Channel 10 reported earlier this month that the Israel Defense Forces and Shin Bet security service had arrested 400 Palestinians who were seen as lone-wolf terrorists. They were “arrested before they set out to carry out terror attacks.” The suspects were identified through the use of a computerized program that analyzed social media posts.
The station's military correspondent, Or Heller, attributed the drop in the number of terrorist attacks to the cyberprogram. What made the story even more bizarre was the fact that a few minutes earlier, journalist Ronen Bergman – who reported the story – had said the exact opposite: That despite the system’s existence, it was impossible to identify lone-wolf attackers.
The Channel 10 report was followed by an item in Haaretz, which claimed the method had "significantly reduced” the number of terror attacks. But it hedged that claim and spoke of 400 detainees who had intended to carry out terror attacks, not that they were actively caught just before setting out to commit them.
But what was lacking in these reports were the burning questions: How is it possible for Israeli security forces to arrest young Palestinians over things they have not done? How can they be convicted of a crime when they haven’t done anything? And how is it that such a dystopian TV report could appear without any criticism?
The repeated claim that use of the system has achieved substantial results is an amazing one that is impossible to verify. By the same measure, those who attribute the drop in the number of attacks to an algorithm could also attribute it to the average temperature this spring in comparison to last.
Even if the system has managed to identify several suspects who may have planned to commit a terror act (that was not carried out), that certainly doesn’t amount to “significant” results that could begin to justify such a grave violation of hundreds of detainees’ human rights.
Explaining the extent to which the approach is baseless and to which it violates any pretense of due process requires first explaining how such systems generally operate, and that there are Israeli companies working on such systems.
The first step involves gathering information in the public (or semi-public) domain from the social networks. Then the decision has to be made over what to do with the large amount of information. Low-level scrutiny could include monitoring of specific keywords on websites – for example, “shahid” (the Arab word for martyr). On that basis, anyone writing such a word could be subject to special attention from the system. But such an approach would be inefficient and yield too many possible suspects. More precise analysis is required.
At this stage, semantic similarities are considered between social network posts and what people who have already been singled out as suspicious are writing. In other words, the number of people singled out as potential suspects is expanded simply based on their style of writing.
There are basically two kinds of systems purporting to produce predictions based on such information. Once the information is received, it can be analyzed to reinforce patterns and correlations among various variables. Let’s assume, for example, that we think someone intending to commit a terror attack writes a social media post that contains the names of others who have already carried out terror attacks in the past and that he changes his profile picture. The large amount of data collected could then be used to examine the assumption and support it. From that point, a person whose user profile is similar should also come up in the system. Information Haaretz has obtained shows that Israeli psychologists have sat in on the interrogation of Palestinians arrested under such suspicions, apparently to develop a collection of rules of thumb.
Another approach involves building a system that looks for patterns that we don’t necessarily recognize. One such example is deep learning systems, one of the best-known being Google’s AlphaGo, which is capable of what would previously have been labeled human intuition.
A similar system, Deep Patient, is currently being used in experiments to diagnose illnesses. Despite its success, there is a lively debate over how to view its results in light of the fact that the system doesn’t explain its predictions – which by their very nature are based on probabilities.
These and other systems always involve a correlation among characteristics but never causation. Even if a system can demonstrate success in 10 percent of cases, it would be unthinkable for a reasonable person to incarcerate someone based on a 1-in-10 chance he would carry out a terror attack.
So what do they do in the security forces with clear information that even military judges won’t accept as evidence? One option that has been used is tracking such a suspect. But that’s not practical to do on a large scale. Another option is administrative detention (detention without trial). Even if it is almost inconceivable to arrest someone under such circumstances – in light of information based on such probabilities – Palestinians don’t benefit from human rights, so this option has also been used.
The problem is that administrative detention has its limits. The Israeli public does accept its use without oversight in practice, but it’s hard to justify such detentions to the rest of the world.
So how is it that 400 Palestinians were arrested by Israeli forces, and another 400 by the Palestinian Authority based on information provided by Israel?
A degree of speculation is required here, but it seems reasonable to assume that the system identifying the suspects included evidence of other suspicions, maybe involving incitement.
Nery Ramati, a lawyer who represents many Palestinians before the military courts, told Haaretz that the real aim of the arrests for incitement is to stave off future crimes. “One can conclude this, for example, from the fact that when the incitement suspects are released, in many instances they are transferred to administrative detention.” In other words, when the weak claims of incitement fail, the appearance of well-founded allegations evaporates and the suspect is then imprisoned without knowing why.
Even in Israel itself, it is almost only Israeli Arabs who are jailed for incitement offenses. Data from a Freedom of Information Law request to the police revealed that over the past year, indictments were filed against 37 Israeli Arabs and no Jewish Israelis. Although a “hate report” issued by the Berl Katznelson Foundation stated that right wingers incite on the internet every three minutes, it is only Arabs who are arrested. In 2014, MK Ayelet Shaked (Habayit Hayehudi) called for the killing of Palestinians, a day before a settler burned a Palestinian teenage boy, but she was not arrested either.
It appears that the claims regarding the results were disseminated by the Shin Bet based on probabilities that are impossible to verify, as well as absurd wishful thinking. The aim of this is to create a consensus among members of the public in Israel and overseas for groundless imprisonment while exploiting fears of terrorism.
Such systems are in use around the world. But it’s not for nothing that a lone individual is the target of such allegations. Even if such systems are used as a statistical tool, imprisoning people on the probability that they will commit an offense is clearly unacceptable. Even if most Israelis are satisfied by the Shin Bet’s baseless claim that lives have been saved thanks to the system, one must remember that such tracking is done in Arabic.
Public Security Minister Gilad Erdan also proposed creating a database of Israelis who support the boycott, divestment and sanctions (BDS) movement against Israel. Ultimately, anyone who objects to the government’s activity could find himself on a list of potential culprits for future crimes.
Thanks to blogger Noam Rotem for help in preparing this report.
Want to enjoy 'Zen' reading - with no ads and just the article? Subscribe todaySubscribe now