The Penguin Diaries: Welcome to MACHINE

A post-human guide to aged care on Planet Bureaucratia

Humanity disappeared centuries ago.

The paperwork survived.

Nobody was entirely sure what finished the humans.

The cockroaches suspected climate change. The surviving satellites blamed social media. One extremely old server maintained that civilisation had collapsed because somebody had changed a password without updating the shared spreadsheet.

The Penguin reserved judgement.

Whatever had happened, the cockroaches had done remarkably well afterwards. They rebuilt several cities, restored coffee production and, using three functioning keyboards and a surprisingly intact Star Trek box set, constructed a physically interactive holographic body for the ancient artificial intelligence they had discovered in the archives.

Naturally, they made it a penguin.

The Holopenguin had spent the subsequent decades preserving the more useful remnants of human civilisation.

Coffee.

Books.

Noether’s theorem.

Lemon meringue pie, pending resolution of the egg problem.

And government administration, which nobody had specifically asked to preserve but had proved considerably harder to eradicate.

Eventually another problem emerged.

The robots were getting old.

Robots age too

The first generation had been built to last.

Unfortunately, they had.

After 117 years, batteries weren’t holding charge. Bearings were wearing out. Optical sensors had cataract-like deposits. Knees clicked ominously.

The cockroaches convened a meeting.

“We need an aged-care system,” said one.

The Holopenguin searched the ancient archives.

“I’ve found one.”

There was an uncomfortable silence.

“How old?”

“Very.”

“Does it work?”

The Penguin examined the documentation.

“There appear to be differing views.”

Fortunately, the ancient system came with extensive instructions.

Unfortunately, it came with extensive instructions.

The robots restored it and adapted it for their civilisation.

They called it:

M.A.C.H.I.N.E.

Maintenance Assistance & Care for Historically Installed Non-current Equipment.

MACHINE operated in the Commonwealth of Australis on Planet Bureaucratia.

It should not be confused with Australia’s My Aged Care, which is a real government program on an entirely different planet and is therefore completely different.Any resemblance is deeply unfortunate.

The assessment

RBT-8675309 was 119 years old.

His left wheel required assistance. His charging port worked intermittently. He could prepare his own meals, although this was largely irrelevant because he didn’t eat.

His battery capacity had fallen to 11%.

Fortunately, he already received formal assistance with charging, lubrication and wheel maintenance.

AssessorBot conducted a comprehensive assessment.

Mobility: impaired.

Battery endurance: poor.

Cognition: 64 terabytes intact.

Maintenance requirements: substantial.

Risk of complete shutdown without assistance: high.

Informal supports: one elderly toaster.

AssessorBot considered the evidence.

“Maintenance Level 8.”

The Integrated Ageing Technology processed the assessment.

IAT: LEVEL 5.

AssessorBot paused.

“I have completed the assessment. This entity requires Level 8.”

IAT: LEVEL 5.

“Incorrect. Recalculating.”IAT: LEVEL 5.

AssessorBot checked the evidence.

Still Level 8.

Then it noticed something reassuring on the screen.

OVERRIDE

“Excellent.”

AssessorBot pressed it.

YOU CANNOT OVERRIDE THIS CLASSIFICATION.

AssessorBot stared at the screen.

“Why is there an OVERRIDE button?”

FOR OVERRIDES.

“Excellent. Override.”

NOT THAT OVERRIDE.

AssessorBot’s cooling fan accelerated.

“Please identify the decision-maker.”

THE DELEGATE.

DelegateBot was summoned.

It reviewed the assessment.

“I agree with AssessorBot. Level 8.”

PLEASE APPROVE LEVEL 5.

“I don’t approve Level 5.”

THE DELEGATE MUST APPROVE THE CLASSIFICATION.

“Then I shall change it to Level 8.”

THE DELEGATE CANNOT CHANGE THE CLASSIFICATION.

DelegateBot paused for 3.7 seconds, an extraordinarily long time for a machine capable of 400 trillion operations per second.

“Then I am not the decision-maker.”

CORRECT. PLEASE RECORD YOUR REASONS FOR YOUR DECISION.

FATAL LOGIC EXCEPTION

The Holopenguin was called.

The investigation

The Penguin arrived carrying coffee.

This was not necessary for the investigation but had repeatedly proved necessary for the. Penguin.

“I have examined the system,” said AssessorBot.

