AI AND THE DEMOCRATISATION OF SCRUTINY

The Australian State, Algorithmic Governance and the Right of Citizens to Question Power

By Andrew Klein
The Patrician’s Watch

The questions we must ask

If you are confident that your decisions serve the public interest, why should the public be prevented from examining how you reached them?

A democracy cannot remain healthy when the governed are increasingly visible to the institutions of power, while the institutions of power become increasingly opaque to the governed.

And perhaps the most fundamental question:

What gives a government the moral legitimacy to exercise power if its decisions consistently undermine the welfare of the people it governs?

I. Introduction: Who Watches the Watchers?

Artificial intelligence is changing the relationship between individuals, institutions and the state.

It offers extraordinary opportunities. It can improve medical research, assist scientific discovery, make public services more accessible, analyse complex information and help governments identify problems that might otherwise remain hidden.

It can also concentrate power.

The ability to collect, combine, classify and analyse enormous quantities of information gives institutions new opportunities to understand the populations they administer. It can improve service delivery, but it can also increase the capacity to monitor, categorise, predict and influence human behaviour.

The technology itself does not determine which outcome will prevail.

That is a question of governance.

In Australia, the Albanese government has embraced artificial intelligence as an important component of economic development, public administration and national infrastructure. It has established an Office of AI, developed policies for responsible AI use within government and entered into collaborative arrangements with major technology companies.

These developments deserve examination.

Not because technological progress should be opposed, but because the institutions responsible for introducing these systems must remain accountable to the people affected by them.

There is a fundamental democratic distinction between using AI to improve public administration and using AI to make public administration less accessible to scrutiny.

The distinction becomes particularly important when automated systems influence decisions about welfare, disability support, taxation, immigration, healthcare, employment and access to public services.

Who controls the information?

Who determines how it is classified?

Who establishes the criteria used to assess individuals?

Who decides what constitutes risk?

Who can challenge an automated conclusion?

And who accepts responsibility when the system gets it wrong?

These are not merely technical questions. They concern the distribution of power within a democratic society.

The argument advanced here is that artificial intelligence should become an instrument of democratic participation and public accountability, rather than simply another means by which institutions exercise authority over individuals.

AI should democratise the scrutiny of power, not democratise the power to scrutinise and control the population.

II. The Transformation of the Citizen into Data

Modern government requires information.

Departments must understand population needs, allocate resources, administer services and assess eligibility. Statistical information is essential to planning and public administration.

There is nothing inherently wrong with collecting or analysing data.

The difficulty arises when the administrative representation of an individual becomes more important than the individual themselves.

A person becomes a welfare recipient.

A disabled person becomes a participant.

An older Australian becomes a consumer of aged-care services.

A veteran becomes a claimant.

A person experiencing homelessness becomes a case.

These classifications may be necessary for administrative purposes. They are not necessarily inaccurate or malicious.

But they are incomplete.

A person is more than the category into which an institution places them.

A person living with disability may experience fluctuating functional capacity. A standardised assessment may fail to recognise the circumstances that determine their actual support requirements.

A person with an unusual employment history may appear anomalous within an income dataset.

A family experiencing temporary hardship may be classified according to historical information that no longer reflects its circumstances.

An algorithm can process the available information with extraordinary speed and consistency while producing a conclusion that is fundamentally wrong.

The problem is not necessarily the calculation.

It may be the information selected, the assumptions embedded in the model, the categories used or the failure to recognise circumstances that do not fit the administrative framework.

This leads to an important distinction:

The data point is not the person. The statistical representation is not the lived experience.

AI can identify patterns within information. It cannot automatically establish that the information is complete, that the categories are appropriate or that the conclusions are fair.

Those questions require human judgement, independent scrutiny and an appreciation of the circumstances of the people affected.

There is a danger that administrative efficiency becomes the dominant measure of success.

A system may process more applications, reduce administrative costs and produce decisions more quickly. Those outcomes may be desirable.

But what happens when the system also produces more incorrect decisions, makes appeals more difficult or excludes people whose circumstances do not fit its assumptions?

Efficiency without fairness is not good administration.

