Earlier this week, Conservative leader Andrew Scheer tweeted screenshot of some curious Google search results. Searching for "Canadian soldiers" returned the photograph of former Guantanamo Bay detainee Omar Khadr, who was accused of killing an American soldier in 2002.
Scheer asked Google to take steps and did not take long before another user could indicate that the whole thing was part of the Russian trolley.
Omar Khadr is a convicted terrorist who murdered a doctor and blinded another. It is not a sacrifice, nor should it be displayed in this way next to real Canadian heroes. @googlecanada: Correct it. pic.twitter.com/qywUGQihVb
Truth? It's much more common.
The episode is another reminder that even the best-intentioned algorithms can unknowingly support the spread of disinformation on the Internet. And with the Canadian federal elections, which are only months away, courses are even higher when it comes to politics.
How did Khadr get there?
The name Khadra has appeared in what Google calls the results of the Knowledge Graph. They sometimes appear above or beyond the usual Google search results when a user asks for a question, looks for some general knowledge, or searches for a known place or the public.
The Knowledge Sketch draws data from a variety of sources – one of which is Wikidata, an open information store hosted by the same organization hosting Wikipedia. Think about Wikipedia as a completed report and Wikitrage about the raw data that was used to write it. Like Wikipedia, everyone can contribute to Wikipedia for better and for worse.
Twitter user Stephen Punwasi pointed out that the datasheet used to make Omar Khadr among the Canadian soldiers seemed to have been pulled out of Omar Khadra's Web site – and that "the Russian troller" was the one who did it.
This is embarrassing.
Google's graph of knowledge draws data from a variety of sources, so I've checked the change that Omar Kadh puts under a "Canadian soldier."
Changes to the Russian account a month ago. 🤦♂️
So it was the work of the Russian trolley?
It does not seem to be the case.
The Changes to the Khadra Wikidas were created by a user named Ghuron. The user appears to be an active data contributor to the Web, and according to his Github account, he lives in St. Petersburg, Russia.
However, Ghuron's activity does not seem to be focused on specific person, ideology, country, or political themes. Rather, it resembles automated cleaning work to improve the quality of Wikidats at a rate much faster than any person can do by hand.
According to Ghuron runs a script that uses machine learning to automatically add and modify large volumes of Wikidata data (such as a person's job). Basically, it's designed to put data into the bucket.
His screenplay will ensure that Street Fighter is correctly classified as a video game, that the Faroe Islands are under a larger category of "islands", or that administrative renaissance artists are properly classified as painters. Just look at yourself.
From time to time it seems that his screenplay is also bad – and other Wikidas users did not care to know it.
Did that happen to Khadre?
Yeah! Using Khadr's The Wikidata page edits history as a guide, here is a short timeline that has returned far beyond Scheer's tweet:
July 26, 2018 was Ghuron's script categorizes the occupation of Omara Khadra on Wikipedia as a "soldier" – part of a larger, automated effort to assign jobs to everyone from Zodiac Killer to the Danish priests.
On September 24, 2018, users start posting to Omara Khadra's Wikipedia section and ask why Google's search results call it a Canadian soldier. However, the phrase "Canadian soldier" never appeared on its Wikipedia page, which means that the phrase was probably extracted from its Wikidata site instead. Google has to explicitly confirm this.
Later that day, data is removed from Khadrov's Wikidas page. The user in Khadrovy Wikipedia wrote: "It was a bug of Google, linked only to their search engine and was not encoded by Wikipedia." Google has corrected the error today and Omar Khadr is no longer a Canadian soldier. "
September 30, 2018 Ghuron's script categorizes the occupation of Omara Khadra on Wikipedia as a "soldier" once again. It's not clear if this Google Knowledgebase ignored this change.
Either way, on December 8, 2018, Ghuron's script then categorizes Khaddar's military rank on Wikidat as a "soldier" – a data that is so different that he has probably returned to the results of the Knowledge Graph. Soon Khadr noticed the results of the "Canadian soldier".
What did Google do about it?
Danny Sullivan, Case Thing Google has an ombudsman search engine, Canadaland replied to Jesse Brown on Twitter.
I'm not familiar with who says what. As has been said, we have simply seen the fears that have been widespread. We've reviewed, and since there was a Knowledge Graph issue, we took steps there (we do not take steps to search listings). And yes, the last time, because it's accurate.
"We reviewed it, and because it was a question with the Knowledge Graph, we took action there," Sullivan wrote.
It seems that Khadro's Wikidas page has not been changed – only the Knowledge Graph processes the data it contains.
CBC News has contacted Google and updates this story if we hear more.
It is normal?
As Sullivan also points out Twitter, Google will not change search results – at least not if it does not have to remove information from the index. Rather, Google is changing its Knowledge Graph results.
Whether this difference is obvious to most users, since its location – especially in situations where the political tension is high – is less clear.