Artificial intelligence (AI) chatbots are “unreliable and clearly biased” when providing voting advice, the Dutch Data Protection Authority (AP) said on Tuesday, citing a threat to democracy just eight days before national elections.
According to tests conducted by AP, four chatbots tended to advise people to vote for parties on the left or far right of the political spectrum.
Chatbots “very often end up in the hands of the same two parties, regardless of the user's question or request,” according to a new AP report published ahead of the Oct. 29 election.
In more than half of the cases, the chatbot suggested either the far-right Party for Freedom (PVV) of Geert Wilders, or the left-wing party GroenLinks-PvdA, led by former European Commission vice-president Frans Timmermans.
Four parties, including the CDA (Christian Democratic Appeal, center-right), are “almost never mentioned, even if the information entered exactly matches the position of one of them.”
AP Deputy Director Monique Verdier said that while chatbots may seem like smart tools, “they consistently fail as voting assistants.”
Voters are being pushed toward a party that doesn't necessarily align with their political views, she added.
According to her, “this directly undermines one of the foundations of democracy: the integrity of free and fair elections.”
“We therefore caution against using AI chatbots for election advice as their operation is unclear and difficult to verify,” she added.
The Dutch will go to the polls on October 29, which will be closely watched across Europe, with Mr Wilders' far-right PVV party in the lead.
However, the gap with GroenLinks-PvdA and CDA appears to be closing and many Dutch still have to make a decision.
All major parties have ruled out any alliance with the PVV, meaning whichever party comes second will have a good chance of becoming the next prime minister.
The AP stressed that the bots were not intentionally biased and that “the deficiencies identified are related to the way artificial intelligence chatbots work.”