It is simply the faster administration of potentially defective decisions.

III. The Architecture of Government: Silos and the Diffusion of Responsibility

The introduction of AI into public administration does not occur within a single, unified institution.

Australian government consists of departments, agencies, statutory authorities, contractors, consultants and multiple levels of administration.

Each has its own responsibilities, information systems, reporting arrangements and institutional priorities.

This fragmentation can make coordination difficult.

It can also make accountability difficult.

Consider a hypothetical automated government assessment.

One department establishes the policy.

Another provides the data.

A technology contractor develops or supplies the system.

A third agency administers the service.

An internal committee approves its implementation.

A minister announces the policy.

An individual subsequently receives an adverse decision.

Who is responsible?

The department may say that it followed the approved policy.

The contractor may say that the system operated according to specification.

The administrator may say that it applied the system’s recommendation.

The minister may say that operational decisions are matters for the department.

The individual is left attempting to navigate an institutional structure that possesses considerably more information and technical expertise than they do.

This is not a hypothetical problem confined to AI. It is a longstanding difficulty within complex public administration.

The introduction of sophisticated automated systems can make it more pronounced.

Responsibility becomes distributed across the architecture of government.

The result can be a situation in which every component of the system has fulfilled its assigned task, yet the overall outcome is unacceptable.

This is precisely why accountability must extend beyond the technical performance of an AI system.

We need to know who authorised its use, who established its criteria, who monitors its operation and who possesses the authority to intervene.

A department should not be able to transfer responsibility to a contractor.

A minister should not be able to transfer political responsibility to a department.

A contractor should not be able to avoid scrutiny by claiming that its technology is commercially confidential.

And an automated system should never become an institutional excuse for the absence of accountable human judgement.

The more complex the system, the clearer the lines of responsibility must become.

IV. The Albanese Government’s AI Agenda

The Albanese government has taken several significant steps towards establishing Australia’s AI policy framework.

The Office of AI was established within the Department of the Prime Minister and Cabinet on 15 July 2026.

Its responsibilities include coordinating work across government to develop national AI standards, establish requirements for AI infrastructure and support the development of AI training capability in Australia.

The government’s objectives include attracting investment, developing domestic capability and ensuring that AI infrastructure operates in accordance with Australian interests.

These are legitimate policy objectives.

Australia should not ignore the economic, scientific and strategic importance of artificial intelligence.

However, the establishment of a central coordinating office raises an important governance question:

Who independently scrutinises the institution responsible for coordinating the government’s AI agenda?

Central coordination can improve consistency and reduce duplication. It can also concentrate policy development within the executive.

There is no automatic contradiction between central coordination and democratic accountability.

But coordination must be accompanied by transparency, parliamentary oversight, independent evaluation and meaningful opportunities for public challenge.

The government’s Policy for the Responsible Use of AI in Government, version 2.1, updated in October 2026, establishes mandatory requirements for Commonwealth departments and agencies.

These include:

  • Designated accountable officials.
  • Public AI transparency statements.
  • Internal registers of AI use cases.
  • Staff training.
  • Agency AI strategies.
  • Risk-based impact assessments.
  • Governance and monitoring arrangements.

This is a substantial policy framework.

It should be acknowledged as such.

However, the existence of a policy does not establish that its implementation is effective, that every relevant system is adequately disclosed or that affected citizens possess enforceable rights to explanation and review.

There are also exclusions for Defence and the national intelligence community.

Some exclusions are understandable given legitimate security requirements. But national security cannot become a universal justification for avoiding independent oversight.

The appropriate question is whether secrecy is necessary, proportionate and subject to independent review.

The distinction matters.

A democracy can protect sensitive information without abandoning accountability.

The government has established a framework. The next question is how effectively that framework is implemented and whether citizens can rely upon it when their interests are affected.

V. Government and the AI Industry: Cooperation or Dependence?

The relationship between government and major technology companies deserves particular attention.

The Australian Government has entered into memoranda of understanding with companies including Microsoft and Anthropic.

These arrangements concern collaboration on AI opportunities, research, skills development, infrastructure and potential cooperation with the Australian Public Service.

