Lesen Sie diesen Text auf Deutsch

In summer 2017, some 12,000 Germans registered to participate in the ZEIT ONLINE project "Deutschland spricht". They agreed to meet up personally with someone from their neighborhood or region who held opposing political views and to have an hourlong discussion with that person. Here is how it happened.

As you read this article, more than 600 pairs of complete strangers are meeting up for what we hope will be a thoughtful political debate – at a café in Berlin, in a church in Dresden or at a beer garden in southern Germany. Participants include those opposed to refugees and those who volunteered to help them, they include opponents and proponents of nuclear power, and they include champions of the European Union and those who would like to see the return of the deutsche mark. They are all strangers to each other and have very different world views. They've been brought together by ZEIT ONLINE with the help of an algorithm that we refer to internally as political Tinder. But let's start from the beginning.

The starting point for our project was a question: If entire swaths of society have forgotten how to speak to each other, how do we get them back into dialogue? How do we get them to talk to each other face-to-face? The latest research, after all, shows that one of the few possibilities for viewing things in an entirely new light is to see them through the eyes of others. The filter bubble, it turns out, is in our heads: We frequently dismiss facts that contradict our own convictions as being wrong – or we simply ignore them.

That's why, at the beginning of May, we posted a question to our homepage: "May we introduce you to someone?" We asked our readers five "yes or no" questions chosen for their ability to clearly separate respondents into opposing camps – questions like: "Has Germany accepted too many refugees?" or "Is the West treating Russia fairly?" We then asked for a few more personal details from respondents, including their zip codes, mobile phone numbers and email addresses. The pledge we made at the time also proved to be a source of worry for us in the weeks running up to the experiment. We said we would find a partner with entirely divergent political opinions for as many people as possible and would challenge the pair to meet up for a debate once a match had been made.

There was plenty that could go wrong

Within just a few hours, some 2,000 people had already registered. By the time we took the online form down four weeks later, that number had grown to 12,000. At that point, we began putting considerable effort into the selection process – an effort driven by a concern. We were attempting to pair up thousands of people who were complete strangers to each other and who held contradictory political views for an unsupervised discussion. There was plenty that could go wrong.

First, we filtered out all the entries for which the mobile telephone number was not located in Germany or could not be reached, thus eliminating around 1,700 participants. A further 4,500 did not respond to the text message we sent to their mobile phones. The ability to reach people on their mobile phones was an important condition because we wanted to be certain that the registrations came from real people. Ultimately, there were 5,500 participants we could introduce to each other.

When it came to actual matchmaking, two conditions were paramount: Dialogue partners should have the most divergent opinions possible; and they had to live within 20 kilometers of each other. The fact that each participant submitted their postal code meant that we had a rough idea of where they lived. (We deliberately didn't ask for actual addresses.) For each potential match, we examined how many of the five questions each person answered differently. The results were arranged in a table, with values ranging from zero to five. Zero meant total agreement. Five promised heated discussions.

The image rendered on our computer screens resembled a spiderweb. All pairs who lived within 20 kilometers of each other were connected by virtual strings. In addition, for each string, we also saved the number of questions that the pair connected by it had answered differently. In the end, there were more than a half-million connections in the network.

We had to select as many strings as possible from the tangle, but also only one string per participant – and one that linked that person to someone else with the highest possible number of differing opinions. "Finding maximum weighted matching in graphs" is a classical mathematical problem. In the mid-1960s, American mathematician Jack Edmonds discovered a quick algorithm for handling the problem. We built it into our software and had the computer do the calculations – a process that still took a half an hour.