Polarization in The Digital Age: Lessons from the Past for the Technology of the Future

As prior work has demonstrated and many historical examples suggest, the level polarization present within a society fluctuates over time. However, the role of shared information between a given society’s polarized elements in this fluctuation has not been subject to thorough study. I seek to demonstrate here that there is a strong argument to be made that shared information plays an important factor in mitigating polarization, and that changes to the dynamics of information sharing and production due to technological advances and resulting market changes have contributed to an exponentially increasing variance in the political information consumed by members of American and Canadian society, which in turn increases polarization. This suggests that an intervention that could mitigate the levels of information disparity between consumers of political information would reduce the level of polarization within this population and allow for political conflicts to be resolved more amicably through more mutually acceptable compromises.

We may begin by first defining what is meant by the term ‘polarization’. Political psychologists and scientists have traditionally not understood polarization to be a single phenomenon, and instead have identified several types of polarization that are understood in the context of this essay to coalesce into the broader concept of polarization. First among these subtypes is ‘political polarization’, which indicates the degree to which political opinions diverge from a center position within a given population. Political polarization is highly related to but does not strictly imply animosity between these divergent opinions and the groups that hold them. This is defined by ‘affective polarization’. Some scholars further make distinctions between ‘political polarization’ in reference specifically to support for political parties, and the ‘ideological polarization’ that defines the actual differences in policy opinions between individuals within a given population. Thus, some would consider a country with many highly ideologically distinct political parties to be ‘politically polarized’ and ‘ideologically polarized’ but not ‘affectively polarized’ so long as they get along, and another country divided by parties whose policies differ only slightly but who nonetheless hate each other to be ‘affectively polarized’ but not ‘politically’ or ‘ideologically’ polarized. To avoid as much confusion as possible, the term polarization within this essay will be used to refer primarily to affective polarization, which is technically distinct but strongly influenced by political and ideological polarization.

Having defined our primary term, we must now note that although polarization has increased starkly in the post-2010 era, as evidenced by diminished political consensus between parties and partisans, stark increases in negative views of opposing partisans, and increases in political violence, polarization does in fact ebb and flow in all complex societies over time. In fact, polarization varies as a function of factors influenced by but not dependent on external pressures or technological advancement. While I will argue that changes to the information economy and the resulting information disparities due to technological advancement in particular have played an exacerbatory role in this process, I will not argue that it is the sole or even primary cause of the present elevated levels of polarization. The role that information disparity plays within this process is that of an interactive effect with already elevated levels of polarization, caused by economic, sociological, and demographic factors, the origins of which lie deeper into our history than the beginning of the digital age.

While this essay is not one advocating the position of historical determinism, it must be acknowledged that scholars such as Peter Turchin have demonstrated that cyclical long term demographic-economic processes play a large, independent, and replicable effect on the level of social unrest and division present within a given society. Much like the business cycle, societies experience regular generations-spanning periods of positive growth and expansion, as well as negative periods of contraction and strife. During a positive period, a society enjoys the high degree of social stability that we would associate with low polarization. It is characterized by higher social mobility, a lowered cost of living for the general population, and lower income inequality. There is polarization, yes, but this is overshadowed by a high capacity for cooperation, allowing the society to effectively mobilize to take large-scale action in response to changing circumstances or higher-order goals without jeopardizing its internal social harmony. But in this period of prosperity, the seeds of future decline are sown.

This has to do with the fact that in all complex societies, class divisions exist. A disproportionately large share of wealth, power, and social prestige is always held by a disproportionately small share of the population, whether by the result of hereditary privilege, meritocratic stratification, social distinction, or some combination thereof. This holds true historically, where this upper strata was defined by classes such as the nobility and lower gentry, and even today in democratic societies where it is defined by the upper and middle classes. Those born out of these segments of the population aspire to join them, and those born within them aspire to retain their position or climb even higher. There are many ways whereby a position within this elite strata can be achieved by what Turchin labels an “elite-aspirant”. Entry into a prestigious profession, the state bureaucracy, and powerful institutions such as the church or military all serve as historical and contemporary examples of how individuals from various social stratum can gain access into the ranks of the ‘elites’. However, the higher one climbs, the fewer available positions there are, and this naturally produces a level of competition among ‘elite-aspirants’.

The problem with this is that because families that do achieve entry into the elite strata naturally want to pass on their increased quality of life and social privileges to their children, they tend to leverage their increased access to resources to secure competitive advantages in the pursuit of these positions for their children, through means such as higher education. At the same time, those not in the elite class still want themselves or their children to gain access. Thus, over time during prosperous periods, when social mobility is high, the pool of ‘elite aspirants’ grows while the requirements of entry into the class get higher and higher.

