Par Shawn Fuller

Rédactrice collaboratrice

Algorithms 101

In its simplest form an algorithm is any set of steps that, if followed, will accomplish a goal. For instance, a recipe is an algorithm that a human follows to make a meal.

The traditional computer algorithm should provide the same answer or do the same thing each time it is given the same input. Increasingly tech companies use machine learning algorithms for many of their systems. Algorithms developed by machine learning don’t follow an ordered set of steps. Instead, they work more like the perceptual systems of the human brain, which processes visual input through successive layers of neurons, each layer specialized to extract different kinds of information. The software that supports machine learning mimics the interconnected neurons, whose behaviour changes with learning. The algorithms of social media take in information about our behaviours, make decisions based on their own hidden learning, and feed back to us the results.

Hannah Fry summarizes the four major categories of algorithms that are used by Google, YouTube, and Facebook in her book, Hello World: Being Human in the Age of Algorithms. They are prioritization, association, classification, and filtering.

Prioritization algorithms rank things according to criteria such as popularity, or ratings. When you search YouTube for videos of hurricanes, cats, TV bloopers, or how to repair your dishwasher it uses prioritization algorithms to bring you the most popular videos, ranked according to the number of times a video has been viewed by other people. It also displays videos that you may also like to watch, based on their association with other videos that people have viewed.

Association algorithms find connections between things. They are what Amazon uses to display other books and products that you may be interested in based on what you just searched for. It is the association algorithms that may create a radicalization pipeline in which the recommendation engine offers increasingly extreme videos. The media scholar Zeynep Tufekci observed that after watching a number of videos of Donald Trump rallies on YouTube, the site began to autoplay videos featuring white supremacist rants and Holocaust denials. When she started watching videos of Hillary Clinton and Bernie Sanders, YouTube started recommending left wing conspiratorial videos filled with allegations of secret government agencies and 9/11 cover-ups. In response to public outcry and media attention over ISIS and extremist videos, YouTube began removing videos that directly preached hate or incite violence. They also adjusted their ranking algorithms to de-amplify those that promote conspiracy theories and pseudoscience. Now, if you look up chloroquine and Covid 19, for example, YouTube will use its association algorithms to connect you to videos by medical researchers, who provide a more factual view of the topic.

Classification algorithms attempt to place you in various categories. The massive data harvesting that social media applications and data broker companies engage in is aimed at placing you in demographic and behavioral categories in order to target ads for products that might interest you. It may comfort you to know that these algorithmic guesses are not entirely accurate (yet). Based on my signed in activity Google thinks I am not a parent. Meanwhile, my daughter’s google account thinks she is interested in cars and basketball (neither is true).

Filtering algorithms remove or exclude information that is considered noise or not of interest. Siri and Alexa need to filter out the background noise in order to recognize your voice commands. Similarly, your phone filters out background noise while you are speaking to someone. This sometimes leads to the perception that your call has been disconnected during a pause in the conversation. (“hello? Are you still there?” “Yes, I’m still here”) The social media apps use filtering algorithms to include only the stories, memes, and videos that match your known interests.

Us

In his Wired article, Steven Levy draws from a 17-page chunk from the (mostly destroyed) journals of Mark Zuckerberg to examine a key point in the evolution of Facebook. One of the preoccupations within this journal was a product that he called Feed (later named News Feed). This signalled a dramatic change to the Facebook concept, which would later have real-life impacts on global events in war and in politics. At the beginning, Facebook required more active effort on the part of users to connect with friends. You had to go to your friends’ profiles in order to see what they were up to and post your updates on their “walls”. This was very much in line with the active participation that defined the web 1.0 world. The news feed changed all of that.

Now your friends’ updates were pushed to you without any effort on your part. The news feed would make it easy for people to see what was important among the friends with whom they were connected. Steven Levy notes that Zuckerberg used the word “interesting-ness” to cover the kinds of stories that should appear in a news feed. They needed to be “centered around your social circle.” There was to be an emphasis on changes in personal relationships and life events. The new feed was to favour these over interesting events or other information outside your social circle.

The news feed when it launched in 2006 appeared to be disastrous since it had major privacy flaws and people protested heavily against it. But it was also clear, once Facebook employees looked at the data, that people were spending more time on Facebook. It was the news feeds themselves that helped the protests to go viral. Facebook fixed the more serious privacy issues, the protests died down, and Facebook continued to grow exponentially.