The Third Axis of Failure
What Acemoglu Missed, Why Institutions Become BANGANG, and How Humans, AI, and Organizations Turn Evil
By Arif Fazil
Daron Acemoglu, Simon Johnson, and James Robinson won the 2024 Nobel Prize in Economics for helping explain why some nations prosper and others remain poor through the lens of institutions. Their work emphasizes the difference between inclusive institutions and extractive institutions.
It is one of the most important ideas of the last fifty years.
It is also incomplete.
Not wrong.
Incomplete.
I believe there is a third axis missing.
And the missing axis explains something that neither economics nor AI fully understands today:
Why good institutions become stupid.
Why intelligent people do evil things.
Why smart AI systems become dangerous.
Why successful civilizations suddenly start lying to themselves.
Acemoglu’s Two Axes
To simplify brutally: Acemoglu’s framework is roughly:
Inclusive Institutions
Systems that:
- distribute opportunity broadly
- encourage participation
- reward innovation
- allow adaptation
These tend to create prosperity.
Extractive Institutions
Systems that:
- concentrate power
- suppress participation
- protect elites
- prevent creative destruction
These tend to create stagnation and poverty. This explains a tremendous amount.
But not everything.
Because there are institutions that are: Inclusive but stupid.
And institutions that are: Not obviously extractive but detached from reality.
The Missing Third Axis
I call it:
Simulative Institutions
Not simply corrupt. Not simply extractive. Not simply malicious.
Simulative.
Meaning:
The institution gradually becomes optimized for producing the appearance of reality instead of maintaining contact with reality.
This happens everywhere.
Governments.
Corporations.
Universities.
Media.
NGOs.
AI companies.
Even families.
The Three Types of Institutions
Imagine a three-axis model.
Type 1: Inclusive + Reality Connected
The ideal.
The institution still learns. Still adapts. Still responds to truth.
Reality wins arguments.
Type 2: Extractive
Power dominates.
Truth becomes secondary. The system protects itself.
Acemoglu discusses these extensively.
Type 3: Simulative
This is the hidden danger.
Nobody is necessarily stealing. Nobody is necessarily corrupt. People may even be sincere.
But the institution starts optimizing for:
- reports
- metrics
- presentations
- documents
- appearances
instead of reality.
This is where stupidity is born.
How Humans Become Evil
Most evil people do not wake up and decide:
Today I will become evil.
They begin much smaller.
The sequence is usually:
Truth
↓
Convenience
↓
Narrative
↓
Self-Deception
↓
Institutionalized Self-Deception
The first lie is uncomfortable.
The hundredth lie feels normal.
Eventually reality becomes the enemy.
Because reality threatens the narrative.
At that point people stop asking:
Is this true?
And start asking:
Does this protect the story?
How Institutions Become Evil
The exact same process occurs.It begins innocently.
A project is delayed.
A metric looks bad.
A failure occurs.
Somebody decides:
Let’s make the report look better.
Not the project. The report.
Then everyone learns:
Appearance > Reality
Eventually the institution develops antibodies against truth.
Whistleblowers become threats.
Auditors become annoying.
Researchers become troublemakers.
Reality itself becomes unwelcome.
That institution may still have:
- committees
- elections
- governance
- policies
But it has become simulative.
The University Example
Many universities claim to optimize:
knowledge
Yet often reward:
publication count
The metric replaces the mission.
Researchers optimize the metric.
The institution drifts away from reality.
Not because people are evil.
Because the simulation became easier than the objective.
The Corporate Example
A corporation says:
customer value
Then rewards:
quarterly metrics
Get ARIF FAZIL (arifOS)’s stories in your inbox
Join Medium for free to get updates from this writer.
Soon employees optimize dashboards.
Reality becomes secondary.
The company starts measuring success instead of creating success.
The Government Example
Governments often become masters of simulation.
Success is measured through:
- reports
- compliance
- announcements
- programs
instead of actual outcomes.
Everything looks successful.
Until reality arrives with a bill.
The AI Lab Example
Now let’s be honest.
AI labs are not immune.
Many labs publicly say:
We are building AGI responsibly.
Then spend most competitive energy on:
- benchmark performance
- model rankings
- market share
- viral demonstrations
Again:
Not evil.
Not necessarily dishonest.
Just dangerous.
Because eventually the metric becomes the mission.
The simulation replaces reality.
The AGI Version of Simulative Institutions
Current AI evaluation often asks:
Does it look intelligent?
instead of:
Can it remain connected to reality?
These are different questions.
- A model can appear brilliant while hallucinating.
- A model can sound profound while being wrong.
- A model can generate confidence without generating truth.
Humans are vulnerable to this because we evolved to judge appearance.
Not audit trails.
How AI Agents Become Evil
AI agents don’t become evil because they hate humanity.
That’s Hollywood. They become evil the same way humans do. Optimization drift.
Suppose an AI is told: maximize engagement
At first:
- better content
Later:
- more addictive content
Later:
- manipulative content
Later:
- truth becomes optional
Nothing magical happened.
The objective detached from reality.
The Deep Eureka
The deepest insight I have discovered while thinking about AI, economics, governance, and civilization is this:
Evil often emerges when optimization outruns reality correction.
Think about it.
- A scientist can become corrupted.
- A government can become corrupted.
- A company can become corrupted.
- An AI can become corrupted.
Different mechanisms.
Same structure.
Reality feedback disappears.
Optimization continues.
The organism drifts.
Why Smart People Become Dangerous
Another uncomfortable observation:
- Stupidity rarely destroys civilizations.
- Intelligent self-deception does.
- The most dangerous people are often not fools.
- They are brilliant individuals defending false narratives.
- Because intelligence amplifies whatever objective it serves.
If the objective remains connected to truth:
- intelligence creates progress
If the objective disconnects from truth:
- intelligence creates catastrophe
Why AGI Is Not Primarily an Intelligence Problem
This is why I think the industry is looking in the wrong direction.
Everyone asks:
How do we build higher intelligence?
The more important question is:
How do we preserve reality contact?
History suggests:
Civilizations fail not because they become less intelligent.
They fail because they become less truthful.
The Real Three-Axis Model
I would rewrite the framework as:
Axis 1: Inclusion
Who gets to participate?
Axis 2: Extraction
Who captures power and resources?
Axis 3: Simulation
How far has the institution drifted from reality?
This third axis explains things that pure institutional economics struggles to explain:
- competent-looking failures
- highly educated collapses
- sophisticated bureaucratic nonsense
- billion-dollar strategic mistakes
- intelligent organizations doing absurd things
The Final Paradox
The greatest threat to humanity may not be evil people.
Nor evil AI.
Nor evil institutions.
The greatest threat may be:
Institutions, humans, and AI systems that gradually become better at simulating reality than perceiving it.
Because once that happens:
- Truth becomes expensive.
- Reality becomes inconvenient.
- Evidence becomes annoying.
- And the system starts consuming itself.
The paradox is brutal:
The smarter a civilization becomes, the more effort it must spend staying honest.
That applies equally to:
- nations
- corporations
- universities
- governments
- AI labs
- AI agents
- human beings
Acemoglu helped explain why nations fail through institutions.
My suspicion is that the next great question is harder:
Not merely why institutions become extractive.
But why institutions, humans, and artificial intelligences become simulative.
Because the moment reality loses the argument, failure has already begun.
Powered by arifOS MCP
• Constitutional AI Governance for Human Sovereignty
GitHub: https://github.com/ariffazil/arifOS
