Family firm with Nazi past funds effort to use AI to detect online anti-Semitism

As social media platforms take steps to block Holocaust denial, an interdisciplinary project is harnessing artificial intelligence (AI) in the battle against online anti-Semitism.

“Decoding anti-Semitism” is a collaboration between King’s College London and the Technical University of Berlin’s Center for Research on anti-Semitism. The project is funded by the Alfred Landecker Foundation with 3 million euros ($3.5 million), the German news agency dpa reported.

The Landecker Foundation was created last year by JAB Holding Company, based in Germany. Owned by the well-known Reimann family, the firm’s Nazi-era predecessor made use of at least 838 forced laborers during the war. Today, JAB Holding Company owns subsidiaries including Panera Bread, Au Bon Pain, and Krispy Kreme.

The company’s wartime leader, Albert Reimann Junior, was a devout Nazi who had an ongoing relationship with the daughter of a Jewish man — Alfred Landecker. Their “forbidden” romance produced three children. Landecker was taken by the Gestapo to a ghetto in Izbica, today’s Poland, where his tracks faded. It is assumed he was transferred to a Nazi death camp where he perished.

In 2016, the Reimann family began to commission research into its Nazi ties. “But the Reimann family was not satisfied with simply uncovering its own past. The company heirs are also concerned with supporting Holocaust survivors and, as a lesson learned from history for today, promoting democracy and human rights to preserve and expand a pluralist society,” reads the Landecker Foundation website.

‘Learn directly from the human’

The newly minted Landecker Foundation’s central mission is to share the history of the Holocaust and fight anti-Semitism, particularly on the Internet. Through the foundation’s “Decoding anti-Semitism” project, an open-source detection tool is being created to align with social media platforms, said lead designer David Allington.

Benckiser Factory in Germany, owned by the Reimann family, the current owners of JAB Holding Company. During Nazi rule, there were 175 forced laborers used at the factory. (courtesy: Alfred Landecker Foundation)

“The likes of Facebook and Twitter’s ongoing efforts to combat hate speech online are well-intentioned but fall short of what is required,” said Allington, a senior lecturer in social and cultural artificial intelligence at King’s College London.

“Where our approach differs to that of [Facebook and Twitter] is that we will use not only AI but also the support of historians, linguists and other disciplinaries to recognize [anti-Semitic] sentiments,” Allington told The Times of Israel.

The “complexity” of anti-Semitic expression on the Internet calls for an interdisciplinary approach, said Allington. For example, the UK has seen a surge in “conspiracy myths” accusing Jews of creating and spreading COVID-19. Much of the language and images are “coded,” said the designer, and couched in other, less familiar terms.

According to AI expert Katie Hall, there are significant limitations when it comes to using AI to identify and combat hate speech. Some of those limitations come from the data sets used by Big Tech companies, she said.

“Men and women and different races and ethnic groups describe themselves differently,” said Hall. “We know these different groups just use language differently, and most historical data is created by specific groups — college-educated and white.”

Katie Hall, founder and CEO of Claira

As founder and CEO of the AI-powered hiring marketplace Claira, Hall helps companies recruit more diverse pools of candidates for their positions.

“We are trying to level the playing field and be deliberate and careful about removing bias,” Hall told The Times of Israel.

Currently, data sets rely heavily on analyzing text, said Hall. Few algorithms have been trained with — for example — audio exchanges or filmed encounters. However, said Hall, there is “movement underway to integrate voice and conversation and video into the data sets.”

According to Hall, “there is a need to learn directly from the human. This is the opportune moment right now — whether we want to figure out a way to detect bias or we don’t. Humanity is so varied, and the data sets are really far behind.”

‘Fighting with one hand behind our back’

“Decoding anti-Semitism” is hardly the first effort to fight hate speech online. In the past few years, dozens of foundations and organizations have attempted to address increases in hate speech against Jews and other groups.

“Whatever these organizations are doing, it isn’t working,” said Andreas Eberhardt, founding director and CEO of the Alfred Landecker Foundation.

“That is why we have adopted such a collaborative approach and over the coming weeks, months and years the partnership will see discourse analysts, computational linguists and historians come together, combining their expertise for one common goal,” said Eberhardt in an interview with The Times of Israel.

Andreas Eberhardt, founding director and CEO of the Alfred Landecker Foundation (courtesy)

“Such collaboration is the only way to make progress on this issue and not doing so would be like fighting with one hand behind our backs,” said Eberhardt, who had led several foundations committed to German-Jewish relations and Holocaust memory.

According to senior lecturer Allington, most defamations against Jews on the Internet are “unsanctioned.” And before defamation can be sanctioned — or punished — it must be identified, he said.

“Once [the hate speech] is revealed, as our project hopes to do, then the scale of the issue will become clear and those organizations that are in a position to remove such content — the platforms on which they are posted — will be compelled to act and we will be putting pressure on them to do so,” said Allington.

From anti-Semitic terror attacks in France to the Pittsburgh synagogue shooting in 2018, online hate speech can lead to real-world violence.

“We see that hate speech online and hate crimes are to some extent always connected,” said Matthias J. Becker, a linguist with the Technical University of Berlin and project lead for “Decoding anti-Semitism.”

“In order to prevent that more and more users become radicalized on the web, it is important to identify the real dimensions of anti-Semitism — also taking into account the implicit forms that might become more explicit over time,” said Becker.