Categories
Perspectives

Can we model violent behaviour – or predict it?

Big data is everything these days.  There is absolutely no question that there is more information out there than ever before and more ways to collect it.  Everyone from companies to police forces are getting into the game with a view to earning more money or making their jobs easier and more efficient.

Nor is there any doubt that having more data is better than having less.  This is especially true when it comes to terrorism studies.  I have long complained that much of the literature to be found is highly theoretical and shows little evidence of being data driven or, where there is some data, it is seldom robust enough to support the claims made.  Of course in a field like terrorism, where those with the bulk of the information are law enforcement and security intelligence agencies, it is not surprising that data is generally unavailable.

This shortcoming has done little to stop well-intentioned individuals from devising models and programmes to describe, explain and,  in some cases claim to predict, violent extremism.  Over the course of my time at CSIS I saw many of these efforts and while I found most to have interesting and useful elements, none were perfect and none could have been applied to a real, constantly-changing operational environment.  In fairness, the most honest providers of these models would admit that their products were tools and not infallible instruments.

There is of course a great danger in seeing any model as Gospel.  Models are only as good as their inputs and the data used to construct them.  If small data samples preclude certain phenomena, the model cannot handle them.  And if that same data stream is heavily weighed in a biased direction it will also fail.  This is the problem known as false positives and false negatives.

Despite these gaps, models keep getting pushed forward as indispensable tools for our law enforcement and security intelligence officials.  Here is an example.  Recently, Correctional Services Canada has been struggling with the very controversial issue of segregation.  The government, and the public, want to see the numbers of inmates kept in isolation lowered and the pressure is mounting on CSC. For its part, the union representing prison staff has raised the danger to their members should the use of segregation diminish.  And in what is even scarier, and more germane to this blog, the union fears that what was once a decision made by humans has now been relegated to a computer questionnaire that determines whether an inmate should be isolated.

In other words, we are allowing a machine with algorithms to decide who poses the greatest threat to staff and other prisoners.  I have no idea what those algorithms are or what data (and how much) was used to create them, but am I the only one who sees this as mad?  A field that requires human input is now left to a programme.

I fear that this rush to automate everything – from manufacturing to driving to medical diagnosis – will end badly.  At a minimum, human livelihoods in the form of jobs are at stake while at a maximum actions will be taken (or not taken) based on models that may not actually be accurate.

Perhaps it is time to recognise that some things, like why people become terrorists, are unknowable in any useful generalised fashion.  Perhaps it is time to acknowledge that for some things “it depends” is as good as it gets.

By Phil Gurski

Phil Gurski is the President and CEO of Borealis Threat and Risk Consulting Ltd. Phil is a 32-year veteran of CSE and CSIS and the author of six books on terrorism.

One reply on “Can we model violent behaviour – or predict it?”

Big Data is a tool, much like intellect, reason, a police car, anything. The choice to use this tool is still in the hands of a human being. No matter what we may like to believe with technology, we know that it is not always correct. I don’t think that predictive analytics purports to stake that claim either. Let’s look no further than the weather, i don’t recall seeing a 100% chance of precipitation ever. However, given the data, they predict with some reason of accuracy how likely it is that it will rain. Similarly, I think it’s time that we use a tool like Big Data analytics to predict who may be headed down a path of violent extremism. It’s a tool at our disposal.

I’m currently finishing up my Master’s at Brandeis University. The last course I’m taking is a really neat one, it’s sports analytics. The idea that we will be able to predict with some accuracy, how many wins a team might have in a season, or which players we would want to sign based on the numbers. Of course the first thought that comes to many people’s minds is Moneyball, and sure while that has something to do with it, there is a plethora of work done throughout the analytics community to start studying what the numbers mean and what they can predict. I think what I’m trying to say is that analytics is a different approach to doing things the same old way. The example I will give is that of coach Kevin Kelley from Pulaski Academy in Arkansas. You can read more about him at https://www.washingtonpost.com/news/sports/wp/2015/08/13/the-highly-successful-high-school-coach-who-never-punts-has-another-radical-idea/. Using analytics, he has determined it is almost always the right idea to go for it on 4th down, and also to always go for onside kicks. That’s completely against conventional wisdom, but it’s working. His record speaks for itself.

There is a growing interest in predicting attacks via social media. There are analytics that are being run daily on social media sites like YouTube, Twitter, and Facebook to predict the next attack. Much like the deep web, I don’t think a great deal is known about the research efforts into analyzing social media for threat communications. Despite what Trump might think, I think researchers and government agencies are keeping their cards close to their chest on broadcasting. I believe Google has come up with an algorithm, or they are working on one to predict terrorist attacks, and as recently as June, there was an algorithm put forth by physicist Dr. Neil Johnson of the University of Miami which argued that based on Social Media scans attacks may be able to be predicted. http://www.nytimes.com/2016/06/17/science/fighting-isis-with-an-algorithm-physicists-try-to-predict-attacks.html If social media is where the terrorists are, then we have to be a step ahead of them in combatting the threat. The government’s haven’t always been synchronous with technology trends. (I think they’re still using BlackBerries) The internet can tell us a lot about what these lost souls may be thinking. It’s a matter of sifting through all of that data quickly. We have already run out of Ipv4 IP addresses and have moved on to Ipv6. Each minute the internet content is growing: 2.5 Million instagram posts , 400 hours of video on YouTube, 3 million facebook posts. These numbers translate into daily advances of 40 million tweets, 4.3 billion facebook posts, 4 million hours of Youtube content, and 6 billion google searches. (https://www.gwava.com/blog/internet-data-created-daily). Somewhere buried in all of that could potentially be the key to preventing an attack. I think we owe it to do what we can.

The numbers whether we’re looking at football, basketball, baseball or shopping, are the probability of the likelihood of any event. At the end of the day if there’s a 90% chance something will happen, it also means that there is a 10% chance that it won’t. This is where I can see the hesitation. If the computer says a person X is headed down the path of violent extremism, and the government has invested money in preventing something from happening and person X is a person of interest because of the computer, then I think a few things happen. Firstly, the need for the computer to be considered right means that everything person X does will be scrutinized in a fashion to make it fit a particular narrative (dangerous). Secondly, if person X is genuinely not doing anything to go down that path and they’re in the 10%, then that was a waste of tax payers dollars (dangerous). Finally, even a margin of error of 10% on 1000 people is still 100 innocent people who are now being surveyed despite being otherwise completely innocent.

All in all, big data analytics has come a long way to predicting human behaviour, but at the end of the day it is human behaviour, there is a margin of error. I don’t think anybody is suggesting to take it as Gospel and follow it blindly, but using it as a tool to predicting the future is something to be strongly considered.

Leave a Reply