AI – It’s Time to Get a Little Loopy

At both WCO Tech (Hanoi) and IBMATA (Prague), many of the speakers and panelists talked or answered questions about the role of artificial intelligence (AI) in frontline operations.

“When will AI replace customs? inspectors?” seemed to be the most pressing question as a result of an unprecedented confluence of world events:

■ Politicians and regulators mandating? 100% cargo inspection and individual identification rates.

■ Growth of illicit cargo and tariff avoidance schemes.

■ Masses of human migration from trafficking and / or refugee crises.

■ Paucity of qualified applicants.

Interestingly, there was wide disparity among the educated answers. The most optimistic opinion was that AI is already here and can serve in some meaningful ways – examples include risk analysis and imagery interpretation. By contrast, several speakers reminded the audience that fully semantic AI, like ChatGPT’s natural language processor or Tesla’s full self-driving neural network video processor (aka v12), are many years away from replacing front line agents.

I am certainly not qualified to even remotely guess “when” semantic AI will happen. Rather, I want to focus on the “what and how” to describe it. To be most efficient in conversations between buyers (government agencies) and sellers (equipment manufactures and service providers), I suggest we adopt two specific linguistic conventions.

Before I reveal them, it is important that we all share a common understanding of the role that humans play when interacting with technology.

The phrase “feedback loop” is used in a large variety of disciplines. In biology, animals, including us humans, have feedback loops that keep heart rates, breathing, and hormones at more or less normal levels – what is referred to as homeostasis. In astronomy, feedback loops are used to constantly monitor atmospheric turbulence to provide instantaneous corrections to adjustable optics. Electrical circuits are designed with feedback loops to keep equipment operating within normal ranges. In business, analyzing customer buying habits and social media comments provide feedback to improve existing products or drive new development. In finance, fund managers use feedback loops to keep their portfolios appropriately balanced.

Some feedback loops necessitate the presence of a human operator. Riding a bicycle requires one to constantly judge the balance of the bike so as not to fall over. When the bike leans left, the operator leans right. When first learning this skill, our feedback loops aren’t mature enough, resulting in jerky movements, over correcting, and the occasional collision with terra firma (or worse). Driving a car, piloting an airplane, performing surgery are all examples of activities that today mandate the presence of a human operator as part of the feedback loop.

This is what we call human-in-the-loop (HITL) operations – the first of the two terms I’m introducing here.

In the border security domain, manual cargo inspections and imagery analysis are a form of HITL operations.

You are likely aware that nearly all modern commercial airplanes have autopilots that relieve the pilot of HITL-style flying. The technology of flight management systems (FMS) has reached such a level of sophistication that many models can taxi, takeoff, cruise, land, and even shutdown with the push of just a single button. This is definitely the case for several models of unmanned aircraft (aka drones) like Northrop Grumman’s Global Hawk. For vehicles, there are many consumer-produced videos online demonstrating how a Tesla with FSD v12 software can drive point-to-point without any intervention from the human operator.

In both cases, regulators and the public still require a human to monitor the performance of the aircraft or vehicle and take over control in the event of an anomalous situation. As an instrument rated, commercial pilot myself, I often describe to non-pilots that my role onboard is not meant to handle the routine but rather to handle emergencies since there hasn’t yet been invented the autopilot that can autonomously handle a “Sully” Sullenberger’s “Miracle on the Hudson” situation (ditching a fully loaded commercial airplane in New York’s Hudson River after a dual engine flameout without a single loss of life).

When the feedback loop is overloaded or incapable of autonomous response, the human must intervene.

This is what we call human-on-the-loop (HOTL) operations – the second new distinction.

As technology gets more mature, more and more activities, including border security operations, have or will transition from human-in-the-loop (HITL) to human-on-the-loop (HOTL).?

HOTL is often far more efficient since the humans can now perform other, more intellectually noble tasks. Imagine a future where instead of two pilots flying onboard a single jetliner, a single HOTL pilot at a ground control station can monitor the flights of two or more aircraft. This is the dream of nearly every airline CEO as it will significantly reduce labor costs.

This is the exact situation we’re facing in border security operations. Can we convert some highly HITL tasks to a more operationally efficient HOTL environment? I submit that examples can already been found. By incorporating S2 Global’s CertScan software with HOTL features, the Port of Authority in Puerto Rico implemented a 100% NII inspection program that increased Customs compliance and virtually eliminated smuggling through the Port of San Juan.

I submit then, instead of asking manufacturers the question posed above, “When will your AI replace my customs inspectors?” you should be asking: “How can your AI solution help me transition from HITL to HOTL?”

To reinforce this point, let me cite two quite absurd examples of failing to adhere to the HOTL principle.?

The first is merely embarrassing. It involves the case of Steven A. Schwartz, Esquire, who, in May 2023, used ChatGPT in a personal injury lawsuit on behalf of his client but failed to HOTL-check the output which included several non-existent court cases as precedent. Contrast that with the financially devastating “flash crashes” of stock markets due to HOTLless trading algorithms running wild.

As the title of this article recommends, stay “loopy” my friends – HITL or HOTL as you see fit.

Jeff Goldfinger wrote this article, originally published in Border Management Today, Issue 011.