. Without the necessary inputs, the algorithm instead makes assumptions. Finds the inputs it can as substitutes/proxies; tries to run its set of guidelines to take what inputs it does have to basically “guess” at a match.

So, rather than taking an even deeper dive into the technology, I will pause here and observe two things. One is that algorithms are not magic. People make them. People create the guidelines. The routines. There is not just math to think about. There are ethics. When those algorithms are in black boxes, we are unaware of its assumptions, and therefore blind to how conclusions are reached. Second, inputs matter. Not only do we need to make it possible for the necessary inputs to be more readily accessible, especially with say oncology, but we also need to have individual control of our inputs. There are ethics here, too. Privacy, security, and more. Not only do we need to know what’s in the algorithms, but we need to know which algorithms are using our inputs. And, that shouldn’t take too much magic.