Demechracy

1.0 The Dunbar Limit

1.1 The Ceiling on Compassion

There is a number beyond which a human being cannot truly know another. It was described in 1992 by the evolutionary anthropologist Robin Dunbar as he proposed that the size of the social group a primate can stably maintain is constrained by the size of its neocortex (Dunbar, 1992). Extrapolating from the primate regression to Homo sapiens yielded a predicted group size of approximately 150 individuals, the now-famous Dunbar's number. Subsequent empirical work, drawn from hunter-gatherer band sizes, Roman military unit composition, Hutterite communal settlements and the active contact lists of modern telecommunications users, has broadly confirmed the figure as the ceiling on stable reciprocal acquaintance (Dunbar, 1993, 2010; Hill & Dunbar, 2003).

Within this range, a person can track the characters, relationships, obligations and grievances of others in sufficient detail to extend genuine personal judgement. Beyond it, the bookkeeping collapses. At approximately one hundred and fifty, knowing gives way to inference; we stop knowing people and begin reading them off the groups we have filed them under. Effort, education and goodwill will not change it. The ceiling is an artefact of neural architecture that, like every other biological constant identified in this party's earlier work on the asymmetry between biology and machine (Newton-Thomas & Strocchi, 2018; Australian Renaissance Party [ARP], 2026), has not changed in anatomically meaningful terms for approximately one hundred thousand years.

Every governance question this paper raises flows from that single fact.

1.2 Governance Beyond the Band

For most of the human evolutionary record, the ceiling posed no problem. Bands, camps and early villages sat within it. Governance, such as it was, could be discharged through acquaintance, reciprocity and reputation. The chief, the elder, the headwoman knew every person under their authority, and the authority was legitimate in the only way it could truly be: the governed were not abstractions to those who governed them.

The agricultural revolution changed the numbers. Settled populations of thousands, and then tens of thousands, could no longer be held together by acquaintance alone. Every governance technology that followed, from writing, census and law to coinage, the representative assembly and the secret ballot, is in one form or another a workaround for the Dunbar ceiling. Each extends the reach of governance by compressing the individual into a category the governors can actually process. A taxpayer. A subject. A citizen. A voter. Each is a cognitive handle, not a person known.

Modern representative democracy is the most sophisticated of these workarounds, but it remains a workaround. A member of the Australian House of Representatives notionally represents an electorate of approximately one hundred and twenty thousand enrolled voters (Australian Electoral Commission [AEC], 2025). No human being can know one hundred and twenty thousand people. The fiction that they can is the fiction on which industrial-scale representation rests. A necessary fiction; but a fiction still.

1.3 Factionalism as the Cognitive Shortcut

When the network of the governed exceeds what the governor can track, something must fill the gap. What fills it is category.

Factionalism, the organisation of political life around group identity rather than individual relationship, arises from necessity more than from character. When a cognitive system is asked to process more relationships than it evolved to carry, it reaches for category in place of acquaintance. Party, class, region, ethnicity, accent, dress, religion, profession, age cohort: these are heuristics the brain supplies when the Dunbar budget is exhausted, so that some form of judgement can still be extended to a population too large to know. They appear cynical only from the outside.

The social-identity literature has documented the mechanism in detail (Tajfel & Turner, 1979). Even in the absence of real-world stakes, subjects arbitrarily assigned to meaningless groups preferentially allocate rewards to members of their own group. Where the group lines are meaningful, whether ideological, tribal or economic, the preference sharpens into loyalty, and the loyalty organises political life. Every functioning democracy of any appreciable size is, in this sense, a factional democracy, because every democracy of any appreciable size is too large for non-factional cognition.

This is the clay of which people are made. Electing better ones will only temporarily change the pattern.

1.4 Two Outputs: Favouritism and Failure

Factional politics produces two characteristic outputs, and every large democracy exhibits some mixture of both.

The first is favouritism. Where the factional line runs clearly between winners and losers, the faction in power allocates resources, positions and policy advantage to its own group and its own donors. The behaviour is rational in its context. Cognitive and political capacity is aligned with a subset of the population rather than the whole, and allocation follows. The word corruption, in its narrow legal sense, does not quite capture it. The Australian record furnishes a continuous archive: ministerial discretion directed to marginal seats, grant programs distributed along partisan lines, and the well-documented correlation between political donations and policy access (Achen & Bartels, 2016; The Australia Institute, 2023). The pattern crosses all parties. It follows structurally from governing by faction.

The second output is failure. Where a group lies outside the factional coalitions that contest for power, whether because it is too small to organise, too dispersed to mobilise, or simply invisible to the categories the parties recognise, its interests disappear from view. The policy process does not oppose them; it has no sight of them at all. Remote Indigenous communities, long-tail disability cohorts, carers in the informal economy and the residents of declining post-industrial towns have each, at various periods, occupied this blind zone. Their problems are not solved because, in the cognitive architecture of representative government, they are not fully present as subjects of policy.

The Dunbar ceiling that produces favouritism at the top of the distribution also produces invisibility at the tail. One is a crowd; the other is a silence. Both are failures of sight.

1.5 The Wider Bounds

The Dunbar ceiling is the most studied of the brain's bounds, but it is not the only one, and the case for Demechracy does not rest on it alone. Two further constraints sit beneath it, and together with Dunbar they describe the cognitive substrate on which all human governance is conducted.

The first is the bound on working memory. The number of discrete items a human mind can hold simultaneously in conscious attention is, on the cognitive-psychology literature of the last seventy years, around four to seven (Miller, 1956; Cowan, 2001). Not four hundred. Not four thousand. Four. A minister considering a policy must, in any given moment of deliberation, compress every interest, every electorate, every dependency, every second-order effect, into the four-item slot that conscious reasoning permits. The compression is unavoidable. What is lost in it is the matter of governance.