The government has stated that these memoranda are non-binding and do not confer preferential treatment in Commonwealth procurement, grants or regulatory decisions.

That qualification is important.

A memorandum of understanding does not establish corporate capture, nor does collaboration with a technology company demonstrate improper influence.

Government needs access to expertise. Technology companies possess capabilities that public institutions may not have developed internally.

However, cooperation creates legitimate questions about the distribution of influence.

Who participates in policy discussions?

Which companies have access to senior decision-makers?

Are independent researchers, community organisations and smaller technology providers given comparable opportunities to contribute?

Are public servants developing sufficient internal expertise to evaluate the products and advice they receive?

Does government retain the capacity to change providers, audit systems and understand the technical architecture of the systems upon which it depends?

What happens when a public institution becomes dependent upon a private provider for the operation of an essential service?

These are questions of institutional independence.

The problem is not that government works with private industry.

The problem would arise if government became unable to exercise independent judgement because its technical knowledge, infrastructure or administrative capacity had become dependent upon commercial providers.

There is also a broader question concerning the allocation of public resources.

AI infrastructure requires enormous investment in land, electricity, water, telecommunications and supporting infrastructure.

The benefits may be substantial, but the costs are also borne by communities and public systems.

We should ask:

  • Who owns the infrastructure?
  • Who receives the commercial benefits?
  • Who pays for the supporting public infrastructure?
  • What obligations do companies have to local communities?
  • What protections apply to Australian intellectual property and personal information?
  • What happens when a commercial provider changes its business model or withdraws a service?

The public interest must remain the organising principle.

Government should be a partner of industry, not its captive.

VI. Curating Public Opinion: When Consultation Becomes Classification

One of the most revealing examples of the democratic implications of AI appeared in the Australian Government’s September 2026 consultation on AI infrastructure.

The consultation paper stated that the government might use AI tools to summarise and compare public submissions, including sentiment analysis to identify patterns and trends.

There is nothing inherently improper about this.

Governments regularly receive large numbers of submissions. Analytical tools can assist officials in identifying recurring concerns, organising information and understanding the range of views expressed.

But the use of AI in public consultation raises an important question:

What happens to dissent when individual submissions are converted into statistical summaries?

Imagine that 10,000 Australians respond to a proposed policy.

An analytical system identifies:

  • 62 per cent broadly supportive.
  • 23 per cent opposed.
  • 15 per cent undecided.

The result may provide a useful overview.

But what if the 23 per cent opposed include people whose lives will be profoundly affected?

What if their submissions identify a specific legal defect, a discriminatory consequence or a danger that the majority has not experienced?

What if a detailed and carefully reasoned objection is classified as negative sentiment?

The statistical summary may accurately describe the distribution of responses while concealing the importance of particular arguments.

This is the danger of confusing the frequency of an opinion with its significance.

A minority objection may identify a serious problem.

A single well-supported warning may be more important than thousands of general expressions of approval.

Democratic consultation should not operate as a popularity contest in which the numerical majority automatically determines the importance of an argument.

The government must therefore explain how AI-assisted consultation is conducted.

It should disclose:

  1. Whether AI was used to analyse submissions.
  2. Which systems were used.
  3. What instructions and classification criteria were applied.
  4. Whether human reviewers checked the outputs.
  5. Whether original submissions were retained and considered.
  6. How minority and dissenting views were preserved.
  7. Whether independent reviewers can examine the analytical process.

The September consultation attracted more than 1,500 submissions. The government subsequently published a report describing the themes and issues identified.

That publication is a welcome step towards transparency.

But the existence of a summary does not remove the need to preserve the underlying arguments and make the analytical process open to examination.

We should be particularly careful when governments use AI to interpret the views of the people they govern.

The purpose of consultation is not to discover how efficiently public opinion can be summarised. It is to understand what the public is actually saying.

VII. Robodebt and the Automated State: Have We Learned Enough?

Australia does not need to speculate about the consequences of defective automated administration.

Robodebt provides a devastating example.

