Recommender Systems are the AI that suggests YouTube videos or Netflix shows you might like, curates social media posts based on your interests, or creepily shows you ads for a book you were just discussing with a friend. This AI combines supervised and unsupervised learning systems, meaning that it works with data sets (unsupervised) and also responds to the decisions you are making (supervised).
Most recommender systems combine three recommendation types: content-based, social, and personalized. Content-based recommendations ignore the user and recommend based on the quality or recency of the content. Social recommendations favour what is most popular, based on likes, subscriptions, amount of purchases, etc. Personalized recommendations are based on what you, specifically, are interested in. For example, a YouTube video might be recommended to you because you have previously engaged with that channel, or because users similar to you watched it, or both.
A few issues with recommender systems include the creation of ideological echo chambers – where we are trapped in a bubble of people who think exactly like us, which can have serious social consequences. Additionally, recommendations can be based on harmful stereotypes. Less serious issues might be missing a show you would have liked, because the AI thought it didn’t match your preferences, or seeing ads for products you just bought, or websites you just visited.
Each of us experiences a different version of the internet. Recommender systems aren’t going anywhere anytime soon, and to coexist with AI in an ethical and knowledgeable way, it is our responsibility to understand how AI influences our everyday lives