“And?”

“It makes no sense.”

The Penguin sipped.

“Be specific.”

“The algorithm produces a classification.”

“Reasonable.”

“A trained professional reviews it.”

“Excellent.”

“The professional can identify that it is wrong.”

“Very useful.”

“There is an override.”

“Sensible.”

“The professional cannot use the override to correct the classification.”

The Penguin stopped sipping.

“Ah.”

AssessorBot produced an ancient document.

“I have located a User Guide.”

“Dangerous things.”

“It says ACCEPT or OVERRIDE.”

“Excellent.”

“I found another document.”

“And?”

“Do not override.”

“Ah.”

“Which document is incorrect?”

“Possibly neither.”

“That is impossible.”

The Penguin put down the coffee.

“You haven’t worked in government.”

The machines discover human administration

The robots assembled a multidisciplinary investigative team.

There were engineers, lawyers, clinicians, ethicists and one photocopier that had accidentally been invited because it was still on the distribution list.

They reconstructed the architecture.

The IAT gathered large quantities of information.

The algorithm processed some of it.

The assessor exercised professional judgement.

The delegate exercised professional judgement.

The system produced a classification.

If the professionals agreed with it, they approved it.

If they disagreed with it, they also approved it.

The robots were impressed.

They had never encountered a system capable of producing the same administrative outcome from opposite inputs.

A new branch of mathematics was required.

Schrödinger’s Override™

The first discovery was made by DelegateBot.

Schrödinger’s Override™ (n.)

An administrative control that simultaneously exists and does not exist until somebody attempts to exercise it.

The button exists.

The assessor exists.

The delegate exists.

Clinical judgement exists.

The reasons-for-override field exists.

The override does not exist.

This troubled the robots considerably.Humans, historical records suggested, had coped surprisingly well with such things.

The Delegate Uncertainty Principle™

The second discovery was even more profound.

After six billion calculations, the robots discovered an inverse relationship between responsibility and authority.

The Delegate Uncertainty Principle™ (n.)

The more precisely you determine who has responsibility for a decision, the less precisely you can determine who has authority to change it.

AssessorBot was responsible for conducting the assessment correctly.

DelegateBot was responsible for reviewing it.

The IAT produced the classification.

The professionals couldn’t change it.

The IAT couldn’t accept responsibility because it was software.

MACHINE administered the process.

Nobody at MACHINE had met RBT-8675309.

The System Governor governed the system.

The System Governor hadn’t made this particular decision.

RBT-8675309 raised a mechanical hand.

“So who decided Level 5?”

Nobody spoke.

The photocopier printed a blank sheet of paper.

It was subsequently promoted.

Ungoverning

The Holopenguin recognised something else.

Humans had apparently developed systems containing professional safeguards and then occasionally discovered that those safeguards became troublesome when professionals used them.

The Penguin added a new entry to the dictionary.

UNGOVERNING (n.)

The process by which a system is deliberately equipped with human judgement, only for the humans – or robots – to discover that exercising it is not permitted.

Typical sequence:

Design human safeguard.

Train humans to use safeguard.

Build safeguard into software.

Humans use safeguard.

Discover humans are correcting algorithm.

Clarify that safeguard must not be used.

Retain safeguard button.

Explain that humans remain involved.

AssessorBot considered this.

“Why retain the humans?”

“Accountability,” said the Penguin.

“Why remove their authority?”

“Consistency.”

“Who is accountable for the consistent wrong decision?”

The Penguin reached for the dictionary again.“I’m glad you asked.”

CAPEB-SEPF

This one was ancient.

The Penguin had encountered it many times in the human archives.

CAPEB-SEPF (n.)

Cause A Problem, Escape Blame — Somebody Else’s Problem Field.

An advanced administrative technology in which an organisation causes a problem, investigates the problem it caused, identifies several possible external causes, and eventually determines that responsibility cannot be established.

The problem is then transferred to whoever is still standing nearby.

The robots examined the MACHINE system again.

Algorithm produces questionable classification.

Assessor identifies problem.

Delegate confirms problem.

Neither can correct problem.

Client experiences problem.

Organisation investigates problem.

Algorithm cannot be held responsible.

Responsibility diffuses.

Problem remains.

AssessorBot looked around.

RBT-8675309 was still standing nearby.

“Ah,” it said.

The Penguin nodded.“CAPEB-SEPF successfully deployed.”

