Thinking With Machines: When AI Can See the Pattern, Who Makes the Decision?

Artificial intelligence is becoming very good at finding patterns.

It can process enormous amounts of information, identify risks, forecast possible outcomes, compare scenarios and surface relationships that a human team might miss.

For Project Management Offices, that creates an obvious opportunity.

But Thinking With Machines: Building AI-Enabled PMOs for the Age of Intelligent Project Leadership by Danilo Arba made me think about a less obvious problem.

When a machine becomes better at seeing the pattern, what happens to human judgement?

That question sits at the heart of the book for me.

Beyond the traditional PMO

The traditional PMO has often been associated with reporting, monitoring, governance, resource management, schedules, budgets and risk.

Arba argues for something more ambitious.

He looks at the possibility of an AI-enabled PMO that can move towards intelligence-driven governance. Machine Learning, Natural Language Processing, Generative AI, Robotic Process Automation, Agentic AI and PMO Copilots are not presented merely as fashionable technologies. They are connected to actual project and programme management problems.

That practical orientation gives the book its relevance.

AI can help identify patterns.

It can support forecasting.

It can help optimise resources.

It can flag risks.

It can process information faster than a human team reasonably can.

But none of that answers the most important question.

Who decides what the information means?

Better information does not automatically create better

decisions

This distinction is easy to miss.

We sometimes speak about AI-powered decision-making as though better information naturally leads to better decisions.

Human organisations don't work that way.

A leader can have excellent information and still make a poor decision.

Perhaps the incentives are wrong.

Perhaps organisational politics gets involved.

Perhaps there is pressure from above.

Perhaps fear influences the decision.

Perhaps the technically correct option creates consequences that the data cannot adequately capture.

Or perhaps the machine has identified the pattern correctly but misunderstood the context.

This is where Arba's emphasis on human judgement becomes important.

AI can assist decision-making.

It does not automatically assume responsibility for the consequences.

And that distinction becomes increasingly important as organisations begin depending on AI rather than merely experimenting with it.

The question of delegation

One of the ideas I found myself returning to while reading the book was not:

What can we automate?

It was:

What should humans refuse to delegate?

That is a different question.

Automation naturally encourages us to think in terms of capability.

If a machine can perform a task faster, why should a person continue doing it?

Fair question.

But leadership is not simply a collection of tasks waiting to be automated.

Some decisions involve values.

Some involve people.

Some involve uncertainty.

Some require understanding what is not present in the data.

And some require somebody to stand behind the decision afterwards.

That last part matters.

A machine can produce a recommendation.

A human being still has to live with the consequences.

The efficiency trap

There is another assumption in the AI conversation that deserves examination.

We often assume that automation will automatically give people more time for strategic thinking.

Sometimes it will.

But organisations have an unusual habit.

Whenever technology creates efficiency, organisations often find new ways to consume that efficiency.

A faster reporting system can produce more reports.

More reports can generate more meetings.

More meetings can create more decisions.

More decisions can create more work.

And eventually, the organisation may become extremely efficient at processing information without becoming noticeably better at thinking.

That tension is particularly relevant to the idea of an AI-enabled PMO.

Perhaps the real measure of intelligent automation should not be how much additional work a system allows us to produce.

Perhaps it should also be how much unnecessary work it allows us to eliminate.

That changes the question from:

What can AI do for us?

to:

What should we stop doing because AI has exposed its lack of value?

Thinking with machines

I like the phrase Thinking With Machines because it avoids two familiar extremes.

AI does not have to replace human thinking.

Nor do humans have to pretend that machines are simply faster versions of themselves.

The interesting possibility lies somewhere between those positions.

Machines are good at certain forms of scale, pattern recognition and information processing.

Humans bring context, judgement, responsibility, ethics and an understanding of consequences.

The challenge is not deciding which side wins.

The challenge is understanding where each belongs.

That becomes particularly important in project leadership because projects rarely fail or succeed on information alone.

People make decisions.

People interpret risk.

People negotiate competing priorities.

People respond to unexpected circumstances.

People decide what matters when everything cannot matter equally.

Technology can change how much information reaches those people.

It cannot remove the responsibility of deciding what to do with it.

Who is this book for?

Thinking With Machines will be particularly relevant to project managers, PMO leaders,

programme professionals, transformation leaders and executives trying to understand how artificial intelligence can be integrated into organisational decision-making.

It is also useful for anyone who wants to move beyond the simplistic question of whether AI will replace people.

The more interesting question is how organisations will divide responsibility between humans and increasingly capable machines.

That conversation has only started.

My takeaway

The book left me with one distinction.

Knowing what is likely to happen is not the same as knowing what should happen.

AI may become increasingly good at the first.

Leadership will still be tested by the second.

And perhaps the future PMO will not be defined by how much it automates.

Perhaps it will be defined by how intelligently it decides what should remain human.