Artificial Intelligence (AI) for Reliable Interdiction
How AI and predictive analytics can be a game-changer for Law Enforcement
Of the millions of people, organizations and activities comprising our world today, which ones represent potential threats that merit interdiction? Selecting the wrong ones wastes precious resources and impedes lawful activities. But the consequences of overlooking real threats is even worse. The public expects to be kept 100% safe against such threats, but has little tolerance for intrusive intervention into their own lawful activities. Dealing with this “Catch 22” is a key challenge for all enforcement agencies – whether law enforcement, border enforcement, child protection, taxation, or others charged with protecting the public.
Some enforcement agencies rely on experience and trade craft to find threats. But experts are scarce and the number of potential targets is very large. Another common approach is random targeting. Random targeting of individuals and organizations for interdiction may have some benefit as a deterrent, but it is a weak deterrent given the extremely low likelihood of being targeted. Random targeting also has a very low probability of success and is demoralizing for officers assigned to such activities with little to show for it. The bottom line is that there simply are not enough skilled people to effectively and efficiently find threats. While perfection is unaffordable, improvement is a necessity.
As a result, agencies are increasingly looking at ways of using systems for selecting targets that have a higher probability of being a genuine threat. Typically, profiling is used to identify individuals and organizations that exhibit characteristics or behavior similar to known threats. However profiling falls short in several key areas:
- Profiling is often perceived as discriminatory and can spark significant public backlash.
- Most profiling can be defeated by criminals who simply change their modus operandi to avoid matching these profiles.
- Since profiling depends on previous experience to identify threats, it is unable to detect new patterns until after the fact. It is analogous to driving a car by looking only in the rear view mirror!
- Lastly, pattern-based profiling generates far too many false alerts – often as many as 90% of the targets turn out to be false.
Much work has been done attempting to improve the accuracy of pattern-based targeting by creating more sophisticated rules that identify potential risks based on a broader set of characteristics. And as new patterns are discovered, new rules are created and existing rules are further refined to detect future occurrences of those patterns. But the lag between trend detection and rule implementation means that the rules are always out of date because they are based on yesterday’s data. Furthermore, the inevitable result is a morass of complex rules that are unmanageable and often self-contradictory.
Artificial intelligence (AI) that leverages the power of Predictive Analytics represents a quantum leap in targeting technology. Rather than depending on prior experience, Predictive Analytics lets the data speak for itself by automatically identifying new patterns through statistical analysis that may not be obvious to even the most experienced officer. Importantly, predictive analytics can identify anomalies in the data that may indicate a threat even though it doesn’t match any known pattern – so no more driving with the rear view mirror.
Typically, predictive analytics is used to “mine” historical data to identify new patterns and insights in support of strategic decisions. For example, with access to the right data it can automatically identify new forms of fraud, contraband transport, or tax evasion that allow enforcement agencies to realign their resources or develop countermeasures. But data mining doesn’t help with individual targeting decisions. The problem is that data mining takes time to digest the data, conduct the analysis, filter out the noise and extract the real insights. It cannot produce actionable intelligence for immediate interdictions. By the time the insights are available, the opportunity for interdiction is long gone.
That limitation vanishes with the advent of real-time predictive analytics tools. Real-time predictive analytics examines each piece of data as it arrives and generates insights within seconds. This provides enforcement officers with the actionable intelligence they need to make interdiction decisions with far greater accuracy than is possible with pure pattern matching approaches.
Real-time predictive analytics is a real game changer for targeting, but AI also provides the opportunity to leverage the power of Machine Learning. Machine learning uses the outcomes from each interdiction to automatically refine the predictive analytics algorithms. The result is an automated feedback loop that provides continuous accuracy improvement.
The combination of these two AI tools, real-time predictive analytics and machine learning, will completely transform the way in which interdiction decisions are made. AI enabled interdictions will be more effective, more efficient, and increasingly more reliable than pattern or intuitive approaches. We can never replace the need for more enforcement officers, but this approach to interdiction will go a long way to better use available resources to offset today’s threats.
John Kendall is the Director of the Border and National Security Program for the global public sector at Unisys.