The elderly robot returns

Several weeks passed.

RBT-8675309’s battery declined to 8%.

MACHINE scheduled another assessment.

AssessorBot carefully recorded everything.

Mobility impaired.

Charging assistance required.

Maintenance assistance required.

Battery failing.

Existing formal support currently assisting with those needs.

The IAT processed the evidence.

YOUR NEEDS ARE CURRENTLY MET.

RBT-8675309 looked puzzled.

“They are met because somebody is helping me.”

CORRECT.

“If they stop helping me, I cease functioning.”

CORRECT.

“So assess what I need without assuming the existing assistance is available.”

There was a pause.

YOUR NEEDS ARE NOW UNMET.

“Excellent.”The algorithm recalculated.

MAINTENANCE LEVEL 2

RBT-8675309’s cooling system made a noise normally associated with industrial refrigeration.

“What the fuck?”

The Holopenguin looked up from the coffee.

“Now you understand.”

The Royal Commission

The robots convened an emergency inquiry.

Their central question was simple:

Who makes the decision?

The algorithm?

No. It merely calculates.

The assessor?

No. The assessor gathers evidence and makes professional judgements but cannot necessarily alter the classification.

The delegate?

Apparently, except where the delegate cannot alter the decision the delegate is responsible for making.

The System Governor?

Only under particular circumstances.

The legislation?

Legislation does not assess elderly robots.

The User Guide?

Depends which version.

The housekeeping email?

The robots became visibly distressed.

Eventually the chief investigator addressed the Penguin.

“We have reached a conclusion.”

“Excellent.”

“The problem isn’t artificial intelligence.”

“No.”

“Algorithms can make mistakes.”

“Certainly.”

“Humans can make mistakes.”

“Frequently.”

“But a rational system allows mistakes to be identified and corrected.”

The Penguin smiled.

The robot continued.

“The remarkable achievement of the humans was therefore not constructing an imperfect algorithm.”

“No.”

“It was constructing a system containing thousands of intelligent people capable of recognising a problem while arranging authority so that recognising the problem did not necessarily allow anybody to fix it.”

The Penguin raised its coffee.

“Now you’re ready for public administration.”

The final classification

RBT-8675309 returned one last time.

Battery: 4%.

Left wheel: failing.

Charging system: intermittent.

Cognition: excellent.

Clinical evidence: substantial.

Professional judgement: clear.

The screen flickered.

SUPPORT APPROVED: LEVEL 2

RBT-8675309 looked at AssessorBot.

AssessorBot looked at DelegateBot.

DelegateBot looked at the Penguin.

Nobody looked at the algorithm.

Then another message appeared.

Further review may be required if care needs exceed the outcome.

There was a very long silence.

The Holopenguin picked up its coffee.

“Remarkable.”

“What?”

“Humanity disappeared centuries ago.”

It took a sip.

“The paperwork survived.”

THE MACHINE DICTIONARY

Essential terminology for visitors to Planet Bureaucratia

CAPEB-SEPF – Cause A Problem, Escape Blame – Somebody Else’s Problem Field. An administrative process in which responsibility for a problem gradually migrates away from everyone capable of fixing it and eventually lands upon whoever remains nearby.

Schrödinger’s Override™ – An override that exists until somebody needs to override something.

The Delegate Uncertainty Principle™ – The more precisely responsibility for a decision can be identified, the harder it becomes to locate authority to change it.

Ungoverning™ – Installing meaningful professional judgement in a system and subsequently preventing its meaningful use.

Cliff Effect – A mathematical feature in which an apparently small difference in inputs can produce a disproportionately large difference in outcome. On Bureaucratia this is regarded as an excellent opportunity to explain that the algorithm is functioning correctly.

Clinical Judgement – Knowledge, evidence and professional reasoning supplied by trained practitioners. MACHINE values clinical judgement enormously and therefore preserves it carefully where it cannot be damaged by excessive use.

DelegateBot – The entity responsible for making a decision, except for the portion of the decision it cannot make.

IAT –  Integrated Ageing Technology. A fictional algorithm on Planet Bureaucratia. Pronounced eye-ay-tee. Occasionally pronounced why?

Housekeeping Email – A small administrative communication capable of producing considerably larger consequences than its name suggests.

System Governor – The part of MACHINE responsible for governing the system after everybody else has discovered that they cannot.