Within our own society, we can see this most clearly at work through the example of the simultaneous devaluation of the university degree’s value and the increase of its cost. Whereas in the 1960s average tuition in the United States cost mere hundreds of dollars and was widely viewed as a job ticket into the middle class, it now costs tens of thousands to obtain a bachelor’s degree considered not to be the qualification for most jobs, but the bare minimum to apply. Even when tuition costs are adjusted for inflation, or calculated relative to the median wage, this reality is undeniable. Despite this, demand for university degrees is not falling, but increasing rapidly. Between the years 2000 and 2018, the share of the American population over 25 with bachelor degrees increased from 29.8% to 48.2%. Those with master’s degrees went from 10.4% to 20.2%, and with PhDs from 2% to 4.5%. This is a dramatic increase in credentialization that has corresponded not with an increase in general prosperity over time but instead a substantial decrease, as cost of living has skyrocketed while real wages have stagnated.

Within Turchin’s model, these are strong indicators that we have entered a period that will be characterized by elevated levels of ‘intra-elite competition’ and ‘popular immiseration’. This is precisely when societies enter their most polarized states, and can even destabilize entirely. It is a time where more and more ‘elite aspirants’ will be frustrated and disillusioned with the society that has failed to deliver on their expectations. It is a time when more and more families, even those who have ostensibly already made it into the middle or upper classes, will be forced by rising costs to cease to enjoy the standard of living to which they have grown accustomed, or see their children fail to achieve it for themselves. Worst of all, it is a time that only seems to worsen year after year. The harder the economic situation gets, the more people compete to obtain the positions that will deliver them from economic insecurity, and the harder the competition gets, which drives up the costs of competing and worsens the economic situation. The more competition for prestigious and well paying positions there is, the greater the pressure on the state and state-funded institutions to create more of them to satisfy the discontented ‘elite aspirants’, which when done leads to higher state spending. This in turn raises inflation and requires higher taxes, worsening the situation in the long run while still failing to keep up with ‘elite aspirant’ demand in the short run.

Under these pressures, social unity is undermined. Class interests begin to diverge. Educated and capable elite aspirants are produced that see rising incentives to attack the system and demand reform at the precise time that existing elites, who see their own positions as more precarious than ever, have every incentive to bitterly resist such demands. These are the conditions that, as Turchin demonstrates, preceded the collapse of republican Rome, the Russian Time of Troubles, the 14th century collapse of the French Plantagenet dynasty, the English War of the Roses, the 17th century English civil war, the American Revolution, the 1789 French Revolution, the American civil war, the collapse of the USSR, and many, many, many more. This is not to say that conflict is inevitable during these periods, though societal re-organization must occur at some point to resolve it before it reaches a crisis point. For example, according to Turchin’s work the United States underwent such a period during the 1920s that was able to be resolved through the implementation The New Deal, which substantially re-ordered the economic arrangements of the day and laid the groundwork for the post-war economic boom that only began to decline in the 1970s.

So where does information disparity fit into this story? It is my suggestion that while long-term socio-demographic variables set the stage for times of increased polarization, it is information disparity that prevents societies from properly understanding the position they find themselves within and achieving political consensus regarding the necessary solutions. This is because variability in political opinions is informed by the variability in which social reality is experienced by individuals. Perceived social reality produces a mold through which individuals form their moral systems, epistemic beliefs, and incentive structures. As such, numerous characteristics that correspond to these differences in which social reality is perceived correlate with differences in political opinion, including age, sex, race, religion, region, sexual orientation, education status, and of course, social class.

As alluded to earlier, several indicators suggest that the United States and Canada have become increasingly polarized in recent decades and especially within the post-2010 era. As recently as 2019 a survey found that approximately one-fifth of American partisans in both parties believed that violence against the opposing party would be at least “a little” justified if their party lost the 2020 election. Between 2020 and 2021 the share of American students surveyed who said violent protest was “never acceptable” dropped from 82% to 76% and at most elite schools it was even lower. Over the same course of time, self-reported affective and political polarization has been increasing rapidly, with studies indicating that the affective polarization between political groups in the United States is on par with racial group polarization, and that partisan identity significantly predicts non-political judgements of the out-group.

