Can he predict the activity pattern of ISIS?
On 13 November 2015, a series of coordinated terrorist attacks, including mass shootings and suicide bombings, occurred in the downtown of Paris, France, causing more than one hundred people died. According to the news from Associated Press and Euronews on November 14th, terrorist attack has already caused 129 people killed, 352 people injured, including 99 were seriously wounded.
ISIS’s brutal actions inflamed the world. However, terrorists’ activities are not random; they have certain patterns. And nowadays, Artificial Intelligence(AI) is the one that is best at pattern recognition.
Paulo Shakarian is a computer science professor in Arizona State University, United State. He had went to The United States Military Academy at West Point for his bachelor, and had worked for U.S. Military as an Intelligence officer. These experience made him very familiar to the activities of terrorists. When he was little, Shakarian thoughts that war was far away; but the September 11 attacks changed his mind when he was a senior at West Point. In 2002, he was sent to Iraq as an Intelligence officer in the U.S.Army 1st Armored Division for 14 months, and then became the platoon commander in the 501st Intellegence Brigade quartered in Germany. During his time in Germany, Shakarian collected, processed, and analysed plenty of intelligence information for the army, and received a Bronze Star Medal.
Due to his job, Shakarian often worked with the intelligence officials in the front line. This made him realize that various intelligence technologies can be applied into this field. Intelligence officers need to analyze all the data they got and set hypotheses for their reasoning, but they barely have time to get back and check, especially during the war. In Iraq, Shakarian started to use the computer science knowledge he learned at West Point to do information analysis, and have ideas about using AI methods to build models for terrorism activities, which was thought to be unpredictable.
After that, he enrolled into Maryland University, and got his master degree in Computer Science. His goals was very clear: using machine learning methods to change information analysis field. Machine learning is also the technology on which self-driving cars relied. After he received his PhD. degree in Maryland University, he continued his research on AI in Arizona State University.
His research incubated many projects that have revolutionary potential, such as the SCARE software the Task Force Paladin used to detect Improvised Explosive Device in Afghanistan, and the social media analysing tools, GANG and SNAKE, the Chicago Police Department used to reduce gang activities.
He published a few paper in this field in recent years. Let us see what he found. Can he predict the activity pattern of ISIS?
Get rid of the leader of terrorists:
In November 2012, Shakarain published a paper, named “Shaping Operations to Attack Robust Terror Networks”, aiming at destroying the leader spawn system in terrorist groups. This paper caught the attention of United States House Select Committee on Intelligence, and got Shakarain himself invited to talk about his discovery.
In order to destroy terrorist groups and revolt networks, safe organizations often intend to attack the fixed points, “High Value Target”(HVT), which are their leaders. However, thousands of examples proved that after the leader died, the revolt network will select a new leader in a very short period of time. For instance, on July 8th 2006, the notorious leader of the base organization, Abu Musab al-Zarqawi, resulted in death, a new leader called Abu Ayyub al-Masri was selected in one week.
How to destroy this leader reproducing system? Shakarain introduced a concept called “shaping” activity, which means that, before attacking the leader, we should first focus on destroying the ability of reproducing leaders in that terrorist group, and then get rid of the leader in a normal way. In this way, the terrorist network will have trouble re-selecting a new leader, and coming back.
Then, how to do that? By removing nodes, Shakarain used network theory to maximum the network centrality. Intuitively, network centrality stands for the critical value of the height nodes. Since the networks with low centrality are easier to be decentralized, it will be easier for them to select new leaders. In this paper, they focus on finding those targeted nodes, which can be maximum centralized, so that the following attack to the fixed points will be more effective.
Before, this problem was proven to be a NP-Complete problem. For practical purposes, they introduced a mixed integer-linear program that was equivalent to greedy heuristic search. After executing such greedy heuristic search, when they were checking 5 real-world terrorists networks, they realized as long as they remove 12% of the overall nodes, the network centrality could be improved by 17% ~ 45%. This algorithm can also be applied to the network that contains 1133 nodes and 5541 edges.
