AI Is Not Neutral

We tend to ask the wrong questions about artificial intelligence.

Is it biased?

Is it fair?

Can it be trusted?

These are not irrelevant questions.

But they are not the decisive ones.

Because they assume something that is rarely examined:

that the systems in which AI operates are already neutral.

They are not.

 From Tools to Decision Infrastructures

Artificial intelligence is still often described as a tool. This framing is convenient.

It suggests control.

It suggests that we can choose when and how to use it.

But it is increasingly misleading.

AI is not simply supporting decisions.

It is becoming an infrastructure of decision-making.

It shapes:

• who is shortlisted
• who is prioritised
• who is visible
• who is excluded

Not occasionally.

Systematically. At scale.

And infrastructures do something tools do not:

they structure outcomes without announcing themselves.

 Bias Is Not the Problem. Continuity Is.

Many organisations implement oversight structurally.

They assign reviewers, introduce approval steps or create escalation processes.

But oversight is not the presence of a person.

A person can be formally present and practically powerless.

They may lack:

• understanding of the system’s limitations
• access to relevant data context
• time to engage critically
• authority to override decisions

Or they may simply assume that the system is more objective than they are.

This is the central weakness of many governance models:

Responsibility is assigned.

Judgement is not enabled.

 

 Automation Bias Is Not Just Cognitive

 

Much of the current debate focuses on bias.

Datasets.

Algorithms.

Outputs.

But in many cases, what we call “bias” is not a failure.

It is a continuation.

AI systems learn from past decisions.

Past decisions reflect institutional patterns.

Institutional patterns embed power, hierarchy and culture.

So the system performs exactly as expected.

The issue is not that AI gets things wrong.

It is that it may get them consistently right — according to a flawed institucional logic.

 

 The Comfort of Technical Questions

There is a reason why the debate gravitates towards technical fixes.

They are measurable.

They are actionable.

They create the impression of progress.

But they also shift responsibility.

If the problem is technical,

the solution is technical.

If the problem is institutional,

the solution is far more uncomfortable.

Because it requires questioning not the system—

but the organisation deploying it.

 

 Opacity as a Governance Condition

AI systems are often described as opaque because they are complex.

But opacity is not only a technical feature.

It is also a governance condition.

When decisions cannot be traced, explained, challenged, accountability does not disappear.

It becomes abstract.

And abstract accountability rarely protects anyone.

 Inclusion Beyond Access

Inclusion has traditionally been framed around access.

Access to education.

Access to employment.

Access to opportunity.

This remains essential.

But it is no longer sufficient.

Because access can coexist with:

• invisible filtering
• automated prioritisation
• decision layers that remain out of reach

Inclusion is no longer only about entering the system.

It is about how the system decides once you are inside.

 The Limits of Regulation

Regulation matters.

It introduces categories, obligations and safeguards.

But it operates under an assumption:

that risks can be anticipated, classified and managed.

This is only partially true.

Because the most relevant failures are rarely violations.

They are normalised practices.

Small exclusions.

Accumulated disadvantages.

Decisions that no one questions because they appear consistent.

No regulation can fully capture that.

 What Cannot Be Delegated

The deeper question is not whether AI should be regulated.

It is whether certain forms of judgement should be delegated at all. Not because machines are incapable.

But because delegation transforms responsibility.

When decisions are mediated by systems, responsibility becomes distributed, accountability becomes harder to locate and legitimacy becomes more fragile.

What Is Ultimately at Stake

The real issue is not whether AI is neutral.

It is whether the systems in which it operates are capable of questioning themselves.

Because AI does not introduce bias into organisations.

It operationalises what is already there.

And once operationalised, those patterns become scalable, consistent and significantly harder to contest

At that point, the problem is no longer technical.

It is institutional.