Recommender systems are the most commercial offshoot of AI. Many of the largest companies, including Meta (Facebook) and Alphabet (Google), make their money from advertising. When the companies get paid by clicks, the profitability is directly related to how well they target advertisements to individual users, and how well they keep people engaged on their platform. Streaming services, such as Netflix, have competitive advantages if good recommendations help to keep people going back to their sites.

These systems use a diverse collection of AI tools for their recommendations. Steck et al. [2021] describe challenges that arise in the Netflix recommendation system, in particular, integrating deep learning into their recommender systems. “Through experimentation with various kinds of recommendation algorithms, we found that there is no ‘silver bullet’; the best-performing method (whether deep learning or other) depends on the specific recommendation task to be solved as well as on the available data. For this reason, different kinds of machine learning models are used to generate personalized recommendations for the different parts (e.g., rows) of the Netflix homepage” [Steck et al., 2021].

The recommendations of social media companies trying to maximize engagement, and so their profits, leads to increased polarization [Sunstein, 2018; Levy, 2021]. People engage more with extreme views than with moderate, considered content, forming filter bubbles and echo chambers where users only communicate with like-minded people [Acemoglu et al., 2021]. Because the content is targeted, public people, including politicians, can tell different people different, even inconsistent, messages. With technologies where the messages are public, many people moderate their messages to not turn off other people. This mix of relative privacy and optimization for engagement makes for toxic online forums.