Corporate Comfort Food

Why "Human in the loop" is NOT enough. Real AI governance needs more substance!

Human in the Loop - HITL

Undoubtedly, the most reassuring FOUR words of the year.

It sounds like someone has our backs.

It is like a corporate version of mac and cheese - warm, familiar, reassuring.

To me, “human in the loop” sounds more like comfort food than anything else.

What does HITL really mean, and where- or when- is the human even being looped in?

Does it mean a recruiter pressing approve on a shortlist of 1000 resumes without any idea what AI is doing behind the scenes?

That’s just button-pushing.

And button-pushing isn’t leadership. It’s like being asked to lay out the plates at the party and not getting to decide what’s on the menu.

If we want real oversight, we need to step up as the party planners.

Explainability Comes First

Governance starts with “E-X-P-L-A-I-N-A-B-I-L-I-T-Y.”

And yes, that is a real word.

If you don’t know how a system works, you can’t govern it.

I learned this the hard way when I built a simple chatbot.

At first, I asked my team to test it like any other tool. There was confusion.

So I decided to host an “Ask Me Anything” session, and the first question was “How does this chatbot work?”

I showed them how it worked under the hood - where the data came from, how the flows were designed, and why certain answers appeared.

Now suddenly, they were asking sharper questions. “Maybe it didn’t answer because this document is missing.” “Maybe we should add this FAQ.”

They weren’t just testers anymore - they were co-designers.

That’s the power of Explainability. It turns humans from button-pushers into governors.

At the company level, “opening the hood” means HR, Compliance, and managers can make sense of what the AI is doing. At the policy level, regulators are beginning to demand it. But we don’t have to wait for laws to tell us: without explainability, governance is just theatre.

It All Depends on What’s at Stake

I came across a governance framework by Tey Bannerman that really got me thinking. In it, he asks: “What’s at stake if this goes wrong?”. That question shaped my thinking in a big way.

• Low stakes: If an AI helps a consultant polish email drafts, nothing much is on the line.

• High stakes: If AI decides who gets hired and promoted? People’s lives are directly affected.

The higher the stakes, the stronger the governance required.

So before asking, “Who’s in the loop?” the real question is: “What’s at stake if this goes wrong?”

Five Ways Governance Should Work

Let’s keep it simple and relatable. AI Governance manifests in five ways:

1. The Express Lane

Think of the airport. Most travelers wait in long immigration lines. But programs like Global Entry or NEXUS (for Canadian travelers) let pre-verified people zip through the express lane.

It looks fast, but it’s not reckless. There’s a background check up front that makes the speed possible.

AI governance works the same way.

• For low-stakes, high-volume tasks like ticket routing or auto-scheduling, you can move fast if the groundwork is done.

• Governance here isn’t about checking every output. It’s about checks up front, so when the system moves quickly, you can trust it.

The danger is when companies skip the groundwork and create “express lanes” with no upfront checks. That’s when speed turns into risk.

2. The Filter Check (Accuracy & Fairness)

Like checking what your air filter caught before you breathe it in.

When accuracy and fairness are critical- candidate rankings, performance scoring, medical scans - transparency is non-negotiable.

Without it, trust collapses into hope.

If HR audits a shortlist of candidates and the system is a black box, all they see are names. But if it’s explainable, they can see the weighting of skills, the criteria used, the patterns over time. Now they can challenge, adjust, and actually govern.

The Filter Check isn’t about babysitting every decision. It’s about auditing the patterns with visibility into the process.