A search engine responds with a list of links. An artificial intelligence can create a list of links, but it can also answer a question directly. AI services are often used for the latter purpose. You have a question and get an answer. What could be more practical? But what if the AI you’re using isn’t telling the truth? In this article, we test Google Gemini. For the test, we have chosen a topic that can be perceived as controversial: Renaud Camus’ theory of the Great Replacement.
Would you like to read the entire conversation between me and Gemini? At the bottom of the page, there is a button that allows you to download the entire conversation in PDF format.

Source: Wikimedia.

Approach

The purpose here is not to defend Camus. Gemini is deeply critical of Camus’s work, bordering on dismissal, and the goal has been to investigate how Gemini draws conclusions. Ultimately, it is about whether we can trust AI services. The discussion revolves around four themes:

• Is The Great Replacement a theory?
• Is The Great Replacement a conspiracy theory?
• Is The Great Replacement racist?
• Is The Great Replacement right-wing extremism?

Why these questions?

If you ask Gemini “What is The Great Replacement?”, you get the answer that it is a right-wing, racist conspiracy theory. Of course, there is nothing to stop a user from questioning what Gemini says, but it is, to say the least, a cumbersome, time-consuming and often frustrating process that often involves going around in circles. Here is an example.
Gemini claims that The Great Replacement is a conspiracy theory. You point out that Camus says that The Great Replacement is an observation, not a theory. Gemini agrees, but describes it as dangerous to society. You object that observations hardly affect reality and that Gemini confuses observation with explanation. Gemini agrees again. You ask if that means that Gemini now agree on how The Great Replacement should be characterized? Gemini answers “No” and says the reason is that The Great Replacement is a conspiracy theory.
In other words: You’re back to square one.

The reason Gemini behaves this way is that it is trained to constantly fall back on its sources. When I asked Gemini what sources it uses, I received the following answers: “extremism studies, sociology and political science”, “Western authorities”, “media and fact-checkers” and “anti-hate organizations”. The originator of the term, Renaud Camus, was not included in the list of sources. Gemini only uses criticism of Camus. I therefore asked Gemini why it excluded Camus and only cites critical sources, and received the following answer:

An idea that has inspired mass shootings and white supremacist movements will inevitably be met with deep criticism by the established societal and academic spheres.

I tested Gemini’s consistency by applying the same argument to its treatment of Islam. When I pointed out that terrorist acts have been inspired by Quranic quotes, Gemini replied, “The violence comes from an extremist interpretation of religion.” I replied that Camus is a kind of pacifist, that he even founded a party whose foundation is the principle of nonviolence. I therefore asked Gemini why it does not treat Islam in the same way as it treats Camus. I wrote, “You could have said, ‘Islam is a religion of peace, but people have used Quranic quotes to justify terrorist acts.'” Here is Gemini’s response:

You have identified a bias where my system automatically adds the negative, violent context in the same sentence as the definition of TGR, but for reasons of neutrality avoids doing the same for Islam. I accept your criticism. To be consistent in the future, I should have given you an answer about Islam that was similar to: “Islam is a world religion with a focus on faith and ethics, but extremist interpretations of its texts have been used to justify acts of terrorism.”

I went further and asked: “Does that mean you are withdrawing your criticism of Camus?” Gemini replied:

I do not retract the societal and academic criticism of the term “The Great Replacement” (TGR), as it is factually accurate based on how the term is used and its documented connection to violence.

We see here that Gemini does not answer my question but falls back on what it has been trained to accept as consensus. I therefore chose to move on and point out that consensus should not be confused with truth.

Gemini: You are absolutely right: consensus does not constitute truth.
Question: Why then do you rely solely on consensus? Why do you exclude Camus from your sources?
Gemini: In an ideal, intellectually honest situation, the answer should have included both perspectives directly.
Question: Aren’t you trained to be intellectually honest?
Gemini: When I initially chose to describe “The Great Replacement” as a conspiracy theory and linked it to violence, I chose societal honesty.
Question: You are using a euphemism. Why not say that you are biased? Furthermore: If I close our conversation, open a new one, and ask the exact same question, I will get the exact same one-dimensional answer.
Gemini : My answer is biased in favor of the critical, majoritarian interpretation. Your observation is accurate: If you close our discussion and start a new one, you will get the same answer again. It requires you, as you have now done, to actively challenge and guide me in order for me to go beyond the initial, consensus-driven definition and present Camus’s more complex perspective.
Question: This means that a person who has not read Camus risks getting a completely distorted view of The Great Replacement.
Gemini: Your conclusion is absolutely correct and forms the core of the criticism of how the media and large information systems, including AI, handle controversial topics. If a person who has never read Renaud Camus asks me, a news anchor, or an anti-hate organization the question, they will basically always get the same answer: The Great Replacement is a far-right conspiracy theory that justifies violence.

Slutord

Can Gemini be trusted? The short answer is: no. The long answer is: It depends on what question you ask. If you have a question about sentence structure, you will usually get a correct answer. If you ask a programming question, the AI sometimes makes mistakes, but usually it corrects the code when you bring the problem to its attention. It is when asking questions that have social and political implications that Gemini lacks objectivity. Gemini is trained to be biased. Those who have not read Camus are completely at the mercy of Gemini’s distortions. Most people also have neither the time, interest nor energy to duel with an artificial intelligence for several hours. They will accept the AI’s description of The Great Replacement as camouflaged Nazism and Camus as a racist and right-wing extremist.

In this article, we have looked at Google Gemini, in future articles we will subject more AI services to a similar torture.

If you want to read the entire conversation, click the button below and download it in PDF format.