No Human In The Loop
Let’s kick off 2026 with a personal reflection.
I have long been bothered by the adage "human in the loop".
It sounds safe and ethically sound, but in reality it’s AI centric speak that serves as a smokescreen for humans.
After reflecting on the potential impact of HITL, I came to the conclusion that we need to revise this concept.
From the machine's perspective, “human in the loop” is actually a solid concept.
It aligns perfectly with the old database law: Garbage In, Garbage Out.
Because AI is supposed to thrive on quality, the system needs training, correction, and continuous improvement. Thus it needs humans in its loop to achieve this. Because we posted so much garbage on the internet in the last 3 decades and because we fail to provide good prompts and guardrails, having a limited understanding of the algorithm, we get hallucinations in return.
In this sense, the "human in the loop” is useful for AI: a human that feeds and corrects the machine to filter the existing garbage out.
However, for our operational brain, for our management, and for our legal liability, "human in the loop" is dangerously vague.
It suggests that accountability floats somewhere within a circular process. But responsibility doesn’t float, it sticks.
AI is not legally liable. Not under civil law, not under criminal law. Only a human, an owner of a tech company, a hacker, a legislator, a manager or an individual user of a an AI system can be held responsible in court of law. Not the AI itself, it has no legal persona.
HITL says nothing about when the human should intervene in the operational value chain. Vague oversight leads to “Garbage Out” and real reputational damage. It leads to rubber stamping sloppy output and risks to even be weaponized by malicious actors.
If humans are merely curators of what AI initiates and creates, we stagnate as civilisation. We should not outsource the "first spark" of creation or the "last mile" of control.
We must break this circle of HITL and replace it with a linear chain of ownership.
The adage I prefer is: Human In, Human Out.
This is the base of our responsibility architecture:
“Human In”
The human always initiates. We determine the context, the goal, and the ethical framework. We are the architects. We ensure no garbage goes in: not in content, not in prompts, not in guardrails. We need to master the skills to originate the process and maintain the knowledge to manage it. We can let the machine do the heavy lifting. Generating, analyzing, and optimizing in the middle ground. As long as we understand what happens inside the engine and are accountable for any garbage that goes in or comes out.
“Human Out”
The human always validates. No output leaves the organization without an explicit human signature.
The human starts, AI works, the human decides. That’s real compliance without evasion.
HITL is fast food for organisational optimisation. We risk getting blindfolded by promises of speed optimisation and cost efficiency. The first principle of economics also applies to AI innovation: there is no such thing as a free lunch. A focus on speed and cost is paid in a flood of AI slop, in reputation damage and ultimately in losing control. The road to faster and cheaper leads us towards getting fundamentally stuck.
Let's instead implement AI to get smarter, to make better decisions, to enhance our possibilities, to create at a much bigger scale, to multiply the effectiveness of our endeavors.
Let’s stop running in loops like lab mice, let’s not mistake speed for progress. Let’s go for more effectiveness rather than more effiency.
Human In, Human Out. A clear resolution for the new year.
© Belgaimage
This blog is written by Patrick Lacroix in a personal capacity. AI tools are used for research, structuring, drafting and language support. All content is selected, verified, and edited by the author, who retains full editorial responsibility.