How artificial intelligence and social media can change the elections

Where did the investigation lead us:

Three scientists have found that reality and social media are two different worlds
Antonis Galanopoulos (Political Scientist): Social Media reflects society: An obvious lie!
Giorgos Kyritsis (Scientist in Machine Learning): The new reality shapes a perfect environment for manipulation by political parties
Petros Ioannidis (Political Analyst, “Aboutpeople”): “Moving inside faces and things of our own “race” on social media, we can not know what is really happening outside of it. We comprehend just a part of the reality, not all of it.”

The scientists also told us:

“Trying to create not just a consumer profile on social media, but a political-ideological profile, becomes dangerous.”
“Parties through artificial intelligence algorithms can effectively invade social media to draw voters.”
“Models and old statistical methods for predicting party percentages can no longer be used in modern society.”
“Because of algorithms, we’re probably thinking less, so we’re being manipulated more.”

By Vassilis Galoupis and Panos Katsachnias

Morpheus gave the “Chosen One” the option to choose the “Blue Pill” or the “Red Pill”. If he got the first one, Neo would forget everything, go back to The Matrix, and go on with his life as if nothing had ever happened. But if he chose the latter, then he would be awakened. But with the price of awakening being high, since among other things, if he chose the “Red Pill”, Neo could never return to his previous state if he changed his mind.

But unlike the story told in the 1999 film The Matrix, the Greek reality of 2023 does not seem to have a happy ending. On the occasion of the election result of May 21, the existence of two or more electorates was revealed. Parallel worlds, that is, that as such never come into contact.

The “bubbles” of social media and their political microcosm are precisely those ecosystems that make users feel comfortable with a narrative or with something they believe, considering that this is a majority current and out in society.

Data Journalists reached out to AI and social media experts to help us understand what exactly happened in the recent parliamentary elections, with results that surprised even the most experienced pollsters, as well as the parties.

The reality is hidden behind social media

One would expect that party staff would be able to read the real picture in time, beyond those presented by social media. For example, many young people have turned to small parties. Or that SYRIZA’s staff never realized that it was losing a third of its voters, proof that it had no real “grounding” with society. What did really happen in the May 21 elections?

“On social media, the most “angry” people make noise,” observes Political Analyst Petros Ioannidis, referring to a fictitious image that was created. “Society is ultimately insecure, not angry,” he concludes.

They are born in a digital world to grow up in the real one

The reality, as a complex mosaic of many aspects, appears only when we go out into our material social world. While digital “bubbles” are maintained as such only on social media.

“Social media has a tendency to group people together and influence public opinion. They can easily do it. I think that the models and the old statistical methods for predicting party percentages can no longer be used in modern society,” says George Kyritsis, an expert in artificial intelligence.

“On social media, people only come into contact with content that confirms their thoughts. People tend to do what the person they like does. That’s why they can be manipulated easily.”

“Ekklesia of Demos” S.A.

Operating in the digital “bubbles”, being isolated and in a friendly environment for their views, also forgetting that like is not a vote, users seem to confuse the digital environment with that of the ancient Ekklesia of Demos” (Assembly of the People) or the “Agora”. They forget that they actually interact on digital platforms owned by private giant companies. They make decisions about who they feature and who they don’t, like Donald Trump and Twitter, and what content they are or aren’t featuring. Posts that are being deleted and users that are blocked for a period of time because they violate the terms and rules of the company.

“The sequence of algorithms is powered by the choices of each user in this environment. From which page he will visit, how long he will stay on it, how he will react, positively or negatively, all these are elements that fuel the interaction with the system,” says Mr. Galanopoulos.

These companies both reinforce the closure of each user in his bubble and limit his right to speak, since this is not a public space for dialogue, but for private businesses. In other words, users use the space it offers them, but the ownership is held by these companies forever. At the same time, they reserve the right for some opinions to give them a bonus, while others to cut them off.

Fake News

In defense of social media, there are at least two arguments: that social media is not responsible for users’ messages and that fake news spreads six times faster on the internet than real news. It is this latter “quality” that groups of views such as that of the “flat earth” exploit in order to grow their audience worldwide and by extension their revenues, since they receive money from the views of the videos and podcasts they upload.