These increases have corresponded with the simultaneous increase in examples of precedent-breaking polarization events such as 2014 Ferguson riots, the 2016 election of Donald Trump, the 2017 UC Berkeley riots, the 2017 Charlottesville Unite the Right riots, the 2020 Black Lives Matter protests, and the 2021 storming of the American capitol building. These can be considered to be paralleled in the Canadian context at least in part by the nationwide 2020 pipeline protests and the 2022 Freedom Convoy protest in Ottawa. What is of concern to us here is that all of these may serve as examples of events that can be considered as political Rorschach tests, where groups ostensibly see the same events yet come away with drastically differing perceptions of what had occurred.

Well, the information disparity hypothesis of political polarization (IDHPP) would posit that this is due to the substantive changes to the information economy that occurred over the period of time in question, pinpointing the true point of division as being sometime in the 2010s. At this time, changes were made to the architecture of Facebook and Twitter, such as the introduction of the ‘like’ button and the ‘retweet’ button that over time would allow these platforms to alter their timelines from being primarily chronological to being run by algorithms designed to more readily promote hyper-virality by catering to user interests. In addition to significantly informing the trajectory of online culture over the coming decade, this change has also been hypothesized to have paved the way for a political landscape increasingly characterized as hyper-polarized and rife with misinformation.

Prior to this shift, the information economy was characterized by a relatively smaller number of actors, with larger media entities commanding a significant degree of consumption overlap between opposing partisans, through mediums such as cable news, newspapers, and radio channels. This consumption overlap acted as a natural incentive for these media entities to remain as impartial as possible, as well as also ensuring that partisans had access to a higher proportion of common facts and perspectives. This information commonality naturally inhibits political polarization by limiting the potential number of disagreements pertaining to a particular political issue between partisan and establishing an agreed upon “battleground of ideas” over which partisans can try to resolve disputes. In other words, access to similar information produces similar conceptions of social reality.

Under such a model whereby political polarization is mediated by the size of gaps in perceived social reality, large enough gaps result in an endemic inability to resolve political arguments as neither side can effectively argue their perspective to the other because scope of the argument cannot be limited to an established set of discrepancies. This issue compounds over time as information gaps between partisans continue to grow and technological and scientific advancement render issues more and more complex. This would explain the sharp increase in political polarization post 2010, as the rise of social media fragmented the information economy to such an extent that large media entities virtually ceased to have overlap in partisan consumption and competition for viewership, engagement, and clicks increased sharply, incentivizing incendiary partisanship and disincentivizing balanced perspectives due to the confirmation bias driven self-selection of partisans into informational “echo-chambers”.

Until now, few solutions have presented themselves, arguably at least partially because there has been an over focus on the role of political misinformation in this process. Misinformation has always existed to some degree within political discourse, but it became a major public concern during the 2016 presidential election when its alleged rampantness was perceived to have been critical to the success of the Trump campaign. The story went that countless Russian bots were spreading misinformation on social media platforms and that this fooled numerous Americans and persuaded them to vote for Donald Trump. While it was true that the Russian government had organized so-called “troll farms” which bombarded social media platforms with misinformation, little evidence has suggested this had any substantial impact on the election. A study conducted within the past year demonstrated that these posts were highly concentrated among highly partisan Republicans, with 1% of survey respondents accounting for 70% of all disinformation exposures, and that there was no relationship between exposure to the Russian foreign influence campaign and changes in attitudes, polarization, or voting behavior. So while it remains possible that misinformation plays a role in polarization, it seems more likely that belief in misinformation is a symptom and not a cause of the larger issue at play.

Other suggested solutions have included mandatory independent fact checks underneath social media posts on certain subjects, and restricting the reach of posts rated as false or misleading. This strategy was tried by Facebook, Twitter, and Instagram, but received substantial criticism for being neither entirely factual nor entirely independent. For example, a video posted by John Stossel in 2021 which argued that the California wildfires were predominantly caused by bad forest management, and not predominantly by climate change, was rated as ‘partly false’ by the Facebook fact-checking group Science Feedback and labeled as such on the platform. In both the Science Feedback article linked to the label and in subsequent communications between Stossel and Science Feedback, none of the facts included within the video were refuted. Worse, the article included the statement “Forest fires are caused by poor management. Not by climate change.” in quotations, despite it never being said in the video or attached post. On the contrary, Stossel readily acknowledged the role of climate change in worsening forest fires within the video. As a result, Stossel launched a defamation lawsuit, in which Facebook’s lawyers defended the website’s ‘fact-checking’ as merely ‘opinion’ and thereby immune from claims of defamation.

