top of page

Why Every AI Ethics Framework Talks About the Same Things Differently


Imagine four experts in a room. One from Brussels, one from Washington, one from London, one from a tech company. You ask them: What does "fairness" mean when an AI makes decisions about people?


All four nod confidently. All four give completely different answers.


This is AI ethics today. Not disagreement about whether fairness matters. Everyone agrees it matters. The problem is that nobody uses the same words to describe it, measure it, or govern it.



The Same Idea, Five Different Names


The EU AI Act talks about "human oversight". NIST calls it "human-AI teaming". The OECD says "human-centred values". The UK uses "human control". Industry says "human-in-the-loop".


They all mean roughly the same thing: a person should be able to understand, intervene, and override what the AI is doing.


But if you're writing a policy, which term do you use? If you're a developer, which specification do you follow? If you're checking compliance, which box do you tick?


The answer is often: all of them, and none of them coherently.



What This Looks Like in Practice


Picture a team deploying an AI system to prioritise benefit applications. They want to do this responsibly.


They pull up three frameworks. One says they need "explainability". One says "interpretability". One says "transparency".


Are these the same thing? Overlapping? Different?


Two weeks later, after lawyers, academic papers, and a webinar, they still aren't sure. They write a document using all three terms interchangeably. It satisfies nobody but gets signed off on because the project has a deadline.


This isn't a failure of competence. It's a failure of the common language.



The Hidden Cost


Fragmentation costs you time. Every organisation does its own translation work, solving the same problem independently.


It costs you clarity. When policy uses one term, and your technical team uses another, requirements get lost in translation.


And it costs you accountability. When something goes wrong, fragmented terminology makes it harder to pinpoint what failed. Was it the "human oversight", the "human control", or the "human-in-the-loop" that broke down? In a post-incident review, that ambiguity matters.



Why This Cascades


If you can't agree on what "human oversight" means, you can't agree on when it's failed. If you can't agree on failure, you can't agree on who should have escalated the problem. If you can't agree on escalation, you can't hold anyone accountable.


The terminology problem becomes the governance problem, which becomes the accountability problem.



What Would Help


The solution isn't another framework. The world has enough frameworks.


What's missing is a translation layer. Something that maps concepts across frameworks. Something that says: when the EU says this, and NIST says that, and the UK says this other thing, they're pointing at the same operational requirement.


That would let organisations focus on implementation instead of interpretation.


Until someone builds it, every organisation deploying AI responsibly has to do that translation work itself. Some will get it right. Many won't.


And in the gap between intention and implementation, real harm happens to real people.

Over the next 12 weeks, I'm explaining RAI monitoring in plain English - what it is, why it matters, and how it works in practice. All free. Follow our page.



Ā 
Ā 
Ā 

Comments

bottom of page