I think any system based on big data that results in analysis, predictions, recommendations and/or insights is prone to be innacurate, biased, or influenced. This is not a new proble is just becoming more evident with tech waves of Big Data, Machine learning among others and unstopable media attention.
Models by definition are exactly like any system garbage in garbage out, models are a reflection of their creators and their environment at least that used to be the thruth, if you see for example credit risk models, they havent change that much over time and tha baseline of criteria is about the same such as income, spenditure rates , tenure in residency and job etc, however how they should evolve in order to align to new reality of society such as low home ownership, high geo and work mobility among others that tends to discriminate younger generations from the benfits of a model.
Recommender systems are more and more being used to help in decision making in things beyond buying your new consumer product or your new movie but more critical things such as prioritizing manufacturing lines, routes for transportation, or even things to lay-offs design or hiring decision support, if the recommender system is not only designed with care but in a multidisciplinary basis such a s professionals beyond datascience it can bring or increase human bias.
Algorithms are well known to amplify bias and negative connotations (e.g. mortgage crisis in the US), so taking care of use data that promote inclusiveness and neutralize things like race, origin, in general social bias generators even on penalty of sacrificing accuracy will be better than constantly amplifying personal bias.