The problem is that fake news can steer users in one direction or another because they are tailored to look like the content that the user likes. If, for example, the user is placed on one side or the other of the political spectrum, then he will believe any news that confirms the correctness of his position. He will consider it objective and not only will he not try to cross-check it, but precisely because he finds it agreeable, he will share it, thus spreading it even more. And here there is no distinction between Left and Right, since they both fall for it equally.

We are all journalists

The so-called “cross-checking of the news” is not done by the entity itself before publication, but it is required to be done by the user himself. This is not done, since not every user can turn into an investigative journalist who adheres to ethical principles before each publication. All of this acts as a feedback loop, or what is commonly called a loop, in which everyone is involved. The platform, the creator of the content, its consumer, and advertisers. And everyone, except the user who is just wasting his time, earns money. “If you don’t pay for the product, then you are the product yourself.” So in this case, the user, under the cover of being entertained, offers his time and data, being at the same time an advertising target, while with all his participation he raises the value of the shares of the companies involved with social media.

Games that are not… a game

Someone who stares at a page with guns, because they play a strategy game on the internet, does not realize that at the moment they are giving critical elements of their psyche or their wider ideological identity to third parties, filling the vast reservoirs of big data and without knowing how they will be used through artificial intelligence and by whom.

If, for example, a user–player is interested in his searches for immigration, the fence in Evros, weapon systems, and Erdogan, then he will “meet” with the corresponding messages emitted by election campaigns of parties with this or a similar agenda. He will be bombarded with matters of these categories in order to be directed toward the specific option.

According to what we know so far, the Russian secret services in 2016 did not need to hack anyone in the US or Britain to achieve their goals. It was enough to guide the audiences they wanted, where they wanted, by highlighting the appropriate, each time for the purpose they wanted to achieve, agenda.

Data Journalists, on the occasion of the recent election result and in anticipation of the “2nd Round” on June 25, spoke with three experts, in order with their knowledge and experience to help in the correct understanding and drawing of safe conclusions.

Antonis Galanopoulos (Political Scientist): Social Media reflects society: An obvious lie!

Antonis Galanopoulos is a Political Scientist and Ph.D. candidate in Political Science at the Aristotle University of Thessaloniki. He explains why the polls were once again disproved and how social media had an impact on the last election.

Mr. Galanopoulos, why was the election result impressive?

“The main thing is the big difference between ND and SYRIZA that takes us back to the first election result of the transition between the first (ND) and the second party (PASOK). With the result of May 21, this is the second time this has happened.

Equally important is that no one expected such a big difference between the first-government party (ND) and the second (SYRIZ), with the measurements that recorded it being questioned as exaggerated in their estimates, to come to the final result to be two and three times larger than them. In other words, there was a huge distance between how one experienced the social and political reality, with that which was ultimately reflected in the election result.

Why were the polls denied, even the Exit Poll?

The failure lies in the overestimation of SYRIZA, the first time since 2012 that it was underestimated because the unpredictable sample of answers in the polls was wrongly weighted. In other words, they made the choice – considering that they underestimated SYRIZA – to weight the sample based on the percentage it had taken in July 2019 and not on what it had taken in the European elections of March 2019. And so we had this divergence. In other words, they made the mistake of taking as a basis the approximately 32% SYRIZA received in July and not the approximately 24% it received in March. And on the basis of this, they counted both the rallying and the indecisive ones.

And with Exit Poll this happened because they continued to work in the same way. They made a mistake in the deliberation, believing that the difference they found was unreal (statements by Dimitris Mavrou of the MRB the day after the elections) and that if they published it they would have a problem. And by that logic, others have continued to make mistakes, or others have continued to tamper with data when they shouldn’t have. ”

Is politics still (as a proposal, content, etc.) the core of the parties, or does communication play a decisive role?

“Communication is part of politics. How one communicates one’s position is crucial to one’s policy. If you do it in a professional way, for example, it may indicate that you are reaching out to the experts every time with an excellent presidential campaign. You can do it like the KKE says that we don’t use communicators. This communication may succeed elsewhere as with the General Secretary Dimitris Koutsoumpas, but elsewhere not. These are choices that are not irrelevant to politics. The fact that communication plays an important role is a given. We live in a time when you can’t avoid it. Because it doesn’t just run on TV channels, as it was in the past. For a long time now, we have had social media which, as has been proven by research, play a key role. Since it seems to be the main source of information for young people under 35. Not necessarily as a means, but as a place to find other means. If a party does not understand this, it will have a political problem, not a communication problem.”

