Social media monitoring combined with Facebook’s AI mistranslating a post led to a man being wrongfully arrested for “suspicion of incitement.”
A smiling Palestinian construction worker posted a photo of himself leaning against a bulldozer and holding a cup of coffee and a cigarette. He posted the photo on Facebook along with “good morning” in Arabic.
Israeli police, relying on Facebook’s translation service, believed the post said “attack them.” Haaretz reported, “The automatic translation service offered by Facebook uses its own proprietary algorithms. It translated ‘good morning’ as ‘attack them’ in Hebrew and ‘hurt them’ in English.”
The Palestinian man’s photo showed a bulldozer on the construction site in the Beitar Illit settlement which is located on the West Bank near Jerusalem. Bulldozers have been used in terror attacks against Israelis in the past.
Israeli police did not consult any Arabic-speaking officer, but instead relied completely on Facebook’s translation. As a result, the Palestinian man was arrested.
Haaretz explained, “Arabic speakers explained that English transliteration used by Facebook is not an actual word in Arabic but could look like the verb ‘to hurt’ – even though any Arabic speaker could clearly see the transliteration did not match the translation.”
The Times of Israel added, “There is only one letter’s difference between the colloquial Arabic phrase for ‘good morning to you all’ and ‘hurt them’.” A spokesperson for the Israel Police’s West Bank district confirmed the arrest of the construction worker to The Times of Israel, adding that the man was released after a few hours when the cops realized Facebook mistranslated the “good morning” post.
Facebook Artificial Intelligence Researchers (FAIR) published a research paperand toolkit for its convolutional neural networks in May. This new machine translation approach reportedly could achieve “state-of-the-art accuracy at nine times the speed of recurrent neural systems.” Wired suggested, “Facebook’s new AI could lead to translations that actually make sense.”
The AI researchers explained that their better translations were a result of multi-hop attention and gating.
A distinguishing component of our architecture is multi-hop attention. An attention mechanism is similar to the way a person would break down a sentence when translating it: Instead of looking at the sentence only once and then writing down the full translation without looking back, the network takes repeated “glimpses” at the sentence to choose which words it will translate next, much like a human occasionally looks back at specific keywords when writing down a translation. Multi-hop attention is an enhanced version of this mechanism, which allows the network to make multiple such glimpses to produce better translations. These glimpses also depend on each other. For example, the first glimpse could focus on a verb and the second glimpse on the associated auxiliary verb.
Better translations are a good thing, but it seems like a human might be consulted about an AI translation before a person is arrested.
Like most government agencies, Israel’s security services conduct social media monitoring; it led to the arrests of 400 Palestinians suspected of planning attacks in April. Israel claimed its social media monitoring had stopped a total 2,200 Palestinians who were planning attacks.
The Palestinian construction worker deleted his “good morning” post after he was arrested and then released.
