When More Intelligence Is Not the Answer

Consider a city transit authority using AI to redesign its bus network.

The system analyzes millions of journeys. It identifies the least-used routes, calculates the potential savings, and recommends eliminating three of them.

The analysis is excellent.

The routes really are expensive. The buses really are often half empty. Removing them really would make the network more efficient.

But one of those routes connects a low-income neighbourhood to the nearest hospital. Another is used mainly by older people who no longer drive. The third is the only practical way for night-shift workers to get home.

The AI has not made a mistake.

It has answered the question it was given:

How do we make the bus network more efficient?

What it has not asked is:

Efficient for whom?

That is the moment when more intelligence is not the answer.

That is the moment when a system may need a point of view from outside itself.

A Human Counterpoint.

When the analysis is complete, but the judgment is not

The transit example is not unusual. Variations of it will soon appear across almost every part of our lives.

Consider a company using AI to shortlist its future leaders.

The system examines the characteristics of executives who have succeeded in the past, then identifies the current employees who most closely resemble them.

The shortlist looks perfectly sensible.

But who asks why every “high-potential” candidate looks like someone the company has already promoted?

Is the system identifying the best leaders for the future—or reproducing the kind of leader the organization already knows how to recognize?

Or consider an AI tutor that improves a student’s test scores.

The system notices every hesitation. It offers a hint at exactly the right moment. The student completes more lessons and earns higher marks.

But is the student becoming more capable—or simply more dependent on never being left alone with a difficult problem?

A Human Counterpoint is not simply a human in the loop

We already have humans in the loop.

They check outputs. They correct errors. They approve decisions. That work is important and, in many circumstances, essential.

But a Human Counterpoint does something different.

Suppose an AI system recommends rejecting a job candidate.

A human in the loop might check that the résumé was parsed correctly, that the correct criteria were applied, and that the rules were followed.

A Human Counterpoint might ask:

Why does every person this system considers promising look like the people we hired before?

The human in the loop asks:

Did the system do this correctly?

The Human Counterpoint may ask:

Should we be doing this at all?

That is a fundamentally different role.

The first person works largely within the objective already given to the system. The second is permitted to question the objective itself.

Why not simply ask the AI to disagree with itself?

We should.

AI systems can generate objections, alternative interpretations, counterarguments, and contrarian viewpoints. Asking a model to challenge its own conclusions is often useful.

But the same system is still generating the question, the answer, the objection, and the reply.

It can simulate another point of view.

It does not arrive with one.

A Human Counterpoint brings a perspective that existed before the assignment. That perspective has been shaped by work, experience, loyalties, mistakes, disappointments, commitments, and consequences.

This does not make human judgment perfect.

Humans are biased too.

The point is not to find a neutral person. There is no such person.

The point is to bring in someone whose assumptions did not come from the same brief, the same dataset, or the same institutional incentives.

Someone who is free to say:

The answer is reasonable. The objective is not.

The better AI becomes, the harder it may be to question

There is a paradox here.

As AI systems become more capable and reliable, we may become less likely to challenge them.

Think about your GPS.

The first few times you use it, you check every turn. You compare the suggested route with what you already know. You remain alert to the possibility that it may be wrong.

Then it gets you to your destination correctly a hundred times.

So you stop checking.

One day, it tells you to turn down a road you would never have chosen yourself.

And you do it.

Not because you examined the route and decided it was best. You do it because the system has earned your trust.

Now replace the wrong road with a hiring decision.

Or a workplace policy.

Or an investment strategy.

Or a recommendation affecting a million customers.

The principle is similar.

The consequences are not.

The greatest danger may not be that AI systems routinely produce obviously bad answers. Those are relatively easy to catch.

The more difficult problem arises when they produce answers that are coherent, plausible, well-supported, and mostly right.

Those are the answers we may accept without asking whether the problem was framed correctly in the first place.

Humans hired for how they see

To date, much of the discussion about agents hiring humans has focused on physical or procedural work.

A human might be asked to visit a location, sign a document, make a telephone call, inspect an object, or complete some other task the agent cannot perform directly.

But there is another possibility.

AI agents may eventually seek out particular humans not for their hands, but for their point of view.

Not someone to click a button.

Not someone to sign a form.

A person paid because of how they see.

One Human Counterpoint might understand workers and notice what an efficiency initiative would mean on the shop floor.

Another might recognize what a decision would mean for children, patients, caregivers, or small-business owners.

Another might challenge the environmental assumptions beneath a proposed strategy.

Another might understand how a technically rational decision will be interpreted within a particular culture or community.

These people would not be interchangeable.

That is the point.

There is no generic human perspective

The future does not need a generic human point of view.

There is no single human point of view.

What matters is the particular perspective being brought into the room.

My own instinct has always been to look from the outside in.

From the customer’s side of the counter.

From the ordinary person’s side of the expert conversation.

I tend to ask:

• Who is being talked past?
• Who has disappeared from this decision?
• What looks sensible at the centre but very different from the edge?
• Does this increase or diminish human agency?
• What will this mean to the people who must live with the consequences?

That would not make me the right Human Counterpoint for every question.

It might make me useful for some.

And that specificity matters.

A Human Counterpoint should not claim to represent humanity. The value lies in offering a particular, independently formed way of seeing—and making that perspective clear enough that an agent or organization can decide when it is relevant.

The next scarcity

AI may soon give us more intelligent answers than we know what to do with.

It can already generate options, analyses, scenarios, and recommendations at a scale no human team could match.

The next scarcity may not be intelligence.

It may be independent human judgment.

The judgment to decide which answers deserve attention.

Which deserve money.

Which deserve reputation and commitment.

Which misunderstand the people affected.

And which should never be acted on at all.

Sometimes, before asking AI for a better answer, we may need a human willing to ask whether we are solving the right problem.