Algorithms are used everywhere in modern society online. Though they appear neutral and really effective, they can be biased.
What are algorithms?
An algorithm is a set of instructions that a computer follows. An example is the Google search algorithm: you type in a word into the search bar, and Google finds the results most relevant to what you wrote. Facebook uses algorithms to choose what posts to show on your feed, and in what order.
What’s the problem with algorithms?
Algorithms are generally good, as they provide a very efficient way to reach an accurate result. A standard Google search takes less than a second. However, they are not perfect. They seem to be neutral as they follow a set of instructions, but in fact they can be very biased. Biased means having a preference for someone or a group of people over another.
How can algorithms be biased?
For example, some schools in the USA use algorithms to judge a teacher’s performance. If a teacher’s students get really bad grades on their exams, the teacher gets fired. The book Weapons of Math Destruction gives the example of Sarah Wysocki. According to all of Sarah’s past students she’s a good teacher. One year, Sarah received a class that had really high grades the previous year, potentially by cheating. These students then get much lower scores the following year, and Sarah gets fired because the algorithm gives her a very low score. This is unfair as she is a good teacher.
Should algorithms even be used to judge a teacher?
In general, judging a teacher on what a few students do in a given year is not enough to determine whether a teacher is good. There are lots of factors that could affect a student’s performance such as their family situation. On top of this, judging teachers on grades creates an incentive for teachers to help their students cheat to get higher grades. Teachers who then get a class with artificially high grades have to try the impossible task of matching false grades.
What’s the greater issue with algorithms?
The issue is that algorithms seem neutral. However, a lot of algorithms can reproduce human biases. This is especially dangerous as algorithms are unable to correct themselves. The people behind the teacher’s algorithm can simply say that the bad teachers have left. Yet in fact, the teachers leaving could be good, such as with Sarah. Especially bad is that the process is not open: Sarah could not appeal the decision. Algorithms can have unfair outcomes and reinforce inequality.
How can we improve algorithms?
The first step to improve algorithms would be to know how the algorithms work! We don’t even know exactly how Google’s search algorithm works, or Facebook’s. Then, we should examine the results more closely: for example, does a teaching algorithm fire good teachers? Overall, algorithms are really useful, but we need to pay more attention to their biases.