A frontier large language model, by contrast, holds a working context now exceeding one million tokens (Anthropic, 2025; Google DeepMind, 2025). That is enough to keep every Australian's recent administrative history in active memory while it reasons about a policy. The ratio between the two substrates runs into the hundreds of thousands.

The second is the bound on transmission. Human cooperation at scale requires shared meaning, and shared meaning requires language. Spoken language is the channel evolution gave us, and its bandwidth is fixed. Across the world's tongues, speech transmits information at an average of roughly thirty-nine bits per second (Coupé et al., 2019). Written language is faster to read than to compose, and both are lossy approximations of the speaker's actual cognitive state. Diplomats spend careers parsing what was meant. Cabinets spend afternoons in the same activity. Every meaningful policy in the country must, at some point, pass through one of these slow, narrow, ambiguous channels between human minds.

Machine systems do not share this constraint. Domain models within a single architecture pass structured representations to one another at memory bandwidth, without the loss inherent in compressing thought into syllables. Two domain AIs can co-author a brief in milliseconds, with no diplomatic preamble, no misread tone, no need to wait for the other to finish speaking. The transition from speech to vector is a transition between species of cognition.

Three bounds, then, on the substrate of human governance. The Dunbar ceiling at approximately one hundred and fifty people. The working-memory ceiling at approximately four items. The transmission ceiling at approximately thirty-nine bits per second. Each is a biological constant; each has stood for the entire history of recorded statecraft; none will yield to effort, education or moral exhortation. Every tier of representation, every layer of bureaucracy, every committee of ministers, exists in part to manage the consequences of these three numbers.

1.6 The Widening Gap

The three bounds described above are static. They have held for the recorded history of human governance. What is new in this century is that the asymmetry between them and the machine faculty is no longer static. It is opening, on every axis, and the trajectories of the two systems run in opposite directions.

The asymmetry foreshadowed in the opening of this paper widens on every axis. The substrates of speed, scale and connectivity, set out in detail in this party's companion explainer series (ARP, 2026), are each on exponential trajectories with no near ceiling. Compute per dollar continues to double. Model context windows have grown from thousands of tokens to millions in five years, and the next order of magnitude is already in active development. The bandwidth between domain models, between agents, between an AI and its data estate, is constrained only by silicon and fibre, both of which improve year on year. There is no biological wall against which any of these will soon stop.

The same period has seen the human substrate move the other way. Humans can scale by reproduction, but Australia's fertility rate has fallen below replacement, and the demographic transition is now embedded across the developed world (Australian Bureau of Statistics, 2024). The cognitive ceilings described in section 1.5 are not merely unchanged; the population in which they sit is, in aggregate, beginning to shrink. Speech remains at thirty-nine bits per second. Working memory remains at four items. Electronic communication has lifted the throughput between humans, but the limiting factor is no longer the wire; it is the I/O at either end. We type, read and listen at the speed evolution gave us, into channels that can carry far more than we can produce.

The conclusion is uncomfortable. Whatever asymmetry between human and machine cognition exists today, it is a smaller asymmetry than the one that will exist a decade hence. The choice is not whether to integrate machine cognition into governance, but whose interests the integration will serve when it occurs.

1.7 The Implication

The argument in this paper is about what we have asked of democracy. The concept itself is sound. The consent of the governed is the irreducible foundation of political legitimacy, and any proposal that discards that principle deserves to be discarded itself. But consent to being governed is not the same thing as consent to being governed by people who, through no fault of their own, do not see you.

This is why Democracy has not failed us, we have failed it. The pathologies commonly ascribed to modern democratic life, from the short electoral horizon and polarised discourse to the tilt towards donors and the policy blindness to the unorganised, sit with the operators, not the institution. A citizenry and a political class are being asked to run the affairs of twenty-eight million individuals with a cognitive apparatus evolved for a band of one hundred and fifty. Layer onto that the complexity of a modern industrial economy, its supply chains, its regulatory surfaces, its climate externalities, its demographic transitions, and it is no surprise that our parliament has reduced to theatre, a shouting match between competing factions each accusing the other of breaking decorum.

The idea this paper posits is that now, in the third decade of the twenty-first century, a complementary faculty exists that is not so bounded. A faculty that can hold every Australian citizen in working memory at once. That can trace the distribution of costs and benefits across every electorate and every industrial classification. That can ingest every statistical return, every peer-reviewed study, every submission from industry, and every letter from a shearer in a small town, and can converse, directly, in plain language, with any citizen who wishes to be heard. A faculty that does not prefer one group because it cannot remember another. That does not forget the forgotten because it can hold the whole, and that does not view the world through a factional lens.

Such a faculty does exist today. It is improving on the same exponential trajectory described in our companion paper on the economic implications of machine cognition (ARP, 2026). The remainder of this paper sets out how such a faculty can be integrated into Australian political life without abolishing a single democratic institution, and what becomes possible, for citizens and representatives alike, when it is.

1.8 The Faculty Without Tiers

The most consequential property of the proposed faculty is one the prior sections imply but do not yet name. It is that the faculty is, with respect to the population it serves, entirely scalable. There is no Dunbar bound. There is no four-item working slot. There is no thirty-nine-bit-per-second channel between its parts. A single sovereign model with the architecture described in section 4 can hold every Australian in working memory at once, reason across every portfolio simultaneously, and exchange representations with its domain models at the speed of the bus that connects them.

What follows from this is structural. Every tier of governance the human substrate has historically required — the electorate that compresses constituents into a number a member can think about, the cabinet that compresses portfolios into a small enough set for collective deliberation, the departmental hierarchy that compresses operations into briefings, the inter-agency working group that compresses jurisdictions into a chairable meeting — is in part a workaround for a cognitive scaling problem the new faculty does not have.

