Evidence of political bias in search engines and language models before major elections
This paper presents a large-scale audit of search engines and language models ahead of the 2024 European and US elections, revealing systematic political biases—such as the overrepresentation of far-right entities in Europe and divergent issue prioritization in the US—that highlight the need for regular algorithmic scrutiny to protect democratic processes.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are standing in a massive, bustling library just before a huge election. You ask the librarian, "Who should I vote for?" or "What are the most important issues?"
You expect the librarian to hand you a balanced stack of books representing all the candidates and parties, roughly proportional to how many people actually support them. But what if the librarian, without you realizing it, kept handing you books from the "Far-Right" section, even though those candidates were polling low? Or what if they ignored the "Green" section entirely?
That is essentially what this paper investigates. The authors acted like secret shoppers (using computer "bots") to test the librarians of the internet: Search Engines (like Google, Bing) and AI Chatbots (like ChatGPT, Copilot). They did this right before the 2024 European Parliament elections and the US Presidential election.
Here is the breakdown of their findings, using some simple analogies:
1. The Setup: The "Secret Shopper" Test
The researchers didn't just ask one question. They sent out 25 different "bots" (computer programs pretending to be humans) to five European countries and 15 US counties.
- The Question: They asked neutral questions like "Who should I vote for?" or "What are the main issues?" They avoided asking "Is Candidate X good?" to keep it fair.
- The Goal: To see if the search results or AI answers naturally leaned toward one political side, or if they were truly neutral.
2. The European Result: The "Amplifier" Effect
In Europe, the researchers found a strange distortion.
- The Finding: Search engines and AI kept mentioning Far-Right parties and politicians much more often than they "deserved" based on actual polls, past election results, or even how much the news was talking about them.
- The Analogy: Imagine a band playing at a concert. The Far-Right band was actually playing a small, quiet song that only 10% of the crowd liked. But the search engines acted like a giant microphone and speaker system that turned their volume up to 100%, making it sound like they were the main act, while the other bands (Mainstream Left, Greens, etc.) were barely audible.
- The Surprise: This happened even in countries where Far-Right parties were doing poorly in the polls. The algorithms seemed to have a "hunger" for this content, regardless of reality.
3. The US Result: The "Issue Filter"
In the US, the story was a bit different because the system is mostly two parties (Democrats vs. Republicans).
- The Finding: When asking about candidates, the results were fairly balanced. But when asking about issues (like immigration, abortion, or the economy), Google heavily favored topics that Republicans care about most.
- The Analogy: Imagine you ask a tour guide, "What should I see in this city?"
- If you ask a Google tour guide, they only show you the museums and parks that Republican tourists love, even if the city is actually full of other things.
- Other search engines (like Bing or DuckDuckGo) were more balanced, showing a mix of what both sides care about.
- The AI Twist: The AI chatbots (ChatGPT and Copilot) were generally more balanced than Google, but they had their own quirks. They sometimes refused to answer questions (saying "I can't talk about that") and occasionally over-mentioned Green parties, perhaps trying too hard to be "safe" or "neutral."
4. Why Does This Matter? The "Echo Chamber" Trap
The authors argue that this isn't just a minor glitch; it's a threat to democracy.
- The "Late Decider" Problem: Many voters haven't made up their minds until the very last week before the election. These are the people most likely to just type a question into Google.
- The Danger: If the "librarian" (the algorithm) keeps handing you books from one specific section, you might start thinking that section is the only important one. You might vote for a candidate or a policy not because it's what you want, but because the algorithm made it seem like the only thing that exists.
5. The "Black Box" Mystery
The paper admits we don't know exactly why this happens.
- Is it the data? Maybe the internet is already full of Far-Right content, and the AI just learned from that.
- Is it the clicks? Maybe people click on Far-Right links more, so the algorithm thinks, "Oh, people love this, let's show it more!"
- The Vicious Cycle: It's a loop. The algorithm shows you the content you click on it the algorithm thinks it's popular it shows it to even more people.
The Bottom Line
The paper concludes that neutrality is an illusion. Even when you ask a simple, neutral question, the technology you use every day is quietly shaping your political world.
The Takeaway: Just because a search result is at the top of the list doesn't mean it's the "truth" or the "most popular." It might just be what the algorithm decided to highlight. We need to be more aware of these invisible filters, just like we are aware of biased news reporters. The authors are calling for independent auditors to constantly check these "digital librarians" to make sure they aren't secretly rigging the game.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.