Extreme views and divided societies have existed long before the Internet. What did not exist until the last couple of decades is the increasing use of AI-based algorithms that hyper-personalize the content being served via social media platforms - extreme content that drives engagement and advertisement sales. These AI-based algorithms are essentially playing against themselves – they continue to make the models better and better, which means that ‘personalized realities’ for individuals are becoming more and more extreme and divisive over time.

Many individuals and groups are well aware of this phenomenon and are brainstorming various solutions that come with the realization that the only way we can fight the speed at which AI-based models are being built, is to build competing AI-based models. For most AI-based algorithms, the builder must define either a target or goal (reward function) that the algorithm is trying to optimize. In our current state of social media business models, that target is ad revenue and the winning strategy is serving more personalized and extreme content. Some of us are suggesting that a healthier target for democracy could be a ‘movement towards common ground’ with a strategy that has not been defined by AI yet because we have not tried to tune the models to aim for this state. While there are questions as to how we can measure this ‘movement towards a common ground’ and how this could be a profitable business model for a social media company, there is a clear need for such a shift: one could just look at various regions of our country and immediately see that things are trending toward destabilization that makes it difficult for businesses to even operate.

To avoid this situation, people must come closer together in consensus on various matters. The goal is not assimilation or brainwashing, but rather, enough consensus so that diverse people with a plurality of views/backgrounds can live prosperously and peacefully.

AI has been a part of breaking this democracy apart, thus now we need to consider ways to bring it back together. If we start to go down this path of building models that drive consensus amongst people within a diverse society, we can start to use those models to assess the content and laws being produced/sponsored/supported by politicians. This could be immensely helpful to individuals who, instead of simply voting according to a party line, are willing to vote for those politicians who bring people together rather than driving them further apart.

We could use the growing body of work focused on making AI algorithms more transparent and explainable to gain AI-based knowledge on what kind of things in society build consensus versus division. We could use these findings to be more informed when speaking about the strengths and weaknesses of the society that we live in. We could also develop AI-based models for ‘consensus building’ from data that is sourced from other countries and cultures, thus giving us a better picture of how we can relate with those that sit outside of our geographic borders.

Does this sound like too wishful of thinking? If so, let me ask you one question: Do you feel comfortable watching our society and democracy continue to be fueled by AI algorithms that are essentially built to promote extreme and even violent views and perspectives?