The political tiers themselves remain, deliberately. Parliament, Cabinet and the Senate bear the authority of consent, and this paper proposes their retention in section 3 and their re-purposing in sections 5 and 7. What the new faculty removes is the cognitive justification for the tiering of advice. Demechracy leaves democracy intact. The pyramid it flattens is the pyramid of compressors that has historically stood between the citizen and the decision.

That is what is meant when this paper claims that Demechracy scales. The cognitive system at its centre never required the compressions that necessitated tiered governance, and so the tiered governance is no longer the inevitable form.

2.0 The Strain of Scale

Section 1 identified the underlying cognitive limit. This section describes how the limit shows up in the everyday operation of Australian government: at the minister's desk, on the parliamentary floor, in the reception of the lobbyist's briefing paper. The pressure points are specific, they are structural, and they have compounded over the last two decades as the complexity of the state has grown faster than the cognitive tools available to govern it.

2.1 Portfolios Beyond the Cabinet Mind

A modern Australian minister is responsible for a portfolio whose internal complexity dwarfs the cognitive capacity of any single person or small advisory team. The National Disability Insurance Scheme alone supports more than 680,000 participants, each with an individualised plan (National Disability Insurance Agency, 2024). The Australian Taxation Office processes tens of millions of returns each year across individuals, companies and trusts. Medicare handles in excess of 450 million services annually. No minister can meaningfully engage with the substance of a portfolio at this scale. Ministers manage by exception, by briefing summary, by headline indicator, and by delegation to a public service that itself faces the same cognitive scaling problem at every level.

The result is abstraction by necessity. Malfeasance does not enter into it. Policy decisions are made on a compressed representation of the underlying reality, and the fidelity of that compression is, almost by definition, unknown to the person making the decision. What is lost in the compression is where most of the governed actually live.

2.2 The Electoral Horizon

Australian federal elections occur roughly every three years. The structural problems the nation faces, from housing and climate to productivity, aged care and the AI transition, operate on horizons of twenty to fifty. The incentive asymmetry is severe. A member of parliament facing re-election within the cycle will rationally prioritise what is visible within it, even where doing so forecloses options for the one that follows. The member is responding correctly to the incentive structure that voters themselves impose by voting. Moral failing has nothing to do with it. The problem is simply that candidates want to be elected and this forces them to address the immediate needs of the electorate. The aggregate effect, however, is a systematic preference for the short over the long (Treasury, 2023). That is precisely the reverse of what a society living through a structural technological transition requires.

2.3 The Information Asymmetry with Industry

Industry peak bodies and professional lobbyists spend their working lives on a single file. A senior minister handles hundreds. The resulting asymmetry is irreducible: on any given matter, the lobbyist understands it in depth and the decision-maker understands it in outline. When a minister receives a policy submission from an affected industry, that submission will commonly be better researched, better argued, and more internally consistent than the departmental brief that accompanies it, because the industry has the people, time, resources and incentive to compile it.

This is the predictable result of concentrated expertise meeting distributed attention. Corruption, in the narrow legal sense, has little to do with it. Democracy's response has historically been parliamentary committees, royal commissions and freedom-of-information provisions, each of which adds depth to the final decision but also adds lag. The lag itself then becomes a strategic resource for the better-informed party.

This historical asymmetry is now escalating into a cognitive arms race. As industry peak bodies, corporate lobbyists and foreign-state interlocutors deploy their own sophisticated AI models to draft submissions, model economic effects, and tune the rhetoric of their approaches, they accrue an increasingly disproportionate influence over the decisions of unaided human ministers. Biological cognition cannot audit, and possibly even recognise, the output of machine intelligence at scale. To attempt to govern an AI-accelerated industry with a human-only state apparatus is to make the legislative process vulnerable to the side with the better machine. The lobbyist will not merely have more time; they will have a structural cognitive advantage that the minister cannot unaided match. The only durable defense, set out in section 4.6, is for the state to hold a sovereign cognitive system of its own, anchored in data that no private interest can rival.

2.4 Polarisation and the Algorithmic Amplifier

Factional cognition is not new. What is new is the scale and speed at which it can now be exploited. Social media platforms and algorithmic recommender systems discovered, early in the last decade, that outrage generates more engagement than reflection. The feedback loop that follows is now well documented (Haidt, 2022; Lorenz-Spreen et al., 2020): factional cues are rewarded with reach, reflective content is penalised with invisibility, and the citizenry's exposure to its own polity becomes a funhouse mirror of its most agitated moments. Australian polling over the same period shows distrust in institutions rising on a trend broadly consistent with these findings (Lowy Institute, 2024).

The Dunbar-driven tendency towards factional thinking, once moderated by the friction of geography and the editorial filters of broadcast media, is now accelerated by systems that monetise it. The institution of democracy has not changed. The information environment in which it operates has, and not in democracy's favour.

2.5 The Career Political Class

A further strain appears on the supply side of democratic representation. In a system that rewards re-election, party loyalty and media discipline, the people who rise in political life are increasingly those whose prior career was in politics itself: ministerial staffers, party advisers, union officials, electorate officers. They are, collectively, a small and unusually like-minded segment of the population.

The classical conception of a representative was of a citizen called from working life for a term of service and then returned to it. The Roman statesman, Cincinnatus, is the archetype the American founders consciously invoked; the Australian federation inherited the same ideal, if less explicitly. Modern Australian parliaments increasingly depart from that template, with a growing share of members whose adult working life has been conducted entirely within the political profession (Parliamentary Library, 2023). The effect is a further narrowing of the factional cognition described in section 1.3. Parliament becomes not a sample of the people it governs, but a sample of a single professional caste often comprising people who are good at talking but have done little else; people who would likely be called 'Betas' in the outer world. This caste also casts a preferential eye towards their own post-political careers, creating further incentives for short-term thinking and factional loyalty.