The Royal Commission into the Robodebt Scheme exposed failures involving policy development, legal advice, administrative processes, institutional culture and accountability.

The scheme used income averaging to generate debts against welfare recipients, despite the inadequacy of that method for establishing individual debts.

The consequences were profound.

People were pursued for debts they did not owe. Individuals experienced distress, financial hardship and damage to their lives.

The Royal Commission examined not merely the technology or administrative machinery but the decisions, assumptions and institutional behaviours that allowed the scheme to operate.

Its Recommendation 17.1 called for a consistent legal framework governing automation in government services.

It proposed that affected individuals should have a clear path to review, that departments should explain the use of automated decision-making and that business rules and algorithms should be available for independent expert scrutiny.

The Australian Government accepted the recommendation for a consistent framework.

However, the March 2026 implementation update still recorded Recommendation 17.1 as ongoing.

This deserves attention.

It is not sufficient to say that government has learned from Robodebt because it now has an AI policy.

Robodebt was not simply a failure of technology.

It was a failure of institutional judgement and accountability.

A system can be technically sophisticated and still be legally defective.

It can be administratively efficient and still produce injustice.

It can be approved by multiple levels of government and still be fundamentally wrong.

The lesson is not that automated systems should never be used.

The lesson is that their use must be governed by enforceable safeguards, independent scrutiny and meaningful rights of review.

An individual affected by an automated decision should be entitled to know:

  • Whether automation contributed to the decision.
  • What information was considered.
  • Which rules or criteria influenced the outcome.
  • Whether a human decision-maker reviewed the matter.
  • How to correct inaccurate information.
  • How to challenge the decision.
  • Which public authority remains accountable.

A review process that merely asks another official to accept the same automated conclusion is not necessarily meaningful review.

Human oversight must be substantive, not ceremonial.

The Robodebt experience should have made this principle unavoidable.

VIII. Bandura and Moral Disengagement: How Institutions Distance Themselves from Consequences

The psychologist Albert Bandura provides a useful framework for examining how individuals and institutions can justify harmful conduct while avoiding personal responsibility.

In his 1999 paper, Moral Disengagement in the Perpetration of Inhumanities, Bandura examined mechanisms through which moral self-regulation can be selectively disengaged.

His framework identifies eight related mechanisms.

1. Moral justification

Harmful conduct is presented as necessary to achieve a supposedly worthy objective.

In government, a damaging policy may be defended as necessary for efficiency, fiscal sustainability, national security or administrative integrity.

2. Euphemistic labelling

Language sanitises the consequences of an action.

Cuts become sustainability measures.

Surveillance becomes integrity management.

Exclusion becomes eligibility rationalisation.

The language may be technically accurate in an administrative sense while concealing the human consequences.

3. Advantageous comparison

A harmful decision is defended because a supposedly worse alternative might have occurred.

The question becomes whether the policy could have been worse, rather than whether it is justified.

4. Displacement of responsibility

Officials claim that they were following instructions, implementing approved policy or complying with established procedures.

5. Diffusion of responsibility

Responsibility is distributed across departments, committees, contractors and administrative layers until no individual appears accountable.

6. Distortion or minimisation of consequences

The harm caused by a decision is minimised, discounted or rendered statistically insignificant.

7. Dehumanisation

The people affected become categories, cases, risk profiles or data points rather than individuals with particular circumstances and legitimate expectations.

8. Attribution of blame

Those harmed are portrayed as responsible for their own circumstances.

An individual who challenges an incorrect assessment becomes difficult.

A welfare recipient becomes suspicious.

A person unable to navigate an administrative process becomes non-compliant.

The language shifts attention away from the institutional decision and towards the supposed deficiencies of the individual.

Bandura’s work is particularly relevant because he observed that harmful outcomes can emerge through networks of legitimate organisations and otherwise conscientious individuals, with responsibility divided across separate functions.

This is not an accusation that every government official is morally disengaged.

Nor does the presence of a particular linguistic mechanism establish deliberate deception.

It is a framework for examining institutional behaviour.

AI can assist in identifying patterns of language, comparing policy explanations and detecting inconsistencies between stated objectives and documented outcomes.