Strategies, policies, and communication failures that explain such a result?

“Style, for example, is a dimension of communication. Opposition SYRIZA leaders have often been accused of arrogance. In other words, the way in which they appeared in the public sphere was at least problematic, revealing arrogance. Even if you think you are right, there are many ways to communicate it. With each of them reflecting something. Surely arrogance is not the best way to communicate any position, since it puts the receiver at a distance. The essence of communication policy is to make the person who is watching or reading you feel close to you. Reduce the distance.

Alexis Tsipras, in one of his interviews on lifestyle shows, said that when you try to impersonate something you are not, you become “cringe”. But in his videos on TikTok, he did not manage to avoid it and not do it, since SYRIZA did not seem well read even if it wanted to address the youngest audiences of this media. Amateur result with poor image and video quality, with a vocabulary of past decades and funny-level Greek films and puns that are not of the time anymore. Especially when the audience of TikTok is 17 to 24 years old. While admittedly, Kyriakos Mitsotakis ran a very good campaign on TikTok. It was direct, relaxed, and interactive, without ever hiding that it was highly professional.

Another example is that of Aphrodite Latinopoulou, who with professionally made videos has thousands of followers on TikTok, numbers that correspond to those of a political leader”.

Are social networks able to influence the outcome of the elections?

“For a long time and to a very large extent. This is because they are critically placed. We rarely turn on the TV to get informed anymore. But our mobile phone that has all the applications follows us everywhere and at some point or moments of the day we will use social media from it. Apps have been built to be personalized as long as we use them and this is what creates the problem with social media.

“The more we use these applications, this personalization reaches a very large extent, resembling more and more ourselves, resulting in a “bubble” being built around us, in which the user is eventually locked away losing contact with the real environment,” points out the Political Scientist, Antonis Galanopoulos. By constantly coming into contact with opinions that agree with his own, the user comes to the point of thinking that they are the majority within the social reality, but we do not come into much contact. That’s because it feels good.

Just as companies build the ads they will see on social media according to their profile, users have built the political and social values that our surroundings believe in on social media. And when that sometimes breaks – as with the recent election result – some are shocked.”

Is this virtual reality being built by the users, or by the AI of the companies that run the programs?

“God helps those who help themselves” applies, since to a great extent the virtual environment is built by the users themselves. There is a reasonable intolerance to opposing views. The majority deletes users whose comments do not agree with theirs. Because they don’t think they have to spend the time they spend browsing social media or reading things they don’t agree with. This means, however, that they are building their environment and that they slowly think that the views they do not agree with do not exist.

The sequence of algorithms is powered by the choices of each user in this environment. From which page he will visit, how long he will stay on it, and how he will react, positively or negatively, all these are elements that fuel the interaction with the system. That is, each user tells it what he likes and it gives him what he likes, in order to keep him longer in it, gaining, on the one hand, more personal information (marketable) about him and on the other hand by showing him more ads. And if in its test suggestions to the user, the system is wrong, because they will not be confirmed by him, then it will get them back and will not show them again.

The attempt, therefore, to create not only a consumer profile but a political-ideological profile, becomes dangerous. All this becomes particularly dangerous, and that is when we talk about freedoms and rights.

We saw it in 2016 with Brexit, the US election, and the election of Donald Trump, accusations of Russian meddling, targeting specific messages at specific categories. Not to those who were already convinced and voted for a party, or were already converted to an ideology or a cause, but to social groups that could be attracted. Like people who felt insecure, or loved guns and were not recorded in the polls. To them they do not aim by displaying party advertisements, but political messages that will indirectly lead them to where they want. This makes it dangerous for Democracy to have algorithms that turn users into guided targets.”

Giorgos Kyritsis: The new reality creates a perfect environment for manipulation by political parties

George Kyritsis is a scientist in “Machine Learning Engineer”. He was born in Cholargos in 1987 and graduated from the School of Civil Engineering. He then decided to expand his knowledge with a master’s degree at the Polytechnic School of Lausanne (EPFL) in Computer Science, specializing in Data Science and Financial Engineering.