Mining the associativity among the ISIS actions
The ISIS’s actions are not completely random; there are some patterns. This is Paulo Shakarian’s incredible discovery. This August, Paulo Shakarian’s team published a paper called Mining for Causal Relationships: A Data-Driven Study of the Islamic State, to present their results in the Conference on Knowledge Discovery and Data Mining.
In this paper, they analyzed the data, provided by the Institute for the Study of War, of 2200 cases relevant to ISIS, and created a descriptive model — — an algorithm that imitates ISIS’s actions. Those 2200 cases happened in the second half of 2014, which includes not only the military operations to surround ISIS, but also fight against ISIS (include the union that Iraq, Syria, and United States led). Among all of these cases, they combined logical programs with cause-effect deduction, trying to mine the causality between cases.
Therefore, they found some rules. The predictions of these rules are the results of combining many atomic propositions (not including the propositions that are parts of the combination, in other words, it is the proposition that can not be broken into other propositions in structure ), filtering these propositions with the rule of comparing the same-order events, and at last getting causality from the filtration. Beside the probability of considering rules (p), they also did research about the measure of causality (εavg). This can be seen as the probability provided by the preconditions of rules will improve at the same time while some other similar rules are considered.
At the end, they did find some patterns:
1.If ISIS has infantry actions with InDirect Firing (IDF) in Iraq at a certain week, then the next week there will be a Vehicle-Borne Improvised Explosive Device(VBIED) in Syria. (p=1.0, εavg=0.92)
2. If ISIS has operations in Tikrit, Iraq, like mass murder, then Iraq and Syria will have a mass amount of Improvised Explosive Device(IED). （p=1.0，εavg=0.97）
3. In the week right after Syrian government are air attacked, ISIS always detent a large number of hostages (p=0.67, εavg=0.91). Also, in their database, similar mass detention always happened after air attacks in Syria.
4. If ISIS is taking actions in Al Anbar Governorate, Iraq, while the union army is air attacking Mosul, Iraq, then in one week after the air attack, ISIS will have drastically more IED actions in Iraq (p=0.67，εavg=0.97). However, if at the same time of the air attack, ISIS is also having operations in Syria, then the use of the IED there will increase as well. (p=0.67，εavg=0.79)
Based on these patterns, they drew a few conclusions below:
- Before having a large scale of infantry actions in the areas beside Baghdad, ISIS might have suicide car bombs in Baghdad to weaken the improved deployment of Iraqi army / police.
- Before a mass scale of infantry actions, ISIS intends to use IDF as the prologue — — matches the traditional army power better; they do not use IDF to make disturbance (while this approach is quite common among the rebel groups during the free action period in Iraq).
- There is some sort of relationship between Union army air attacking and the use of IED, but it has not stimulate other bigger weapons to occur(for example, car bombs). This might mean that after this kind of actions, ISIS relies more on the strategies that are scattering and rebellious.
Shakarain claimed, this paper shows 「The Evidence of Concepts」, which did not really count as the real 「Big Data」, but the results are remarkable. Due to the limited data, they can only build models by focusing on the actions in the past.
The United States Department of Defense was very interested in his research. This might imply that the Pentagon will accept more and more of this kind of data-driven computer science research methods. They will continue research on more complicated temporary relationships, possibly use more detailed time zone time, and analyze a few time units at once. In addition, they will research on the environmental variables (such as weather, information, social media, political situation, etc.) to find more pattern, which can give anti-terroristic strategies certain help.
Moreover, Artificial Intelligence will complete more anti-terrorist task, like face perception, and analyzing the human behaviors in the supervisory video recording, etc.. Hope more researchers participant in this field, and help machines to protect our lives. Nonetheless, only use machines to maintain peace is not possible, what is more important is people’s minds. As Einstein stated,
Peace cannot be kept by force; it can only be achieved by understanding.— — Albert Einstein
If you want to know more about Paulo Shakarian and his research, please click : Big Data vs ISIS(1); Big Data vs ISIS(2).
Report Members: Jiaxin Su, Gabrielle, Wangwang | Editor: Synced
