Data to Predictions
What’s the catch?
“I didn’t see that one coming,” said no data scientist ever. Not because data contains answers to every question. But because they are prepared for any finding that the data might throw at them. With the advent of data-centric civilization, data analysis has become a new tool that everyone must-have. A lot of organizations believe data analysis can do wonders for them. So the question that arises now: Is that accurate? Can data answer anything? In this article, we learn about an approach to stack data technology against crime prevention.
There have been instances in history where data could have been used to predict upcoming events. From wars to famine, to eclipses, everything was within the reach, if only one looked close enough. One of the most traumatizing events in US history was the attack on Pearl Harbor. It was quick and sudden and those who got entangled in the ‘Tora Tora Tora’ couldn’t have imagined the US being attacked. So much so that, even during the attack most soldiers thought it was an air-force practice drill. But was it sudden? The answer is No. The investigations done in the aftermath of the Pearl Harbor attack reveal that a series of events across the islands of Hawaii were all building towards it. The masterminds behind the attack, the Japanese General Yamamoto had been studying US Navy’s practice drills, ocean patrolling and the harbor’s ships schema for a long time. They had used this knowledge of past behavior and patterns to predict that the morning of Dec 7th, 1941 a quiet Sunday would be the best day to strike. They predicted that the entire base would be off guard, with their planes parked and rifles lined up in cases. They were right, and their attack was one of the most successful ones ever made in modern history. The lesson learned was that patterns reveal a lot. Past data is instrumental in determining what the future holds, with a confidence interval.
In the current world scenario where world wars seem a distant possibility, countries are struggling with the scrimmage within their population. It could be anything ranging from planning to overthrow governments, or human trafficking, or just random psychopaths planning their next radicalized move. The question now becomes, can such behavior be detected in advance? The simple answer is Yes and the more complicated one is Almost. One of the recent publications in Cosmo magazine talks about an elite investigator who studies and susses out when exactly an irrational hate comment will transcend into something violent. She collects data on internet users who have exhibited violent behavior in some shape and form. This data is analyzed for patterns to predict which person is just bragging and which one is actively plotting mass killings. Her ability to sense that difference is unsurpassed. But to keep herself safe, she has to work on the computer which makes the job impossibly hard. One of the issues she faces is that there’s a lot to sift through. Hence apart from internet data, she uses her intuition and photographic memory. She reports that she is stretched thin trying to stay on top of it all.
But she doesn’t have to anymore. What we call photographic memory and finding patterns is very much achievable through machine learning. There can be a framework designed to imitate the agent’s profiling behavior.
This will obviously come with risks of its own. Firstly because it’s hard to imitate human intelligence. Secondly, it’s hard to decipher if the data has been intentionally scrambled to say what we want to hear. The solution could be to use the active and passively labeled datasets to extract similar word embeddings from the target website.
To build this framework one might not always need new data. It’s with existing data’s vigorous peregrination that the desired outcome will be achieved. Like former FBI director Bob Mueller says, “where you have partial facts, analysts, agents are always trying to interpret what those facts mean, and extrapolate from them, rather than collecting more”.
Of all the different aspects intelligence analysis is about, the most interlocked with data analysis is perception. Perception is the screen or lens through which we interpret the world around us. To illustrate, here is a picture. It shows a young and an old woman. Once our brain is able to see one, it’s hard to see the other. Even when we recognize both perspectives, it’s hard to go back and forth. But this aligns with natural human behavior.
Hint: The old woman’s nose, mouth, and eye are, respectively, the young woman’s chin, necklace, and ear.
Perception is strongly influenced by our past, education, values etc. This trains our brain to expect certain patterns in our analysis, subconsciously. Henceforth, we develop a mindset through the collection of such patterns, that predisposes us to think in certain ways. And although these might be hard to change, the pitfalls we encounter with data analysis forces us to think beyond what our mind is used to. It is also good training to look at things from the other’s point of view, both from a technical as well as the social aspect. To learn from him again
“There’ll be differences of opinion in just about every intelligence analysis that you make.” — Robert Mueller
Data analysis that’s inclusive of different perceptions can be very instrumental in problem definition and planning. In the case of the investigator, taking into account her style of working and understanding the users she is monitoring are both to be considered when designing the ML algorithm to detect patterns. In the case of national security issues, the perceptions of both nations must be kept in mind when determining the future of a rift. While data and ML can be a powerful tool in predictive crime analysis, one must keep in mind that with great power comes great responsibility. While it can prevent bad it can also prevent the good from happening.