Machine learning is the development of algorithms and models that allow computers and systems to learn from data without having to be rigorously programmed. The idea is that systems can autonomously develop the ability to recognize patterns, make predictions, and make decisions based on the data provided.

Among other things, he says, speaking to Data Journalists on the occasion of the recent parliamentary elections and the surprising results:

“Mr. Kyritsis, could an artificial intelligence expert like you see the trend that was taking place in the Greek elections of May 21 and where the results surprised pollsters and parties? What were the “signs”? Do you think that lockdowns and fear of the pandemic affected?

“Of course, an artificial intelligence scientist can spot these new trends. Machine learning and data analysis models can be powerful tools for predicting and detecting trends in political elections. However, when a model shows significant deviations from previous results, as in the case of the 21 May elections, it is natural to raise doubts and reservations. It may be “accused” of trying to manipulate citizens.

When you say model, is it different from that of pollsters? Is there anything else? Can you explain it to us? What exactly are you looking at?

“In my work, what’s really going on is training a model that tries to find patterns in past data so that it can then use that to predict the future. So as you can see, the predictive capability of the model is based on the availability and quality of the data used. In addition, we make sure that the data we use is varied as the variety and quantity of data affect the predictive ability of the model. The more complete and representative the information available, the more accurate the predictions for the future can be.”

Do databases come from social media, can you give us an example?

“Yes, they can come from there. Social media is a rich source of information. Millions of people publish and exchange information, opinions, and feelings through them. Using data mining and social network analysis techniques, it is possible to collect data from various social media platforms, such as Facebook, Twitter, Instagram, and others. This data may include texts, comments, photos, videos, and other content posted by users.

However, the use of data by social media raises issues of privacy and ethics. Researchers must comply with data protection regulations and ensure the anonymity and security of the data they use. Facebook doesn’t give you data from private accounts, it only gives you data from public accounts.”

How did you see the trend taking shape? Did the world show through social media or by your own data that is moving in one direction or another? And when was this trend formed? During the lockdowns? Later?

“Trend shaping and the impact of social media is a complex phenomenon and is influenced by many factors. People in general are also indecisive. They decide at the last minute. It is very difficult for the models to catch this trend, basically, you have to prepare them appropriately to be able to catch these trends. Especially for young people who decide at the last minute. The lockdowns, I don’t know if they played a role in the outcome.”

How much does technological development affect elections so far and how much do you think it will affect with the rise of artificial intelligence? What can we expect compared to today?

“Technological developments have had a significant impact on the way elections are conducted. Now, through the internet, political parties can get in touch with a large number of people. Parties through artificial intelligence algorithms can effectively invade social media to draw voters. Polling, the election results in the future may be much more unpredictable than they were in the past.

Social media has a tendency to group people together and influence public opinion. They can easily do it. Models and old statistical methods for predicting party percentages can no longer be used in modern society. Perhaps the old statistical methods should be enriched with tools that are capable of analyzing and processing big data”.

An increasing number of people, especially young people, are consuming news only on social media. Essentially, what they consume as information is what their friends “upload”. A microcosm. What kind of voters does this reality create? And how can a party that wants to promote its own positions “invade” this microcosm?

“That’s true. Social media to increase your interaction and stay on the platform tries to serve you content that you agree with or that your friends agree with. That is, people only come into contact with the content that confirms their thoughts. People tend to do what the person they like does. That’s why they can be manipulated easily.

This reality can create voters who are prone to affirming their views and find it difficult to take into account information that contradicts their beliefs. This reality creates a perfect environment for manipulation by political parties. For example, there are artificial intelligence tools that can process a large volume of text in order to condense the information into bullet points. Bullet points will be the dominant themes and opinions of users in this microcosm. So a political party could invade this microcosm with content that is “close” to these bullet points”.

Practically, are they manipulated by the algorithm of e.g. Facebook and Twitter, or by themselves, i.e. by what they want to see?

“The algorithm is not interested in manipulation, it just has an objective. The algorithm wants you to stay on the platform as long as possible. So, the algorithm, trying to maximize the time you spend on the platform, tries to serve you articles and posts based on your friends’ likes. This creates the grouping. People tend to see what they agree with on the internet. This further strengthens his view. He thinks that what he believes is what happens in society”.

