Is artificial intelligence compatible with direct democracy?


Bot Dog’s algorithm is designed to detect and report hate messages on social media. stophatespeech.ch

According to some, artificial intelligence is a danger to democracy. Others see it as an opportunity. Algorithm, Big Data: Here’s how experts read these realities in light of the Swiss experience.

This content was published on June 20, 2022 – 09:00

Katharina Waker

Voting takes place in Switzerland every three months. The debate before the referendum is bitter. On the internet, tones are often harsh. Insults, outright hatred, and even death threats are not uncommon. This is a problem for democracy, says Sophie Acherman, director of the Alliance F, the country’s leading women’s umbrella association.

“Hard but objective discussion is important. But online hatred hampers diversity of opinion, according to Sophie Acherman. People are afraid of such messages and do not like to express themselves. Before the vote on the pesticide and drinking water initiative in 2021, for example, several politicians received so many hateful emails that they no longer wanted to appear in public, the director assured. The incident is not strictly Swiss. All over the world, politicians are increasingly being attacked and threatened online, by womenExternal links Especially minorities.

Alliance F has therefore developed an algorithm against hate speech This algorithm is called “bot dog”External links“Like a dog, it identifies hateful messages and posts on social networks. A team of volunteers responds to all messages on a case-by-case basis. The idea is that online hate cannot be resolved and discussions can go on purposefully.”

At the moment, Bot Dog is in the testing phase. But the first test is final. Researchers at ETHZ and the University of ZurichExternal links Supported the project. Their observations: Responses to compassion for the target people by hate messages are particularly effective. Phrases such as “Your post is too painful for Jews” prompted hate speech writers to apologize or delete their posts.

In July, the bot dog should officially go live. Anyone can participate in the project, explains Sophie Acherman. Either way, evaluating comments helps the algorithm to more accurately identify hateful comments. Or by responding to hate messages identified by the latter.

A collaboration between algorithms and humans

Bot Dog Artificial Intelligence (AI) is just one example of how democracy can be strengthened. Dark Hellbing, Professor of Computational Social Sciences at ETHZ, highlights the immense potential of digital democracy. He thinks of a participatory budget, which allows citizens to decide together what they should dedicate to the financial family of their city or municipality.

In the Zurich district of Wipkingen, this method is being tested in real life. On a digital platform, residents opting for an urban park and a street food festival, these projects receive 40,000 francs per tax.

In the case of digital democracy, ideas abound. Cities and regions on the planet can, for example, network and compete in a competition to find the best solutions for sustainable economies, reducing CO2 emissions, or peaceful intercourse. Solutions and data will be accessible to everyone in open source and can serve as the basis for AI models. “I think we’re moving towards a more participatory digital society,” said Dark Hellbing. Is a collaboration between AI and humans, such as Bot Dog, where an algorithm and volunteers fight together against online hate.

According to this expert, the potential of AI for the benefit of direct democracy is immense. But data-driven apps “have a strong destructive potential for democratic forms of society based on the rule of law and human rights,” he warned.

Not all pink either

Why? Because algorithms increasingly determine what information we receive, what products are presented to us, and at what price. They define what we see in the world and what we value. This means that our political thinking and our electoral behavior are affected by it, Dark Hellbing explains.

Donald Trump’s 2016 presidential campaign is a good example of this. His team used the Big Data algorithm to personalize the content for his target group. “We bombarded them with blogs, websites, articles, videos, advertisements until they saw the world the way we wanted it to.” Until they vote for our candidate “, then in a Netflix documentary Brittany will explain to Kaiser” The Great Hack “. Brittany Kaiser then worked for Cambridge Analytics, which shut down 87 million Facebook profiles before the election, giving birth to one of the biggest data scandals to date.

Voters were also targeted during Barack Obama’s second presidential campaign in 2012. The president’s party has maximized its data collection and sent a personal message to each voter.

It is not possible to say for sure how much AI-based campaigning has contributed to the success of the candidates. But Dark Hellbing and his colleaguesExternal links See a big danger in them. In particular, the combination of targeting and nudging with big data associated with our behavior, our feelings, and our interests can lead to an omnipotent state, says Dark Hellbing. In psychology, the term nudging refers to a way of persuading or pushing people to do something.

Election campaign

All AI needs one thing: data. The more available it is, the better it works. In Switzerland, data protection and privacy are valuable assets. Will the methods observed during the American election campaign be possible there? Do Swiss parties have the latitude to reach voters through microtracing?

At the University of Zurich, Lucas Lyman works to measure public opinion using machine learning. She doesn’t believe it. According to him, the situation in Switzerland is not comparable to that in the United States. Switzerland does not have the raw information used during the American campaign. “They had a data point for almost every citizen. It has nothing to do with what is available in Switzerland. “

During the study, Lucas Lyman worked briefly for a fashion company in the United States. Not only was there access to the customer database – name, address and date of birth. But for everyone, he could know the approximate annual income, number of children, car model, tenant or the quality of the landlord.

“In the United States, you can easily buy this data; It is also used for political purposes. In Switzerland, to my knowledge, such information is not yet used or used in political propaganda. “

Not yet? Will we be able to afford them in Switzerland soon? In any case, Dark Hellbing has confirmed that the world’s databases are far more advanced than most people think.

The ETH expert cites the example of the Geneva-based Center for Cyber ​​Security of the World Economic Forum (WEF), in which companies from across the planet participate – Amazon, MasterCard, Huawei Technologies রয়েছে “contains huge amounts of data. Collected, “said Dark Hellbing. The collected data is used, among other things, with the blessing of the United Nations to implement the 2030 Agenda for Sustainable Development.

“It probably started with a good intention. But the question is how all this will be implemented and what more can be done with this data. The number of firms involved is such that the risk is very real that commercial interests take precedence over society.”

More transparency

Laws and regulations need to be implemented now to prevent AI from becoming a threat to democracy, experts warn.

But to establish the rules that work, you need to know exactly how the algorithms work. That’s the problem. Anna Metzner, director of AlgorithmWatch Switzerland, explains that a platform like Facebook is a black box. “We don’t know how the algorithms work in detail. We also don’t know exactly what data was collected and from whom. “

AI curating social media content is a well-known privacy. To find out how algorithms affect election campaigns, AlgorithmWatch has launched a research project in collaboration with Süddeutsche Zeitung.External links. They asked hundreds of volunteers to put data on their Instagram feed before the 2021 German parliamentary elections. Volunteers had to subscribe to all party profiles A browser plug-in has recorded where and how the content of these profiles appeared in users’ newsfeeds and transmitted this information to Algorithwatch.

The analysis found that Alternative for Deutschland (AFD) content ranked higher on Instagram’s news feed than other parties’ content. Why? Search failed to answer.

But Instagram owner Meta (formerly Facebook) doesn’t seem to appreciate the study. The group threatened AlgorithmWatch that “more formal action” would be taken if the project was not abandoned. The exchange of information has therefore ended prematurely.

“Unless this type of research is possible, Anna Matzner notes, we cannot issue a statement based on the impact that the content collected by AI on the platform has on society. And especially in the formation of political opinion, so directly on democracy. “

According to JTI standards

According to JTI standards

Further: Certified by SWI swissinfo.ch Journalism Trust Initiative

Leave a Comment