But AI should not become the moral authority.

It can identify possible mechanisms of disengagement. It cannot independently establish intention or determine the final moral significance of a decision.

That remains a human responsibility.

The purpose of moral scrutiny is not to condemn people automatically. It is to prevent institutions from becoming so detached from human consequences that responsibility disappears.

IX. Zhengming and Tianming: Does Government Language Correspond with Reality?

There are two Chinese philosophical concepts that provide an interesting perspective on the relationship between political authority and public responsibility.

They should not be treated as interchangeable with modern democratic principles. They emerged from different historical and philosophical traditions.

Nevertheless, both raise questions that remain relevant.

Zhengming (正名): The Rectification of Names

The term zhèngmíng is generally translated as the rectification of names.

It is associated with Confucius and appears in Book 13, Chapter 3 of the Analects.

The principle concerns the relationship between language, responsibility and reality.

If a ruler calls something just, is it just?

If an institution calls something transparent, is it actually transparent?

If a government describes a decision-making process as accountable, can the people affected identify who is responsible?

If a policy is described as consultation, have the views expressed genuinely influenced the decision?

The principle is straightforward:

The name must correspond with the reality.

This is an exceptionally useful test for modern government.

Consider the language surrounding AI.

A system is described as responsible.

What does that mean in practice?

A department announces transparency.

Can citizens examine how the system works?

A government describes an automated decision as objective.

Have its assumptions, data and criteria been independently tested?

A minister describes a policy as protecting vulnerable people.

Has the policy actually improved their circumstances?

The words are not enough.

The reality must be examined.

Tianming (天命): The Mandate of Heaven

The second concept is tiānmìng, commonly translated as the Mandate of Heaven.

In traditional Chinese political philosophy, the legitimacy of a ruler was understood to depend upon moral conduct and the welfare of the people.

The mandate was not simply an unconditional entitlement to exercise power.

A ruler who governed unjustly could be understood to have forfeited the moral legitimacy of that authority.

We should not equate this historical concept with contemporary constitutional democracy.

But its underlying question is worth considering:

What gives a government the moral legitimacy to exercise power if its decisions consistently undermine the welfare of the people it governs?

Modern democratic governments derive their legal authority through constitutional and electoral arrangements.

But legal authority and moral legitimacy are not identical questions.

A government may possess the lawful authority to make a decision while still being required to justify that decision to the people affected.

Elections do not remove the obligation to govern responsibly between elections.

Nor does parliamentary authority eliminate the need for transparency, accountability and independent scrutiny.

The two concepts offer complementary tests:

Zhengming: Does the government’s language correspond with reality?

Tianming: Does the exercise of authority serve the welfare of the people?

Both are relevant to the governance of artificial intelligence.

X. The Democratisation of Scrutiny: AI as Democratic Infrastructure

We have spent considerable time examining how AI might increase the capacity of government to collect information, classify individuals and administer decisions.

But there is another possibility.

The same technology can give ordinary citizens the ability to examine institutions that were previously difficult to scrutinise.

This is where the most constructive opportunity lies.

For much of modern history, detailed examination of government activity has required considerable resources.

A journalist needed access to documents, specialist knowledge and time.

A community organisation needed researchers and legal assistance.

A citizen challenging a government department often faced an institution with considerably greater access to information and technical expertise.

AI can change that relationship.

A citizen can compare ministerial statements with departmental expenditure.

A journalist can examine procurement contracts and identify recurring suppliers.

A disability advocate can compare policy announcements with changes to eligibility and service delivery.

A researcher can examine thousands of pages of government reports and identify inconsistencies.

A community organisation can compare public consultation outcomes with the original submissions.

A parliamentary committee can use AI to identify patterns across large volumes of departmental material.

These activities do not require AI to become an independent moral authority.

They require it to function as an accessible analytical instrument.

The distinction is fundamental.

AI should not replace journalists, researchers, auditors, lawyers, public servants or democratic institutions.

It should increase the capacity of those institutions and individuals to examine evidence.

Nor should AI-generated conclusions be accepted without verification.

Models can produce errors, misinterpret context, reproduce bias and invent connections.

The original documents must remain accessible.

The analytical process must be open to examination.

The conclusions must be capable of independent verification.

The purpose is not to replace one form of authority with another.

It is to distribute the capacity for scrutiny more widely.

A fourth path

Much of the international discussion about AI governance is organised around competing priorities.

The United States has generally emphasised innovation and commercial development.

The European Union has placed considerable emphasis on rights-based regulation and risk management.

China operates within a system of strong state coordination and political control.

These are broad descriptions, and none captures the full complexity of the respective systems.

But there is another question that deserves consideration:

How can AI be organised to strengthen the capacity of citizens to hold power accountable?

This is a fourth path.

It does not begin with the interests of the technology industry or the administrative convenience of government.

It begins with the democratic rights of the people.

AI becomes democratic infrastructure when it enables citizens to understand, examine and challenge the institutions exercising authority over them.

The lens should not belong exclusively to those who govern.

It should belong to the governed as well.

XI. The Expectations of the Governed: Ten Democratic Safeguards

If governments intend to use AI in public administration, citizens should possess meaningful rights concerning its use.

These should not depend upon the goodwill of individual departments or the voluntary disclosure of technology providers.

They should be supported by clear legal and administrative protections.

The following ten principles are proposed democratic safeguards. They should be treated as a framework for reform, not as a claim that every right already exists in Australian law in this form.

1. The right to know

Citizens must be informed when AI materially contributes to a government decision affecting them.

This should include decisions involving eligibility, entitlements, assessments, enforcement and access to essential services.

2. The right to explanation

People must be able to understand the information and reasoning behind a decision.

A statement that a computer system produced an outcome is not an adequate explanation.

3. The right to human review

A suitably qualified person must be able to reconsider the matter independently.

The reviewer must possess the authority to depart from the automated recommendation.

4. The right to challenge the data

Individuals must be able to correct inaccurate information, challenge inappropriate classifications and provide relevant information that the system may have overlooked.

5. The right to independent scrutiny

Algorithms, business rules, impact assessments and relevant technical documentation should be accessible to independent experts, subject to properly justified privacy, security and commercial protections.

Commercial confidentiality must not become a blanket exemption from public accountability.

6. The right to public scrutiny

Government use of AI in consultation, policy development and public administration must be transparent.

Where AI is used to analyse public submissions, the methods, criteria and treatment of dissenting views should be disclosed.

7. The right to accountability

Government cannot transfer responsibility for its decisions to a computer program or private contractor.

The responsible public authority must remain identifiable and answerable.

8. The right to dissent

Citizens must retain meaningful opportunities to challenge the assumptions, classifications and decisions of public institutions.

Dissenting views must not disappear simply because they are statistically uncommon or difficult to classify.

9. The right to remain a person

A statistical category, risk score or algorithmic prediction must never become a substitute for considering the actual circumstances of an individual.

This is perhaps the most important safeguard of all.

The purpose of public administration is to serve people.

People do not exist to make administrative systems easier to operate.

10. The right not to have to understand the machinery

A citizen should not need a degree in data science, administrative law, information technology or public policy to challenge an incorrect government decision.

Government designs the system. Government selects the information. Government establishes the rules. Government chooses the technology and determines how it will be used.

The responsibility for explaining that system must therefore rest with the institution that operates it.

A person should not have to discover which department supplied an inaccurate record, which contractor developed an algorithm, which administrative rule produced an adverse outcome or which technical classification caused their circumstances to be misunderstood.

Those are matters for the institution to investigate and explain.

The citizen should be required to establish the circumstances of their case, not reverse-engineer the machinery of government.

The citizen should never have to become an expert in the machinery of government simply to establish that the machinery has made a mistake.

That is not a demand for special treatment. It is a basic principle of procedural fairness.

The greater the technical complexity of government administration becomes, the greater the responsibility of government to make its decisions understandable, accessible and open to challenge.

Technology should reduce the burden placed upon citizens, not increase it.

And where an automated system makes a mistake, the institution responsible for that system must carry the burden of investigating, explaining and correcting it.