This case is hardly an outlier, and it and others have served only to further undermine public trust in these so-called “fact checks”. Recently X, formerly Twitter, has pioneered a newer model of such fact checks, which instead of employing third party organizations to unilaterally determine the factuality of a given post, provides something called “community notes”. Community notes are contextual facts added by contributors to potentially misleading posts, then viewed and rated by other contributors. Once enough contributors from differing viewpoints rate the contextual fact as helpful, it appears publicly to users. This serves to insulate the feature from claims of outright bias, or as serving the interests of a single group. Whether or not it has any effect on reducing polarization, however, remains an open question, though its cross-partisan nature shows promise.

Others have proposed services that exist entirely separately from social media, that instead strive to directly provide users with cross-spectrum news coverage. The service “Ground News” serves as one such example, distinguishing itself by identifying some of the key problems with reliance on any individual news source or social media platform in information provision. That is, individual news services do not simply report all possible facts on all possible news stories. This is partly because news organizations are themselves not immune to biases, with journalists, editors, and higher-ups all potentially having influence over what stories get covered, what facts are included in these stories, and what tone coverage takes. As alluded to earlier, this is not just because of the opinion biases of these individuals, but also the practical market concerns that media outlets have to consider. If a particular outlet has a customer base that is 85% right wing, confirmation bias dictates that certain stories will be less likely to be liked or engaged with should they cover content that a) is not interesting to that demographic or b) contradicts their preexisting beliefs.

Ground News addresses this issue by compiling news sources across the political spectrum and showing the user all the outlets which cover particular stories, broken down by political orientation. This way users see stories, not articles, right away and can choose which source they would like to read the story from while simultaneously being made aware of their bias. Additionally, Ground News has a service called “Blindspot” which shows only articles covered overwhelmingly by outlets of a certain political orientation, so users can see what right/left wing outlets are not talking about. This feature in particular can be understood as an attempt to bridge the gaps in perceived social reality between right-leaning and left-leaning partisans, as recreating the shared-consumption feature of the pre-digital information economy.

While innovative, this service is not without its limitations. First and foremost, it is limited by the willingness of users to actually examine the perspectives and opinions of those with views and opinions contrary to their own. As we know from the existence of confirmation bias, an unwillingness to do this in the first place is likely a major factor in worsening political/affective polarization. However, if we were to accept that affective/political polarization is in large part caused by gaps in perceived social reality, and that this can only be exacerbated by individuals being exposed to overwhelmingly partisan content that due to system incentives is overwhelmingly likely to be anger-provoking, than it would stand to reason that if an individual was somehow able to switch from this type of content to that which contained more balanced perspectives that that individual would experience less political/affective polarization.

This is something that through the harnessing of the power of artificial intelligence, could be relatively easily created as an entirely new method of information consumption. By utilizing large language models to continuously analyze news articles from across the political spectrum in real time a service could be created that does not show articles to users directly but gives all available information on the stories those articles are meant to cover. This way individuals would have immediate access to all agreed upon information regarding a particular news story, an instant understanding of where the sources in question disagree, and upon what contingents these disagreements rest. This would not only remove much of the human bias present in individual news articles but also would solve the problem of reduced consumption overlap by giving users an understanding of the opposing perspective on any given controversy at the same time that they learn of it. Such a technology would, if sufficiently developed, be able to not only determine through language analysis the points on which any political disagreements are premised on, and identify what information, should it emerge, would lend itself to supporting or detracting from the sides in question in a fashion superior to any individual journalist, but also to directly answer clarification questions on a story that the user may have.

Let us take for example a hypothetical news story that could plausibly occur within the Canadian political context. In this example, the controversy surrounds actions taken by a provincial government during a protest led by first nations groups over the planned construction of a liquid gas pipeline that would pass through indigenous territory. In this scenario, the company has obtained an agreement with the formal leadership of the relevant indigenous nation, the band council, in which a substantial sum of money is agreed to be paid to the tribe through both a premium for the use of the land and a partial profit sharing agreement, in addition to guarantees surrounding employment for tribe members. However, despite the agreement passing through the band council, who under Canadian law have the legal jurisdiction to consent to such actions, some of the leaders of the traditional power structure of the tribe, the hereditary chiefs, vehemently disagree with the decision and have organized blockades that prevent work from being done on the pipeline. Both sides point to the adoption of the policies of the United Nations Declaration on the Rights of Indigenous Peoples by both the Federal government and the relevant provincial government, which specifies that the usage/development of traditional indigenous territory requires the consent of the relevant indigenous people, claiming that they are the legitimate authority that has the power to give consent.

Such a case, even without the countless other details that would be present in a real-world scenario, is inherently deeply complex, offering few clear answers and raising serious moral questions on Indigenous rights, environmental concerns, and resource development. We could easily imagine how such a story could spark considerable polarization within the Canadian body politic. Those that disagree with the protests and side with the development company may emphasize the economic benefits that both the nation and the Indigenous tribe in question would receive being stalled or jeopardized by the protests, they may point to the legality of the actions of the company and the consent of the band council, and point out how environmentally friendly natural gas is by comparison to available fossil fuel alternatives such as coal or oil. Those that agree with the protestors may emphasize and talk about entirely different aspects of the controversy, such as the potential illegitimacy of the legality of the band councils considering that they ultimately were an imposition on the Indigenous peoples by colonial authorities, the long history of wrongdoings by Canadian governments against Indigenous peoples in the name of resource development, and environmental concerns surrounding the continued investment in fossil fuels.

Both perspectives have their respective merits, but the dynamics of the modern digital information economy dictate that many individuals of opposing political orientations will not be exposed to anything more than fragmentary excerpts of the other side’s arguments and supporting facts therein, preventing them from being able to meaningfully engage with the points raised by the opposition. Thus, as a consequence of a difference in informational inputs, two different social realities are constructed, and the opportunity for productive public discourse and achievement of a potential negotiated compromise diminish ever further. Keen observers will notice that this case is not merely a hypothetical, but is based on a real event that occurred in February of 2020 when hereditary chiefs of the Wet’Suwet’En nation organized weeks-long railway blockades in opposition to the Coastal GasLink pipeline project. Though this event was largely overshadowed by the Covid-19 pandemic which would lead to a state of emergency being declared in March 2020, its significance cannot be understated. The Coastal Gaslink project was and remains one of the largest private infrastructure projects in Canadian history, costing well over 14 billion CAD. Additionally, at the time the blockading of critical infrastructure by Indigenous and environmental rights groups was a political norms violation largely unprecedented in scale in Canadian history, with the blockades stranding an estimated 425 million dollars worth of commerce for each day of the shutdown. Dozens of protestors were arrested during dispersal efforts by the RCMP, where many alleged several acts of police brutality took place.

Despite the significance of the event, even 3 years later it is immensely difficult to find articles that comprehensively cover all relevant facets of the situation, with most selectively covering the facts that most support their orientation and omitting those that are contradictory, especially those released at the time of the events. Accordingly, in line with the IDHPP the Canadian population was evidenced to be deeply divided in their conceptions of the situation at the time. In an Angus Reid poll of 1,500 Canadians in February 2020 56% of those polled believed the most important issues at stake were those of the economic impact of the protests as well as law and order, while 44% believed that it was issues of Indigenous rights and the environment. If a population cannot even agree over what issues must be settled, then it is difficult to see how a settlement could have been achieved between these differing viewpoints. Further polls revealed that this division occurred predominantly over party lines, with 89% of Conservative leaning respondents endorsing economic concerns and law and order as the primary issues whereas 76% of NDP leaning respondents endorsed Indigenous rights and the environment as being the primary issues. Supporters of the centrist party, the Liberals, were accordingly split 55–45 in favour of endorsing Indigenous rights and the environment as the primary issues.

This represents just one of many issues that broiled across the canadian political landscape before abruptly being swept out of the news cycle, and thereby public consciousness, by the next big thing, leaving in its wake a populace with opposing conceptions of what had occurred, and further differing conceptions of social reality as a result. Within such an informational landscape, access to a technology that could synthesize all available information on a given news story is critical to fostering a population that is sufficiently informed to resolve conflicts via democratic means of open debate and discussion. Confirmation bias cannot be eliminated, it is an adaptive feature of the human mind, we necessarily attend to information we believe to be correct more so than to information we believe to be incorrect, but it can be mitigated. This technology stands the best chance of doing so, as it does not require the individual to actively seek out the opinions of those who they may vehemently disagree with, and can present the totality of the available information on a subject more impartially than could be reasonably expected by any individual human being.

In time, such a technology could encompass all social media as well, scanning the words not just of our journalists, but of all of us, and using the sum total of our exposited beliefs to formulate guides that would allow anyone to gain an understanding of even the most complex of issues. Questions could be asked, and answered by such a system instantaneously. Those that disagree with each other may finally have a chance at understanding why they disagree, and an understanding of what it is that must be resolved between themselves for agreement to occur. For a democracy in the digital age and beyond, facing issues requiring ever-more sophisticated understandings of technology, science, and morality, there is no other way. We know the democratic citizen of the future will not be informing him or herself on the matters of state by opening up the morning newspaper. It is time we realize that neither will they be doing so by scrolling listlessly through a social media timeline designed to keep them engaged, not informed.