Can an expert see the trends shaped by all these microcosms? Whether the trend has greater penetration in society, or not. Is there any way to know what’s really going on? To see the big picture of all these subsets?

“Yes, it does exist. For anyone managing big data, of course. There are data analysis tools that can identify trends and interactions between users. You need a lot of data, though. That’s why I mentioned the term big data, to extract an objective view of the dynamics of all trends.”

Is it easier for an AI researcher to discern trends today than for a classical pollster?

“It is difficult to answer that. Maybe it is. It is enough to have access to big data and quality data.”

Algorithms determine what you see. And seeing what you step on creates a universe of information tailored to each of us. Our own “bubble”. What are the risks, both socially and politically?

“There can be a dominance of one or two opinions and there can be no pluralism. We’re supposed to be in the information overload phase, though. As you said, the algorithms adapt the content based on the interests, preferences, and behavior of each user. People tend to prefer and give more weight to information that confirms what they believe while ignoring information that challenges their prejudices. So each of us is in a very customized/personalized environment.

Politically, this can amplify the polarization and diffusion of fake news, as users are only exposed to content that fits their biases. Socially, this phenomenon can lead to a strengthening of existing inequalities. People may have limited exposure to information and opinions that differ from their own. This can lead to the strengthening of prejudices and isolation from different social groups.

Often, the algorithm makes it easier. Like streaming platforms that successfully recommend songs or movies you like. But on the other hand, you’re probably thinking less, so you’re being manipulated more.”

People are trained in a “see title, scroll” pattern. Although these media outlets earn their money by drawing your attention, they actually train you to have less attention, for example, to read a story, to distinguish what a valid report is from fake news. Is the world becoming more or less manipulated?

“More manipulated. That’s why educating the public on information reliability and recognition of fake news is an important issue. Indeed, this way of operating social media, such as viewing attention-grabbing headlines and the constant flow of information, can lead to a decrease in attention and a decrease in critical ability.”

Can a limit be placed on fake news or photos produced by artificial intelligence?

“Yes, it is possible to put limits on fake news and photos produced by artificial intelligence. Efforts are being made to develop algorithms that can detect disinformation and the communication of false information. In the future, people may be more adept at identifying fake news. With regard to photos produced by artificial intelligence, image analysis, and verification techniques can be applied to identify possible misleading or manipulating or forged images.”

Petros Ioannidis: Earthquake in two halves

“Moving within persons and things of our own ‘race’, we cannot know – at least precisely – what is really happening outside,” explains Petros Ioannidis, Political Analyst and Director of the company “Aboutpeople”:

Mr. Ioannidis, how do you evaluate the election result?

“In any case, we are talking about a big victory for ND, since the 41% rate is higher than what it got in 2019. Something that’s never happened before. That is, after a 4-year government term, to win the elections again with a higher percentage than the previous time.

A crushing defeat for SYRIZA, which lost about 11.5 points from what it had gotten in 2019.

The increase recorded by PASOK, almost 50% of what it had received in 2019, ranks the party of Mr. Androulakis among the winners of the race. However, I believe that, depending on the loss rate of SYRIZA, this increase could be even greater.

For the three parties that have approached (NIKI, COURSE OF FREEDOM, MeRA25) but have not managed to exceed 3% in order to enter Parliament, we will have to wait for the result of the elections on June 25 to finally see what will happen. Usually, parties that do not enter the Parliament the first time it is difficult to do so the second time. However, because the scene is still not clear since the result of the previous ballot box caused an “earthquake”, it is still unknown what will eventually happen.

There are three options for the upcoming elections:

Whether ND will maintain its very high ratings
What will happen in the field of the center-left: PASOK will approach SYRIZA, and SYRIZA will increase or decrease its percentage, that is, what will be the correlation of forces in the center-left after the next elections
What will happen to the three parties that are struggling to enter parliament.”

Why did the polls fall out once again in their prediction?

“Polls are not predictive tools. The predictive tool is ExitPoll. Polls show trends at the moment. Meaning the views of citizens that may change along the way. That’s what it looks like now, with about a fifth of the public deciding in the last week. I would say that the polls caught the general trend, but not the huge downturn of SYRIZA.”

But even the Exit Poll, which is a predictive tool, did not predict the outcome.