The machinery exists to serve the citizen.

The citizen does not exist to serve the machinery.

XII. Conclusion: The Lens Belongs to the People

Artificial intelligence presents governments with extraordinary opportunities.

It can improve public administration, assist scientific discovery, support economic development and make information more accessible.

But technological capability does not confer democratic legitimacy.

The fact that government can collect more information does not mean that it should collect information without adequate justification.

The fact that a system can classify people does not establish that its classifications are fair.

The fact that an algorithm can produce a decision does not establish that the decision is lawful, reasonable or humane.

And the fact that a government can summarise thousands of public submissions does not mean that it has understood the people who made them.

We must be particularly careful when institutional efficiency becomes the dominant measure of success.

A government may process applications more quickly while making it more difficult for people to challenge incorrect decisions.

It may collect more information while disclosing less about how that information is used.

It may describe a consultation as inclusive while allowing the complexity of its analytical process to obscure minority objections.

It may introduce transparency statements while leaving citizens without sufficient information to challenge the decisions affecting them.

These are the contradictions that deserve examination.

The Albanese government’s AI policy contains commitments to responsible use, transparency and accountability. Those commitments should be acknowledged.

But commitments are not the same as demonstrated outcomes.

The existence of an AI policy does not remove the need for enforceable rights, independent oversight and meaningful parliamentary scrutiny.

The continuing implementation of the Robodebt Royal Commission’s recommendations demonstrates why this matters.

We should not wait for another major administrative failure before establishing protections that could have prevented it.

Nor should we accept the proposition that the technical complexity of AI makes public scrutiny impractical.

If anything, increasing complexity makes independent scrutiny more important.

The relationship between government and citizen must not become one in which institutions know progressively more about individuals while individuals know progressively less about the institutions making decisions about them.

That is not technological progress in any meaningful democratic sense.

It is an imbalance of power.

There is a profound opportunity to do something different.

AI can give ordinary Australians access to analytical capabilities that were once available primarily to governments, large corporations and well-funded institutions.

It can help people examine public expenditure, question policy assumptions, scrutinise procurement, compare official statements and identify discrepancies between government promises and government performance.

It can make democratic participation more informed and more accessible.

But that opportunity depends upon public access to information, independent verification and a willingness to protect dissent.

We should not confuse the ability to collect information with the wisdom to use it.

We should not confuse the ability to predict behaviour with the right to control it.

And we should never confuse an administrative classification with an understanding of the person being classified.

Perhaps this is where the ancient principle of Zhengming becomes particularly relevant.

If government calls something transparency, let us examine what it actually discloses.

If it calls something accountability, let us establish who can be held responsible.

If it calls something consultation, let us determine whether the voices of the people have genuinely influenced the outcome.

If it calls something efficient, let us examine whose lives have improved.

The names must correspond with the reality.

And perhaps Tianming offers another question, one that every government should be prepared to answer:

Does the exercise of authority serve the welfare of the people?

Democratic government is not simply the possession of power through electoral mandate.

It is the continuing responsibility to exercise that power in the public interest.

That responsibility does not disappear when decisions become technologically sophisticated.

It becomes more important.

We should welcome artificial intelligence as a tool for expanding human understanding, improving public administration and strengthening democratic participation.

But we should resist its use as an instrument that makes people easier to manage and institutions harder to question.

The future of AI should not be determined exclusively by governments, technology companies, investors or the institutions that possess the greatest quantities of data.

The people whose lives will be affected must have a meaningful place in determining that future.

And that brings us back to the fundamental question:

If you are confident that your decisions serve the public interest, why should the public be prevented from examining how you reached them?

A democracy cannot remain healthy when the governed are increasingly visible to the institutions of power, while the institutions of power become increasingly opaque to the governed.

And finally:

The citizen should never have to become an expert in the machinery of government simply to establish that the machinery has made a mistake.

The machinery exists to serve the citizen.

The citizen does not exist to serve the machinery.

AI should make power more accountable, not make accountability more difficult.

The lens belongs to the people.

Andrew Klein
The Patrician’s Watch

Leave a Reply