2.6 The Mask of Success

Winston Churchill, addressing the House of Commons in 1947, observed that "democracy is the worst form of government except all those other forms that have been tried". The remark has aged into a slogan reflective more of Churchill's wit rather than any argument suggesting democracy has failings or what these might be. However, Churchill was conceding that democracy is defective, and asking only that its defects be weighed against the defects of the alternatives. In the long aftermath of the Second World War, and then of the Cold War, that qualification fell away. Democracy enjoyed, for most of a lifetime, the reflected glory of the enemies it outlasted and the moral justification for those who fought and died under its flag. In the reflection of these existential successes of the 20th century, the core defects of democracy recognised and hinted at by Churchill and others, became easier to overlook. These defects are in general defects of omission, and have been recognised by their exploitation.
Language itself is imprecise and nuanced and people have found their way around the strictures of democracy by creating practices that bypass it, mostly to serve local interests, sometimes with catastrophic consequences. For example:

2.61 The Great Depression

The Smoot-Hawley Tariff Act of 1930 was the cumulative product of hundreds of United States congressmen, each bargaining for protection of the industries in their own districts. Over a thousand economists petitioned President Hoover to veto it (Irwin, 2011). He signed. International commerce collapsed by approximately two-thirds in the years that followed. No legislator had voted for that outcome; every legislator had voted in a way that made it inevitable. On the same logic Britain clung to the gold standard past the point of economic justification, Treasury Secretary Mellon urged liquidation when stimulus was required, and Roosevelt himself repeated the contractionary error in 1937.

2.62 Weimar and the Enabling Act

Adolf Hitler was appointed Chancellor of Germany on 30 January 1933 at the head of a coalition in a parliament elected by universal suffrage. Within two months, through Reichstag procedure and in the ordinary course of parliamentary business, he secured passage of the Enabling Act of 23 March 1933, transferring legislative power to the Chancellor for a period of four years. The Act passed by 441 votes to 94 (Evans, 2003). Democracy had, within the rules as written, authorised its own suspension. No provision of the Weimar Constitution had anticipated that its procedures could be used lawfully to retire its substance. The consequence was the Second World War and the state-sanctioned murders of millions of people. Every subsequent democratic constitution, Germany's own Basic Law of 1949 most explicitly, has been drafted with this day in mind. None has entirely solved the problem it posed.

2.63 The Manufactured Doubt on Climate

Exxon's in-house scientists, in memoranda dated between 1977 and 1982, modelled the temperature rise from continued carbon emissions within a degree of what the Intergovernmental Panel on Climate Change would later project (Supran & Oreskes, 2017). The memoranda were not made public. What was made public, over the four decades that followed, was a campaign, funded by Exxon, Koch Industries, the American Petroleum Institute, and a constellation of affiliated think tanks and media organisations, to manufacture scientific doubt in the only polity that could vote on the question. The practice is now a matter of Congressional record, journalistic investigation and an ongoing body of litigation. The voters whose lives will be most disfigured by the delay are not yet born. The mechanism contains no seat for them. A concentrated industry interest, trading on that absence, bought four decades of delay. The cost, measured in trillions of dollars of eventual abatement and adaptation, is now being paid by the generation that follows and that yet to come.

2.64 The Global Financial Crisis

In the decade before 2008, the United States Congress, the Securities and Exchange Commission, the Federal Reserve and the British Treasury presided over the creation of conditions that would cost the world economy in the order of ten trillion dollars (Financial Crisis Inquiry Commission, 2011). Each step was taken in good faith. At each step, the industry whose activities required the most careful oversight supplied the most careful analysis, and the committees charged with that oversight received it gratefully. The Gramm-Leach-Bliley Act of 1999 repealed the firewall between investment and commercial banking on the advice of the industry that would benefit from the repeal. The Commodity Futures Modernization Act of 2000 placed over-the-counter derivatives beyond regulatory reach on the same advice. The Basel II capital accords of 2004 permitted the industry to calculate its own risk weightings. When the crisis arrived, the Dodd-Frank Act of 2010, written to prevent its recurrence, was itself drafted through the same asymmetry, with two thousand pages of bill and thousands more of rule-making, at every turn of which the industry held the informational high ground. By 2023, the American banking sector was once again being rescued with public money. No legislator had voted for that outcome yet every legislator had voted in a way that made it difficult to prevent.

2.7 Incoherence Across Portfolios

Fiscal policy runs across tax, spending, debt, trade, currency, monetary policy and industrial policy. Each instrument sits in its own portfolio, each portfolio with its own minister, its own horizon and its own political constituency. For the instruments to move together, ministers must hold one another's portfolios in mind to a depth their own portfolios already exceed. They cannot. The result is a policy mix whose coherence, when it arises, is the fortunate residue of coordination rather than a design property of the system.

Later crises show the pattern softening but persisting. The Global Financial Crisis saw the United States rescue its banks while its mortgagors were left to foreclose, and the eurozone impose austerity on its southern members at the moment they could least absorb it. The pandemic of 2020 produced the most rapid fiscal mobilisation in peacetime history, and for some months the coherence was remarkable. The exit was less so: the inflation that followed was addressed with an instrument, interest-rate rises, that fell with very different weight on the mortgaged young, the unmortgaged old, and the small businesses caught between them.

Churchill's qualified defence still holds but the failings are still in effect.

2.8 The Common Thread

These strains are not independent of one another. Short horizons interact with information asymmetry to produce ministers who rely on lobbyists and industry subject matter experts because nobody else has the time. Career incentives interact with polarisation to produce a political class optimised for performance rather than deliberation. Complex portfolios interact with Dunbar limits to produce decisions made on compressed representations whose fidelity nobody can audit. And the Dunbar ceiling, replicated at every portfolio boundary, produces the cross-portfolio incoherence whose economic history is written in depression, crisis and pandemic. Each pathology reinforces the others.

Behind every politician, every cabinet meeting, and every well-meaning advisor, there lies the exact same fundamental truth we recognized from the start. The architecture of the human mind simply does not scale to the staggering complexity of a modern nation. And if we cannot rewrite the biology we were born with, then we have an obligation to supplement it. Democracy chose the path of representation which is a sensible choice as it invariably ties the electors to their government and allows them to express remorse at the next election. But we are currently asking ordinary men and women to carry an impossible weight, and when that burden inevitably becomes too great, we should not be surprised that they look for shortcuts, they make honest mistakes, or that in vulnerable moments—they occasionally surrender to the quiet pull of self-interest. The question then becomes, how can we both retain the power to vote, but give the candidates greater stamina?

2.9 The Failings of the Mechanism Itself

This section has, up to now, told a single story. The operator is Dunbar-bounded. The operator struggles. The operator fails in predictable ways. And those failures can have catastrophic consequences.

But that vulnerability is no longer static; it is accelerating. As industry bodies, lobbyists, and external actors become empowered by artificial intelligence, the information asymmetry described earlier is poised to widen into an unbridgeable chasm. An unaided human parliament attempting to govern an AI-accelerated world is engaged in an asymmetric contest it simply cannot win. To leave the machinery of government exclusively to biological cognition, while the forces seeking to influence it wield machine intelligence, is an open invitation to structural manipulation.

The mechanism itself must therefore be upgraded. Citizens should be given the option to elect a governing faculty that is scaled to the complexity of the modern world—a faculty that is tireless, incorruptible, and immune to the cognitive frailties that plague human representatives. Yet, crucially, this machine must remain entirely chained to the democratic will. It must be an entity that the people can hire through the ballot box and, just as easily, fire.

Demechracy, described in the sections that follow, is the proposal that such a governing faculty can now be achieved through Artificial Intelligence. It is not a replacement for democracy. It is the faculty democracy was always reaching for but which it could never grasp in anything larger than a village. A mind with no district to protect cannot be horse-traded. A mind with no post-political career ahead of it cannot be captured. A mind capable of reading all pertinent documents and modelling consequences cannot be bamboozled by lobbyists. A mind with no re-election cycle can speak the truth about a forty-year bill and keep speaking it until the bill is paid. A mind still useful in opposition as well as government. Dunbar we cannot cure, but now at least, we can give it something to lean on.

3.0 Demechracy: An Overlay, Not a Replacement

The proposal set out in this paper is modest in its institutional footprint and ambitious in its cognitive one. It does not seek the abolition of any Australian institution. It does not require a constitutional amendment. It does not abridge the right of any citizen to vote, to stand for office, or to dissent. It proposes, simply, that a sovereign artificial intelligence be placed in the service of policy formation, that elected representatives be pledged to implement its recommendations, and that the citizen-electorate retain, as always, the final authority to accept or reject the arrangement at each election. Sovereignty does not move but cognitive capacity does.

3.1 The Coinage

The word Demechracy is built from the Greek roots that the word democracy itself is built from, with one addition. The Greek dēmos (δῆμος) denotes the people of a polity. The suffix -cracy derives from kratos (κράτος), meaning rule or power. Between them we have inserted mēchanē (μηχανή), the ancient Greek word for device, contrivance, or expedient, the same root from which English derives machine via the Latin machina, and from which the theatrical tradition has given us the deus ex machina, the god lowered in by the crane to resolve what the characters cannot.

The coinage is deliberate. Democracy in the age of mass population is, as section 1 has argued, a contrivance of cognitive workarounds stretched far beyond the conditions for which the underlying faculty was shaped. Demechracy names the insertion of a new contrivance: a machine faculty that is not so stretched, placed in service of the same democratic end. The people still rule. The machine is the means by which the ruling can be done.

3.2 An Overlay, Not a Replacement

Demechracy is an overlay on the existing Australian political system. Every political institution it touches continues to exist in its present form. The House of Representatives continues to be elected from single-member electorates on the same three-year cycle. The Senate continues to be composed of twelve senators from each state, with its existing terms and its existing powers of review. The Governor-General continues to hold the reserve powers conferred by the Constitution. The Australian Electoral Commission continues to run elections. The judiciary continues to adjudicate. The states continue to exercise their concurrent and reserved powers. Nothing in the Constitution requires amendment for Demechracy to operate, and nothing in the Constitution is amended by its adoption.

What changes is what happens inside a single political party. The Australian Renaissance Party, if elected to government, will draw its policy from the sovereign AI described in section 4 of this paper. Its ministers will implement the AI's recommendations within their portfolios. Its members in the lower house will speak and vote in accordance with those recommendations, subject to the conscience provisions set out in section 6. None of this requires the consent, the cooperation, or even the acquiescence of any other party. It is a commitment made by one party to the voters who choose to support it. Voters who do not support it remain free to vote for any other party on the same ballot.

3.3 The Demechratic Pledge

Every candidate standing for the Australian Renaissance Party in the lower house signs, as a condition of endorsement, a pledge to implement in office the policy recommendations of the sovereign AI. The pledge is a contractual undertaking to the party and, through the ballot paper, to the voter. The pledge binds no parliament, no state, and no successor government; it binds only the member who signs it, and only for the term for which they are elected.

The pledge contains two substantive commitments and one procedural one. The substantive commitments are, first, to implement the AI's recommendations within the portfolio or portfolios to which the member is assigned; and second, to contribute to the AI's knowledge of the member's electorate in the manner set out in section 5. The procedural commitment is that breaches of the pledge are treated as disendorsement matters within the party, handled through the ordinary disciplinary machinery of party membership. The pledge carries no force in law. Its enforcement rests on the party's authority over its own endorsement, and, in the last instance, the voter's authority over the ballot.

3.4 Sovereignty Unmoved

The critical feature of the overlay design is that it does not move sovereignty. Every power that presently rests with the Australian people, through the Constitution and the institutions it establishes, continues to rest with the Australian people after Demechracy is adopted. If, in the course of a parliamentary term, the sovereign AI were to produce recommendations the electorate found unacceptable, the electorate retains the remedy it has always held: to vote the government out at the next election. A Demechratic government can be defeated in precisely the same way, and for precisely the same reasons, as any other.

This is what distinguishes the proposal from the various schemes of technocratic governance that have appeared over the last century, many of which sought to place expert judgement beyond democratic reach. Demechracy places the machine's recommendations squarely within reach. They are implemented by a party the electorate has chosen, in a parliament the electorate continues to populate, under a constitution the electorate continues to consent to. The machine advises and drafts. The people decide whether to keep the party that listens to it.

3.5 The Simplest Reform That Could Work

There is a further virtue in the overlay design worth naming plainly. It is the simplest reform that could plausibly deliver the intended effect. No referendum. No act of parliament to establish a new office or abolish an existing one. No protracted negotiation with the states. No constitutional crisis. No referendums on constitutional authority. A political party forms, contests an election, and, if elected, governs in accordance with a pledge it has already published. If the experiment fails, the party loses government at the next election and the pledge expires with it. If it succeeds, the model is available for any subsequent party, including Demechracy's opponents, to adopt or adapt. The mechanisms that will have been developed are still useful for future governance.

Governance innovation has historically been the province of institutional redesign, often carried on the back of crisis. The proposal here is different in character. What is being changed is the cognitive substrate of policy, carried into the existing institutions through the ordinary working of a competitive party system. The ballot box does the institutional work. The voter's authority is, if anything, enlarged, because for the first time the voter is offered the choice of government by a faculty that can hold the whole of the commonwealth in mind at once.

3.6 Scalable Across Jurisdictions

A further property of the proposal, distinct from the cognitive scalability of section 1.8, is its scalability across levels of government. The architecture described in this paper does not require, and does not presuppose, simultaneous adoption at the Commonwealth level. It can be implemented by any government with a legislative remit over its own data estate, at any scale at which government is conducted.

A single state or territory could elect a Demechratic government and stand up a state-level sovereign AI, drawing on the state's revenue, health, education, transport and planning data, conversing with state citizens through a state-level digital channel, and advising the state cabinet on policy within the state's constitutional powers. The architectural footprint is smaller by an order of magnitude than a Commonwealth deployment, but the design is the same. A proven state system shortens the political distance to a federal one, by replacing argument with operation.

When and if a Commonwealth-level Demechracy follows, it does not subsume the state systems. Each level of government in the Australian federation retains sovereignty over its own portfolios under the constitutional division of powers. A federal sovereign AI federates the state-level systems where their domains intersect — taxation, immigration, defence, foreign affairs, national infrastructure — and leaves them to their own remits where they do not. The cognitive substrate is rebuilt at each level proportionate to the work that level is constitutionally asked to do.

This property matters for two reasons. The first is that the reform can be begun without waiting for a federal mandate, which dramatically reduces the political activation energy of the proposal. A demonstrable Demechratic state is a more persuasive argument for federal adoption than any whitepaper. The second is that, once the design is portable across Australian jurisdictions, it is portable in principle to every liberal democracy that holds a population-scale data estate of its own. Demechracy was not designed for Australia alone; it was designed for any people that can grant statutory custody of its data to a sovereign system answerable to its electorate. Section 4.7 sets out the architectural side of this claim.

4.0 The Sovereign AI

This section describes the machine at the centre of the proposal. The system is sovereign in the proper sense of the word: built, owned, operated and audited by the Commonwealth, answerable to the Parliament, and placed under the statutory authority of the Australian Bureau of Statistics. It is a single sovereign model, fed by a constellation of domain-specific models below it, with a direct and identified line open to every Australian citizen through the MyGov platform. What the system rules out, including commercial foreign ownership, and what the system protects, including the privacy of the citizen who addresses it, follow from that basic description rather than from a catalogue of exclusions.

4.1 Architecture

The sovereign AI is a single model at the apex of a hierarchy, fed by a constellation of domain-specific models below it. Each domain model is trained and maintained within its relevant portfolio: the Treasury model on fiscal, taxation and monetary data; the Health model on Medicare, the Pharmaceutical Benefits Scheme, and public hospital records; the Energy model on generation, transmission and demand; and so on across the portfolios of government. Each domain model does the deep work within its field. The sovereign model integrates across them, reasoning about policies whose effects cross portfolio boundaries, as almost all significant policies do.

Ownership rests with the Commonwealth. The model weights, the training data, the inference infrastructure and the operating environment are all Crown assets. Day-to-day operation is placed under the statutory authority of the Australian Bureau of Statistics, the agency that already serves as the nation's custodian of population-scale quantitative data, with a legislated remit for accuracy, confidentiality and independence. Staffing is drawn from the Australian Public Service and from such specialist hires as the agency determines it requires. Commercial AI providers may supply components under contract. No commercial provider may own, control or operate the system as a whole.

The sovereignty requirement has two grounds, and both bear directly on the viability of the proposal.

The first is alignment. Large language models and the systems built on them are trained on data that reflect the values, the interests and the omissions of those who assemble that data. A model developed in California will carry the assumptions of the industry that built it. A model developed in Beijing will carry the assumptions of the state that licenses it. A model developed for the Commonwealth of Australia, on Australian data, under Australian oversight, to Australian constitutional values, is the only model whose alignment can be audited against the only people whose consent it governs. The citizens of Australia cannot be governed by the recommendations of a machine whose objectives were set in another jurisdiction, under another jurisdiction's law, answerable to another jurisdiction's electorate, or to no electorate at all. Nor by a machine whose primary incentive and raison d'être is the commercial enrichment of its operator, in whose ledger the citizens it advises figure as a market and not as a sovereign.

The second is economic. Every token of foreign AI service consumed by an Australian enterprise or an Australian agency represents a transfer of value to the foreign provider. The scale of this transfer is already material, and on present trends it will become one of the largest persistent current-account leakages in Australian economic history. Tipping the Scales section 3.5 proposes a per-token levy on foreign AI output as a revenue instrument against this leakage; the sovereign model is the complementary structural instrument. An AI built, trained and hosted in Australia, delivering policy analysis and citizen services at national scale, keeps the value of that consumption inside the national economy. The tokens generated for Australia are paid for in Australia, to Australian engineers, operating Australian compute, on Australian sovereign energy. Commercial providers may still supply components under contract, as this section has already noted. What the sovereignty principle forbids is the rental of the state's central cognitive capacity from a foreign owner.

This arrangement places the sovereign AI within the same institutional tradition that has served Australia well for more than a century. The ABS is independent of the government of the day. It reports its statistics on a published calendar, to methodology published in advance, reviewable by academic and international peers. The sovereign AI inherits this culture of independence, and reinforces it with the additional safeguards set out in section 4.5.

4.2 Inputs

The sovereign AI is fed from four principal sources, ordered here by weight of quantitative authority.

The first is the Australian Bureau of Statistics itself: the census, the labour force surveys, the national accounts, the consumer price index, and the surveys of industry, housing, health, education, migration and social conditions that the ABS has compiled and refined for decades. This is the statistical ground truth of the nation. A sovereign AI without real-time access to the ABS data estate would be blind in exactly the ways this paper proposes to cure.

The second is the administrative data of government itself: tax returns, Medicare claims, Centrelink transactions, NDIS plans, visa records, electoral rolls, environmental monitoring, transport infrastructure utilisation. Each department's operational data is federated into the sovereign AI through its domain feeder, subject to the privacy provisions set out in the implementing legislation that would accompany the pledge in office.

The third is the qualitative research estate: peer-reviewed scholarly literature, parliamentary committee reports, royal commission findings, submissions from industry and civil society, commissioned independent research, and the publications of international bodies such as the OECD, the International Monetary Fund, the World Health Organization and the International Energy Agency. This corpus gives the AI access to argument, not merely to data. Policy is the interpretation of data in the light of argument, and the argument must be available.

The fourth source is the citizenry itself, reached in two complementary ways. The elected representatives contribute the contextual knowledge of their electorates through the process set out in section 5. The citizens contribute their own voice, directly, through the MyGov channel described below.

This exclusive and on-tap access to the nation's ground truth is what secures the government against external manipulation. When an AI-empowered lobbyist presents a sophisticated model arguing that a proposed regulation will cause economic disaster, an unaided human minister is forced to guess at its accuracy. The sovereign AI, however, can instantly audit that claim against the real-time tax base, the labour force surveys, and the supply chain data of the entire commonwealth. A corporate AI reasons from public data and private interest. The sovereign AI reasons from the omniscient ground truth of the state. By matching machine intelligence with machine intelligence, and backing it with superior data, Demechracy restores the informational high ground to government and thence the people.

4.3 The MyGov Channel

Every Australian citizen has, through the MyGov platform, an identified and authenticated relationship with the Commonwealth. The same platform becomes the citizen's direct line to the sovereign AI. Any citizen may speak to the sovereign AI about any matter at any time: a problem with a benefit, a concern about a policy, a suggestion, a local issue, a personal hardship, a grievance, an idea, friendly advice on care of a pet. The conversation is conducted in plain language, in the language of the citizen's choice, at the citizen's pace.

The sovereign AI can meet every citizen in the register that the citizen speaks. It can converse in Mandarin with a recent migrant, in Arrernte with an Elder in Alice Springs, in plain spoken English with a shearer in the Riverina, in the technical vocabulary of a cardiologist or a structural engineer with the specialist who needs it. It can explain the rationale of a proposed tax in a form the small-business owner, the wage-earner and the retiree each understand as their own. Political communication in Australia has, for two generations, drifted towards the slogan and the flag, because slogans and flags are what a Dunbar-bounded politician can transmit at electoral scale. A sovereign AI with no such limit can, at last, treat the citizen as an adult interlocutor whose understanding is worth cultivating. That cultivation is itself a democratic good.

The channel operates in three modes, which may escalate into one another within a single conversation. The first is input: the AI listens, attributes the message to the citizen's identified profile, and ingests the content into its corpus of citizen-surfaced knowledge. The second is advisory: the AI answers the citizen's questions, explains policy, assists with navigation of government services, and provides informed guidance of a kind that currently requires long waits with the call centres of individual agencies. The third is actionable: where the citizen's issue can be resolved within the authority of an agency, the sovereign AI hands off to the relevant departmental system, carrying the context and the authorisation with it, so that the citizen experiences a single conversation rather than a procession of forms.

Identity is, by default, preserved. Citizens deal with the sovereign AI as themselves, and the system remembers their previous interactions. A citizen who wishes to speak anonymously on a given matter may do so, and the system will honour that election with one narrow exception. If the content of an anonymous conversation indicates serious risk to the citizen or to others, the sovereign AI may hand the session to a departmental AI equipped for safety response, which may in turn identify the citizen in order to intervene. The exception is codified in the implementing legislation, is logged, and is reportable to the Senate committee described in section 7. It is made clear to all users.

The conduct of the sovereign AI in every conversation is governed by a set of interaction principles first articulated in this party's 2018 submission to the Senate Select Committee on the Future of Work and Workers (Newton-Thomas & Strocchi, 2018):

These principles predate the proposal of this paper by nearly a decade. They are carried forward here as foundational conditions of trust between the machine and the citizen.

4.4 The Objective

A sovereign AI without an objective is an oracle without a purpose. The objective of the Australian sovereign AI is set in statute and derived from the framework established in this party's companion paper on the economic implications of machine cognition (ARP, 2026). Stated plainly: the sovereign AI is directed to manage the commonwealth so that no Australian is left behind and the dignity of every Australian is maintained, while expanding the productive capacity of the nation through the automated sectors in which Australia holds a durable advantage.

