Experts are programming your AI, but who is reprogramming your employees for HITL?
In the credit union industry, we love a good “people-first” narrative. It gives the board and the marketing department a warm, fuzzy feeling to say that our new AI strategy has a “human in the loop” (HITL). We like to imagine this human as Sarah Connor—the gritty hero protecting our members from a cold, unfeeling machine that could go full mimetic polyalloy at the drop of a semicolon.
But what if we have it exactly backward?
In reality, a well-tuned AI is your reprogrammed T-800—a disciplined, unwavering guardian of the rules. It doesn’t get bored, it doesn’t have a “bad Friday,” and it doesn’t hold subconscious bias against a member’s zip code.
The real threat? The “mimetic” human who hasn’t been reprogrammed for this new era. Without a fundamental shift in how we train our staff, HITL isn’t a safety net—it’s a significant potential point of failure just waiting to fail.
The Danger of the “Autopilot” Mindset
When we don’t “reprogram” our people, we can fall into one of two deadly traps that the aviation and medical industries learned the hard way:
The De-skilling Trap (Air France Flight 447): In 2009, when a technical glitch disconnected the autopilot, the pilots—who had spent years as the humans in the loop of a near-perfect system—had forgotten how to manually fly the plane. They stalled it into the Atlantic. Oops.
The CU Risk: If your underwriters only “review” AI decisions, they lose the core skill of manual analysis. When the system eventually hiccups, will they even know how to take the wheel?
The Blind Trust Trap (The Therac-25 Tragedy): This radiation machine accidentally delivered massive overdoses. Operators ignored the patients’ literal screams of pain because the computer screen said everything was “normal.” They trusted the machine more than the screeching human in front of them. Double oops.
The CU Risk: There’s a term for this: automation bias. If the AI flags a loyal, 20-year member as “fraudulent,” will your staff have the guts (and the training) to override the machine, or will they “rubber-stamp” a member’s nightmare?
The “Emotional Dregs” Paradox: The Burnout Trap
Management often talks about AI taking on mundane tasks as if it’s doing the employee a big favor. But by automating the “easy stuff,” you’re removing the mental breaks that keep your staff sane.
In any job, those routine, low-stakes tasks are the “down-shifts” that allow for recovery. When you give the easy wins to the AI, the humans are left with a 100% concentrated dose of high-conflict, high-complexity stress. You aren’t just changing their workflow; you are increasing their cognitive load to a level that leads to rapid burnout. If you ask a human to spend eight hours a day solving only the problems a machine couldn’t, you aren’t managing a “loop”—you’re running a pressure cooker.
The “Reprogramming” Reality Check
If you’re betting your reputation on HITL, you need to realize that you aren’t just “plugging in” a person. You are asking a worker who was trained for execution to suddenly switch to audit and supervision.
That requires a new set of directives:
Adversarial Training: Train your “human in the loop” to be a skeptic who actively looks for where the machine is wrong.
Value the Specialist: You cannot pay “execution” wages for “adversarial auditing.” If the job is now 100% complex decision-making, the compensation must reflect that high-stakes status. (And if you don’t pay them like specialists, I bet I know a couple dozen fintechs who will.)
Engineered Recovery: Management must build “low stakes” downtime back into the day to replace the mental breaks that AI took away.
Without this, you don’t have a safety net. You have a mimetic polyalloy variable that will melt like a T-1000 in a vat of molten steel the moment the heat is on.
The future is not set. There is no hope for your AI strategy but what we make for our people.