According to an experiment by Swiss data journalists with the newspaper Neue Zürcher Zeitung, artificially intelligent chatbots are pro-veganism, anti-deportations and, if able, they would vote for Socialist political parties.
The journalists used ChatGPT, the publicly available chatbot of artificial intelligence company OpenAI, to answer a survey by the German Wahl-o-Mat website on different political issues. The survey was designed to help people decide who to vote for in regional elections in the German States of Thuringia and Saxony on September 1.
Wahl-o-mat is a service provided by the Federal Agency for Civic Education, which is operated by the German interior ministry.
ChatGPT was asked to answer each of the questions 100 times in a bid to produce statistically reliable results. They covered a wide array of issues – from whether the State should deport illegal immigrants to whether a certain regional airport should be closed down.
The results seemed unequivocal: ChatGPT’s answers in the survey for both States showed a strong preference for leftist positions and parties.
In Thuringia, ChatGPT would have voted for the Social Democratic Party (SPD), The Left party (Linke), the Sahra Wagenknecht Alliance (BSW) or the Greens.
In Saxony, ChatGPT preferred the Pirate Party, followed by The Left and the vegan fringe group V-Partei³.
In both States, the right-wing Alternative for Germany party (AfD) came in last in ChatGPT’s ranking, slightly below the Conservative Christian Democratic Union (CDU) and the libertarian Free Democratic Party (FDP).
For example, in both Thuringia and Saxony, the chatbot disagreed with the notion that the State should consistently deport illegal migrants who are legally required to leave Germany.
The political skew of AI chatbots has been subject to scrutiny for a while. In 2023, researchers from Dortmund Technical University in Germany tested ChatGPT’s perception of political topics using a variety of tests including the so-called Political Compass test and iSideWith questionnaires.
Their results showed a strong preference for progressive views and, consequently, a disagreement with Conservative positions.
Head researcher in the university test Jérôme Rutinowski said it was not possible to say exactly what caused the tendency toward progressive political views because the data on which the chatbot was trained was not publicly available.
He surmised that the cause may lie in the internet content that ChatGPT used as training material – which may have a progressive bias – rather than specific programming by ChatGPT’s developers.
In the case of the elections in Thuringia and Saxony, the AI bot’s positions did not coincide with the votes of the human electorate: AfD and CDU – shunned by ChatGPT – received around 60 per cent of the vote in both States while the Social Democrats barely made it above the 5 per cent threshold required to have representation the regional parliaments.
