Ahead of every election campaign, dozens of polls are published, typically based on just a few hundred respondents. Polling institutes tell us they have carefully selected a sample representing all eligible voters, taking into account variables such as age, gender, area of residence, education, and level of religiosity. Yet time and again, these polls struggle to accurately reflect the actual results.
In the era of artificial intelligence and data science, the way we measure public opinion may also need to change. Instead of asking a few hundred people how they plan to vote, algorithms can analyze the digital behavior of millions of people to uncover patterns that humans would find extremely difficult to identify.
What our digital footprint reveals
Every digital action we perform leaves behind a data footprint. The likes we give, the videos we watch, the searches we conduct, the purchases we make, and the pages we follow are each almost meaningless on their own. However, when billions of these data points are connected over months and years, a surprising picture emerges of who we are: Our interests, our habits, and even our core values and political views.
This question is no longer theoretical. In recent years, studies have shown that the information people leave behind on social networks can also indicate their political stances. One of the best known studies, conducted at Cambridge University in 2013, showed that the political affiliation of Facebook users could be estimated with approximately 85% accuracy based solely on an analysis of their likes.
The algorithm does not necessarily look for a like on a specific political party. Rather, it is the combination of music, sports teams, television shows, brands, and other interests that makes it possible to identify patterns characteristic of people with similar views. Later studies showed that our hobbies, the pages we follow, and even our activity in non-political forums can provide significant clues about our political tendencies.
This raises the truly interesting question: Can this knowledge be used to measure public opinion more accurately?
Can AI replace traditional polls?
Election polls are based on a simple premise: If we ask a representative sample of the population, we can estimate how the entire country will vote. In practice, this method has well-known limitations. A poll is conducted at a single point in time and relies on answers provided by respondents. Some refuse to participate, others hide their true views, and some answer with what seems socially acceptable or desirable. The wording of questions, their order, and sometimes even the interaction with the pollster can also influence the outcome.
By contrast, methods based on data science do not rely on an answer given during a brief phone call. They analyze behavioral patterns accumulated over time across much larger populations, focusing on what people actually do, rather than just what they choose to say at a given moment. This approach also has limitations and is not immune to bias or error, but it opens a new possibility for measuring public opinion on a scale that was previously impossible.
Artificial intelligence is already changing how we work, learn, search for information, and make decisions. Entire industries are undergoing transformation, making it natural for the world of election polling to evolve accordingly.
The question is no longer whether technology is capable of learning about our political preferences, but whether we will continue to measure public opinion through a few hundred telephone responses or adopt tools capable of learning from millions of behavioral patterns in the digital world. At the same time, we will need to ensure that these same tools are not used merely to measure public opinion, but also to shape it.
The author is the head of the Data Science Program in the Faculty of Computer Science at the College of Management Academic Studies.