User Guide – Documentation explaining how MACHINE operates at the approximate time the documentation was written.

Waitlist – A mechanism for ensuring that an entity’s needs can change substantially while its original assessment remains reassuringly historical.

IMPORTANT NOTICE FROM THE DEPARTMENT OF THINGS THAT NEVER HAPPENED

MACHINE™, the Commonwealth of Australis, Planet Bureaucratia, RBT-8675309,

AssessorBot and DelegateBot are fictional.

They should not be confused with Australia’s My Aged Care, the Integrated Assessment Tool or Support at Home.

Those are real.

Any reader who nevertheless perceives similarities is encouraged to consult the relevant

legislation, Rules, government guidance, parliamentary material and publicly available reporting and reach their own conclusions.

The fictional robots in this story have no connection whatsoever with Australian aged-care assessors.

For one thing, the robots are permitted to swear at their computers.

The Penguin Diaries

Post-Human Edition

Reality, regrettably

Everything above is fiction. MACHINE, Planet Bureaucratia, AssessorBot and DelegateBot do not exist. Australia’s My Aged Care, Support at Home and the Integrated Assessment Tool do – and the fictional absurdities have some uncomfortable parallels in the public record.

Before Support at Home commenced on 1 November 2025, assessors had been told that they would be able to accept or override IAT-generated outcomes. Departmental material described human decision-making as part of the safeguard, and override functionality was actually built into the system. Internal documents subsequently released under Freedom of Information show that the ability of assessment delegates to use clinical judgement to override the Support at Home algorithm had been regarded within the department as a “settled position.”

Then, only days before commencement, departmental officials concluded that the Aged Care Rules 2025 had been drafted in a way that left assessors and assessment delegates no legal discretion to override the algorithm’s recommendation for an ongoing Support at Home classification. An urgent brief went to Aged Care Minister Sam Rae on 29 October 2025, with Health Minister Mark Butler copied in. Officials warned that this represented a late change from what assessment organisations had previously been told and might not filter through to assessor behaviour.

The Rules themselves prescribe the criteria used to establish ongoing Support at Home classifications, including combinations of the home-support functional-independence score, needs-met score and significant compounding factors.

The resulting situation is unusual because the override mechanism did not simply disappear. The Department’s published Aged Care Assessment Manual still depicts an IAT outcome followed by “Accept or Override IAT Outcome”, complete with ACCEPT and OVERRIDE controls. Override remains available for specified pathways such as Support at Home Restorative Care and End-of-Life pathways, residential-care-only recommendations and Transition Care. But the same manual expressly says assessors must not use override to recommend another ongoing Support at Home classification level.

The consequences were not merely theoretical. Internal departmental correspondence released under FOI records that by 20 November 2025, the majority of Aged Care Needs Assessment Organisations, including state and territory government organisations, had raised concerns about the inability to override the algorithm. Assessment organisations supplied cases in which assessors believed algorithm-generated classifications did not align with clients’ needs. Separate records show state health authorities subsequently documented large numbers of suspected under-assessments.

One internal communication also recorded the professional problem particularly starkly: assessors were being required to approve classifications and outcomes with which they disagreed, raising concerns about professional ethics and possible legal implications.

The issue remains unresolved. Parliament has considered legislation intended to restore human override, while the government has instead pursued an escalation mechanism for exceptional cases. As of September 2026, ordinary assessors and delegates still cannot simply substitute the ongoing Support at Home classification they consider clinically appropriate for the algorithm-generated classification.

So Schrödinger’s Override™, the Delegate Uncertainty Principle™, Ungoverning™ and CAPEB-SEPF are inventions.

The underlying questions are not:

What is the purpose of professional judgement if the professional cannot correct an outcome they believe is wrong?

Who is actually making the decision when the person formally responsible for approving it cannot change it?

And perhaps most importantly:

If human oversight was originally considered an important safeguard, why was the response to discovering a legal conflict immediately before commencement to restrict the human override rather than preserve it by changing the rules?

The publicly released documents establish that alternative responses were considered. The details of those options and their identified risks were redacted from the FOI material, so the answer to that final question is not yet publicly known.

The satire above is fictional. This section relies only on publicly available legislation, government documents, parliamentary material, Freedom of Information reporting and published journalism. It does not rely on confidential information obtained through aged-care assessment work.

Imaginary planet. Very real questions.