“The Exit Poll that comes out as soon as the polls close (19:00) is at a rate of 80% since it does not include those who voted in the afternoon. An hour later (20:00) the final result comes out. In this, predictions fell out only in terms of the extent of SYRIZA’s downfall”.

Did the “silent” election period contribute to this?

“Silent” elections were a sure thing. There was also no public expectation of anything in this election. In 2015 there were voters – mainly SYRIZA – who hoped for a change. In 2019, there were voters – mainly from ND’s side – who also hoped. In this election there was no tendency to change, we did not see enthusiasm and hope in the whole electorate.”

Were the near majority of ND (41%) and the percentage of SYRIZA (20%) known?

“Except for the period of the Tempi railway accident, all the polls consistently reflected – certainly not for 20% – the lead of ND. In Poll of Polls, the percentage difference was around 7-8%, so it caught the trend.”

What factors determined the outcome of the elections?

“Four main ones:

ND as a whole (and Kyriakos Mitsotakis and his cadres) expressed a clear and coherent message ahead of the elections.
SYRIZA in its election campaign was constantly changing its own message, confusing the citizens.
Throughout the four years, SYRIZA has had an opposition that seemed angry, misreading the society that was in fact anxious and insecure. So even its oppositional tactics did not catch the pulse of society.
The electoral system of proportional representation. Where ND voters, listening to Kyriakos Mitsotakis who spoke of a majority government, voted with the logic of enhanced proportional representation, while voters of the other parties voted with the logic of proportional representation and governmental cooperation. That is why we had a large percentage (16%) of the parties outside Parliament when in 2019 it was 8%.

Has the system of proportional representation been defeated?

“The system of proportional representation is under certain conditions the most democratic system. So the system of proportional representation was not defeated, the way in which the parties managed the system of proportional representation was defeated.

Is Greek society still regarded as progressive with an ‘anti-right’ sign?

“For decades, if we compared the percentages of right parties with the ‘anti-right’ – even when ND was a government – the latter prevailed. After the elections of May 21, this seems to be changing, but let us wait for the next elections on June 25 to have the final picture.

How the system will end up in the center-left with the percentages that SYRIZA and PASOK will receive, will determine the central political scene afterward, since there is the possibility of having only one major political party and two more middle ones to follow. In other words, we will no longer have the classic form of bipartisanism in Greece. “

Were there parallel electorates in this election due to social media?

“The social media in which each of us moves represents a reality (virtual or not) but certainly not the real situation as a whole. It’s a bubble in which we chat with users and read things that we mostly agree with. So by moving through the faces and things of our own “race,” we cannot know – at least precisely – what is really going on outside of it. We comprehend just a part of the reality, not all of it.”

Can a fictitious image produce measurable results?

“On social media, the most angry people are making noise. Therefore, I do not exclude that SYRIZA was led astray by this and thought that these express the whole of society. Perhaps this fictitious image influenced SYRIZA not to see the real one and to think that society is “angry” while it is insecure. Thus leading to an “angry” opposition and aiming wrongly. This was shown by SYRIZA’s political advertisements, which were wrong because they were “black” and did not bring out optimism and perspective.

Does all this indicate a lag behind politics in relation to communication?

SYRIZA submitted a program but did not focus so much on the discussion about the future but on the harsh criticism of the Mitsotakis government. This was both a communication mistake – because people vote mostly for the future – and a political miscalculation.”

On the basis of the first result, can the outcome of the next election be predicted?

As they say Mitsotakis, Tsipras, and Androulakis “the ballot box is empty” ahead of the coming elections. At the moment it is still early days, no predictions can be made about the exact result, but it is clear that ND has a huge lead.”

Can we now talk about regularity after the “cycle” of the 2012 and 2023 election results?

“No. The normality would be that the new bipartisanship (ND-SYRIZA) was at high rates of 30 and 35%, each since the old one reached a cumulative 80-85%. Based on the election result of May 21, we cannot talk about a return to the past after the new reclassification that was recorded. Overall, we will be able to discuss the new image that is being formed after June 25 and the election results at that time. In any case, in the midst of two elections, it is dangerous to talk about a return to any normality, let’s be patient for a few weeks to get the full picture.”

What do you expect to be the turnout in the next elections that take place this summer?

Elections at the end of June and an election that does not appear to have a strong stake based on the outcome of May 21. Participation should be smaller. We are waiting…”