In a world that runs on machines, algorithms are all around you. Whether you are reading this article, watching some video, or browsing the web, it is because an algorithm brought it to you. When you open any social media platform, algorithms decide what you see. When you search for a photo, algorithms simplify the process of finding it. When you buy something from E-commerce websites, algorithms help adjust future prices for companies while also checking your bank account information to safeguard against fraudulent transactions.

While you may be aware that algorithms are everywhere, knowing is only scratching the surface of understanding how algorithmic bots work. In order to function autonomously, algorithmic bots are first given specific instructions by humans in the form of if-then conditional statements. For example if a user wants to search for a picture of a balloon, then internet algorithms will filter through images until the best balloon pictures are displayed. In fact, the capabilities of algorithms are so great that they can function efficiently even when problems are too large and difficult for humans to even write into a simple form. For example, algorithms can identify fraud transactions in seconds, offer recommendations unique to each user, and set airline seat prices at the maximum price that users will pay. While the algorithmic bot answers to these questions may not be perfect, these answers are far more accurate and efficient than if they were computed manually by humans. This is why the functionality of these bots is still a mystery even to the people who created them.

So how can a bot be built if humans do not understand how their brains function? To begin the process, a specific task such as differentiating a pen from a pencil in an image is chosen. While this step is simple to teach humans, complications arise when teaching it to bots because bots can only understand machine language. In order to resolve this obstacle, two specific types of bots — builder and teacher — are deployed.

After it is created, a builder bot constructs bots called student bots, which are sent to the teacher bot to be refined. The teacher bot accomplishes this by giving the student bots a series of tests and measuring their accuracy. Initially, student bots will score less marks on the test as they were built randomly; however, once the builder bot receives the test results, it can separate bots with good performances and recycle the rest. The builder bot then modifies the left over bots and continues to build newer bots that are better equipped to pass the teacher bot tests. As this process continues, the average student bot scores as well as the standards needed to pass the teacher bot tests will continue to rise, ultimately resulting in highly trained and effective bots. Student bots originally trained to differentiate a pen and pencil in a picture will be able to flawlessly differentiate between a pen and pencil even from pictures never seen before once they “graduate.”

The strange thing about this process is that neither the teacher nor the builder bot knows exactly how the student bot learns. In fact, not even the humans who build the first bot know.

After being generated at random and constantly changing, the network inside the student bot’s brain becomes very complicated resulting in another road block for the bot: it is very good at what it does but inefficient when the problem given to it is slightly changed. The bots taught to differentiate a pen from a pencil in the previous example would likely struggle to differentiate them in videos and inverted pictures. Humans remedy these roadblocks by including bot responses for unknown situations and continuing to accumulate data as more data results in more tests which makes the bots better.

Even though we only have a rough understanding as to how exactly student bots gradually learn to ace the teacher bot tests, the fact of the matter is that behind every website or platform there are tests to increase user interaction or pick the posts that you and your friends will like and share the most. We must be aware that our algorithmic bot buddies are all around us as we continue to refine our understanding of them.

‍