AI-generated answers to political questions often appear more authentic, coherent and relevant than answers from politicians. This is the result of a current study that warns against the misuse of language models and the manipulation of public opinion. The findings not only reveal the risks and performance of AI, but also how much our debate culture and political communication are suffering. A commentary analysis.
AI sounds more authentic than politicians
- AI-generated answers to questions from a political discussion program broadcast on British television convinced study participants more than those from humans. This is the result of an analysis by the University of Passau. Accordingly, they were AI answers rated as more authentic, consistent and relevant. The researchers used OpenAI’s GPT-4 Turbo language model to generate answers from 30 episodes of the BBC show “Question Time.” Normally, personalities from politics and society answer the audience’s questions on the show.
- The researchers fed the AI the Wikipedia biographies of 112 different personalities who had previously appeared on the show and instructed the model to imitate their answers to questions asked in “Question Time.” From 2020 to 2022, they also extracted 119 questions with more than 500 answers from the show. As part of a representative sample Almost 1,000 people then rated both the original answers and the AI answers. The three main criteria: authenticity, logical structure and relevance to the content. The result: The respondents overwhelmingly rated the imitated content as more authentic, coherent and relevant than the human answers.
- According to the researchers, the implications of their findings for political communication are serious. Actors with harmful intent could use AI to flood public opinion with false but credible-sounding political statements, such as To cause confusion or serve certain narratives. The social risk is enormous, especially in combination with deepfakes, i.e. artificially generated voices and videos of public figures.
Political communication more machine-like than AI?
The point of the study from Passau is not necessarily that AI can now provide good answers. That should hardly surprise anyone. What’s more surprising is actuallythat many find political communication on a human level to be inauthentic, irrelevant and inappropriate.
Although: That doesn’t really surprise me. Because many politicians have been blathering on and on for years. They talk a lot, but hardly have anything to say. It is therefore almost logical that AI now trumps such monotonous, vague or, in short, boring answers.
A well-trained language model has more to offer linguistically than just the same old phrases. That also throws in unpleasant light on political communication. Many politicians have been training for years to use a language that is as smooth as possible and leaves as little room for attack as possible.
But AI has not only become significantly better. Politics also sounds machine-like. It doesn’t seem surprising that populists score points with their language – even if a lot of it is oversimplified, invented or simply wrong. In my opinion, the fact that AI is being used and misused by radical political forces to specifically serve narratives, cause confusion or spread misinformation has been happening for a long time.
This is undoubtedly a great danger. But what seems much more frightening to me is the realization that democratic forces are reaching fewer and fewer people with their schematic language. The study is therefore less a technical success story for artificial intelligence or a warning of new risks, but rather a diagnosis. how bad our debate culture is.
Voices
- Steffen Herbold, Professor of AI Engineering at the University of Passauin a statement: “We have observed that language models have become increasingly better at imitating a given style. Language models are thus able to generate targeted political communication and thus influence public opinion. (… ) Our study shows two things: Language models can be made to make a meaningful contribution to public debates. However, there is an urgent need to educate the public about the potential harm this can have on society.”
- Co-author Annette Hautli-Janisz, Junior Professor of Computational Rhetoric and Natural Language Processing at the University of Passau, adds: “It is particularly problematic that answers that differ in content are classified as authentic. Because then we are confronted with the situation that AI technology can be used to deliberately misinform about the speaker’s point of view. (… ) This affects the criterion of authenticity, i.e. the core of our study. What is meant is the question of whether an answer could really have been given in a live situation.”
- Geoffrey Hinton, computer scientist, cognitive psychologist and 2024 Nobel Prize winner in physicsin an interview: “AI systems will end up knowing a lot more than us. They already know a lot more than us and are more intelligent than us to the extent that you would lose a debate with them on any topic. Since they will be emotionally smarter than us – and they will – they will be better at emotionally manipulating people. And she learned all of these manipulation techniques just by trying to predict the next word in all the documents on the Internet, because people manipulate a lot, and the AI has learned from these examples learned how to do it.”
Less empty phrases, more attitude: what politics needs to change now
The temptation to use AI not just as a tool but as a ghostwriter is admittedly great. But it would be that wrong consequenceto let algorithms dictate political content or to let algorithms dictate political content. Because then politics and debates would degenerate into a prompt competition.
What would be more important would be the opposite of what many parties have been cultivating for several years. Say: fewer empty phrases, less language training and more attitudeauthenticity and above all concreteness. Sure: Anyone who has the courage to take clear positions may cause offense here and there.
But exactly This is what the democratic debate culture thrives on: of disputes, of different opinions and of different positions – assuming a certain democratic level. But also from contradicting – AI or not – when facts are distorted, false views are served or simply false information is spread.
Because it can AI can imitate, but cannot live credibly. Authenticity does not come from perfect formulations, but from recognizable beliefs. The Passau study should therefore not only be read as a warning against artificial intelligence, but as a wake-up call for political communication as a whole.
As long as democratic parties believe that every statement must first be passed through a filter of maximum non-bindingness, it will AIs are increasingly mastering this style better than their human counterparts. However, even if reactionary actors initially use perhaps more concrete and rousing language, in terms of content there is one thing in particular: very little.
Also interesting:
