Expert urges voters to question and verify AI during election campaign

Sweden’s September election will be the country’s first in which artificial intelligence plays a role in political debate, SVT reports. A survey by the Internet Foundation found that one in 10 Swedes discusses politics with language models.

Måns Johansson, an internet expert at the Internet Foundation, said voters should use AI tools carefully and ask specific political questions.

“If you want to talk politics with AI, it is good to ask concrete political questions,” Johansson said.

AI technology has created new opportunities for political communication, but also for disinformation. Almost anyone can now produce a relatively convincing deepfake, meaning fabricated video or audio.

Johansson said large amounts of AI-generated material were produced to smear politicians during Hungary’s election this year. Sweden has also seen early signs of a similar development, although at a slower pace. So far, there have been a few examples of generated images of well-known politicians showing facial expressions they did not actually make.

Sweden’s established political parties have so far shown relative restraint in their use of generative AI, Johansson said. However, he suggested that some parties could loosen their restrictions towards the end of the campaign if they became desperate.

AI content must be labelled

The European Union’s new AI regulation took effect on August 2 and requires AI-generated content on the internet to be labelled. Johansson said more than 80 percent of Swedish voters support clearly labelling political information generated by AI, according to the Internet Foundation’s survey Svenskarna och internet.

Using language models to discuss politics has become increasingly common. One in 10 people now uses AI as a political sounding board. It is more common among young people than older people, and men are more interested than women in discussing politics with AI.

Johansson advised voters to ask AI to explain political terms and proposals, then conduct their own research. He also said every piece of information from AI should be checked, just as information from an online search engine should be.

“Give me three examples for or against this proposal” is one prompt he recommended. He said AI is often good at providing that kind of explanation.