Investigating Future Security Threats with AI, ML and Data Fusion

AI and ML complemented by sophisticated data fusion and analytics capabilities introduced a new era

The emergence of artificial intelligence (AI) and machine learning (ML) technologies has been transformative for security organizations. With their arrival, organizations are no longer limited to reactively investigating and resolving criminal acts that have already been committed. AI and ML – complemented by sophisticated data fusion and analytics capabilities – have introduced a new era of proactive, predictive threat assessment that can help us identify and mitigate threats before they occur.

The ability to analyze past criminal behavior remains hugely important, of course. What's changed with AI/ML is that the data and intelligence gained via past investigations can now be effectively pivoted and wielded to predict future criminal trends. This is achievable with a newfound ability to harness numerous, disparate data sources – both structured and unstructured – where previously an analyst might struggle to crunch a considerably smaller number of data points via conventional methods.  

With AI/ML and data fusion, we now have the means to define and correlate a myriad of data sources such as population and vehicle registry, financial records, and even images and videos, with automated efficiency and precision. This opens the door to a fuller, multidimensional understanding of threat vectors in the future since we are getting a holistic view from a variety of sources and eliminating 'blind spots.'

AI/ML does not, however, herald a future of Minority Report-like criminal investigation whereby suspects are accused and apprehended before crimes have been committed. Rather, AI/ML aggregates the available intelligence to help law enforcement and security personnel make data-driven decisions based on all the available data rather than simply relying on past experience and hunches. As a result, decisions will ideally be made in a more objective manner that will also assist in operating in a dynamic, fast-changing environment with high levels of uncertainty.

PATTERNS OF BEHAVIOR 

The power to identify trends related to individuals, companies and objects with increasing precision – the core promise of AI and ML in advanced analytics – is well-suited to forecasting criminal activity. Humans are predictable by nature, prone to patterns of behavior, and research demonstrates this. Likewise, crimes don't typically materialize out of thin air. They are largely predicated on the past acts of criminals who have gradually established a criminal comfort zone over time. 

This criminal 'precursor' intelligence can now be effectively distilled from big data, leveraging AI and ML to surface criminal behavior patterns to the attention of analysts. But the AI/ML algorithms must first know what they're looking for. This requires careful attention upfront; ML models must be written with precisely defined data parameters in mind.

To illustrate by way of a law enforcement example, consider an investigation of organized crime. AI/ML algorithms can help analysts identify similar entities, for example, in situations where they need to identify not only the known suspects but also suspects that share similar characteristics. That similarity can manifest in similar patterns of suspicious money transfers, phone calls to the same destinations at the same time and location, and more. The similarity algorithm analyzes the available data, extracts hundreds of features out of it, and surfaces to the analyst entities which might be similar to the known suspects and therefore linked to the same organization or illicit activities. When the similarity algorithm is combined with threat scoring algorithms, the analyst can also prioritize which potential suspects may pose the highest threat. 

Such an investigation would, at minimum, draw from available law enforcement records, such as arrest records and prison records, to account for past criminal activities in a given geography. A fuller understanding would be revealed, however, if investigators incorporated data points distilled from open-source data, business and financial reports, transportation infrastructure and travel patterns, and criminal distribution patterns, to name just a few. This data could be used to help law enforcement agencies to better analyze and prepare for future organized crime trends. 

Another example can be found in the financial domain. Here AI/ML can also be utilized to apply carefully sourced statistical data to look for similarities indicating a potential risk of future financial crimes. This manifests in risk scoring techniques that can help a tax authority investigate tax evasion, assess risks and improve their decision-making process through a data-driven approach, for example, by assessing bank records, financial documents, invoices, etc.

In the case of money laundering investigations, ML models can be trained to spot anomalies among countless financial transactions and activities. ML algorithms are designed to conduct risk scoring, which calculates hundreds of different features that relate to potential entities. The calculated data, together with the ML algorithm scores the risk of the potential entities and surfaces insights to the analyst who can then focus their decision-making process, now based on only relevant data.

THE FUTURE OF SECURITY INVESTIGATIONS 

The ability to synthesize structured and unstructured data has long been a sticking point in investigations. But much progress has been made here and security analysts are no longer limited to analyzing structured data residing in tables and spreadsheets. With advanced data fusion capabilities, analysts are now also able to harness unstructured data including text documents, audio, video and images, for example. Assets like these are often invaluable to include in criminal and homeland security investigations.

With AI/ML and data fusion, analysts have gained the capability to source, tag and organize a deluge of data in a meaningful way. This alone is a major step forward for decision intelligence since the analyst and the decision maker now have a comprehensive intelligence picture and not fragmented data points that might lead them to the wrong decision – and it's just the tip of the iceberg. Today, many organizations are already gearing up to use ML in their analysis. From a recent survey we conducted with over 200 CIOs from 14 countries, we learned that 88% actively use ML or have it on their technology roadmap.

This frees the analyst to deal with the actual analysis of incoming information which is already fused, aggregated, sliced, and diced in a manner that improves dramatically the efficiency of the analyst. As AI/ML capabilities will improve, so too will the analyst be able to investigate not only past events but also point to a possible future and make informed and data-driven decisions.

We also anticipate that the full evolution of AI/ML-guided criminal investigations will invite some significant challenges down the road. AI/ML does not replace the experience of investigators and analysts, they enable them to make data-backed decisions rather than rely on intuition. However, security organizations must ensure that the data models and algorithms they employ are trained on high-quality data sets in order to avoid inadvertent bias – and they should be as transparent as possible, to avoid a situation where analysts rely on 'black boxes' that simply churn out conclusions. 

The ability to predict crime trends has arrived, and it is transformative. AI/ML and data fusion can enable sophisticated statistical analysis leveraging precisely defined parameters across numerous data sources to help inform, augment and prioritize threat assessments.

This is the future of decision intelligence. The time to prepare for it is now.

Author:

Noam Zitzman, Chief of Intelligence Methodology, Cognyte

Noam Zitzman is the Chief of Intelligence Methodology for Cognyte's Analytics Platform. He has over 25 years of experience in several intelligence domains, both tactical and strategic. Noam holds an MA in history of the Middle East and is co-fellow at the Jerusalem Van Leer Institute as part of the Forum for Regional Thinking.

Social Media Profiles:

Noam Zitzman – https://www.linkedin.com/in/noam-zitzman-295511120