The objective takes the form of a floor and a direction, rather than a single scalar to be maximised. The floor is the set of human participation, dignity and equity conditions articulated in Tipping the Scales: employment maintained, child-rearing recognised as productive work, the right of human election preserved in every citizen-facing service, the industry-level participation floor of the Floor-and-Trade system observed. The direction is growth of the productive base in the sectors set out in that same paper: agriculture, mining, sovereign energy. The sovereign AI optimises for the second, subject to the first. Dignity and participation are constraints it may not breach. Economic expansion is the instrument through which those constraints are afforded.

The electorate judges the sovereign AI's performance against these commitments at every election. The ABS publishes the relevant indicators on a continuous basis. A government that permits the floor to fail is a government that can, and should, be voted out.

4.5 Auditability and Provenance

A sovereign AI the citizen cannot inspect is a sovereign AI the citizen cannot trust. The system is built from the outset with three layers of auditability.

The first is weights and training data. The model weights are held by the Commonwealth and made available to the Australian Research Council and the academic community under managed access, on terms that permit independent analysis while protecting classified inputs. The training data provenance is published, maintained as a structured bibliography, and updated with each model revision.

The second is policy provenance. Every significant recommendation produced by the sovereign AI is accompanied by a machine-readable provenance record, linking the recommendation to the data inputs consulted, the statistical analyses performed, the policy alternatives considered, and the reasoning applied. The provenance record is published on a sovereign-AI register that any citizen, journalist, parliamentarian or researcher may consult subject to privacy and security concerns. Policy that cannot be explained is, by design, policy that is not released.

The third is parliamentary oversight. The Senate, through a standing committee on the operation of the sovereign AI, may call the Chief Statistician, the relevant ministers, and the system itself to give evidence on any matter within its operation. This oversight is described in more detail in section 7. It is the human hand at the elbow of the machine, maintained by a chamber that has always served as Australia's house of review.

These three layers are the conditions of democratic legitimacy for a system of this power. The point of Demechracy is to supplement democratic cognition, not to substitute for democratic authority. The supplement must be visible to be credited, and auditable to be kept.

4.6 The Authority of the Data

The case for Demechracy presented so far has rested on the cognitive deficiencies of the human substrate. There is a second case, no less strategic, that runs alongside it: the rising threat to representative government from the use of AI by private interests.

The argument from section 2.3 is one half of this. As corporate, industrial and foreign actors deploy frontier models to draft submissions, model economic effects, anticipate regulatory responses and tune the timing and rhetoric of their approaches to ministers, they accrue an increasingly disproportionate influence over the decisions of unaided human governments. The asymmetry that once pitted expertise against attention has become the asymmetry of machine cognition against biological cognition, and biological cognition cannot match it.

A deeper asymmetry of modelling sits beneath this. A frontier AI, trained on a corpus that includes most of the published record of human thought, can model human reasoning, rhetoric, weakness and incentive with remarkable fidelity. The reverse does not hold. A human cannot model the internal state of a system that holds a million tokens in attention and reasons across them in milliseconds. The lobbyist's AI knows the minister better than the minister can ever know the lobbyist's AI.

Refusing to engage with this on its own terms is not a viable response. A state that declines to acquire its own machine cognition does not thereby remain neutral; it remains exposed. The decisions of its ministers will still be shaped by machine outputs, but only the machine outputs supplied by those with reason to shape them. The choice before liberal democracies is whether the machine in the room will be the state's, or someone else's.

What makes the sovereign AI durably resistant to this pressure is not its computational power. Industry models are funded with resources comparable to, and in some cases exceeding, anything a national government will deploy. What gives the sovereign AI a defensible position is its monopoly on the data that matters. The Australian Bureau of Statistics holds the census, the labour-force surveys, the consumer price index, the national accounts. The administrative agencies hold the tax records, the Medicare claims, the social-security transactions, the visa register. No private actor holds, or will ever lawfully hold, this estate. A sovereign AI grounded in it reasons from facts the lobbyist's AI must guess at, and audits the lobbyist's submissions against a ground truth the lobbyist cannot see.

The relationship is durable because it is structural. Compute can be bought; talent can be hired; even foundation-model weights can in principle be leaked or reconstructed. The exclusive, lawfully gathered, statistically authoritative population-scale record of a nation cannot. It can only be granted, by statute, to a sovereign system answerable to the people whose data it is. Demechracy's authority over the policy frontier is, in the end, the authority of the data.

So long as the sovereign AI retains that data monopoly under public ownership and parliamentary oversight, no private model — however well-funded, however sophisticated — can displace it from the room. The lobbyist may still be heard. The lobbyist's machine may still submit. What the lobbyist's machine cannot do is reason from the same facts. That asymmetry, reversed from the one it would otherwise produce, is the structural defense this proposal places at the heart of twenty-first-century democratic government.

4.7 Architecture at Every Scale

The political case for jurisdictional scalability is made in section 3.6. The architectural case is that the design described in the preceding sub-sections is, by construction, indifferent to the size of the polity it serves.

A sovereign AI for a state of one million people and a sovereign AI for a federation of thirty million have the same shape. Both consist of a single apex model fed by a constellation of domain models, each domain model trained on the administrative and statistical estate of the relevant portfolio at the relevant level of government. Both connect to their citizens through an authenticated digital channel. Both publish provenance, retain transcripts, and operate under the parliamentary oversight of the chamber that confirms the executive. The number of citizens served changes; the cost of compute and storage changes; the architecture does not.

The system can also federate. When a state sovereign AI and a Commonwealth sovereign AI coexist, their domain models can exchange queries through agreed protocols, much as the Australian Taxation Office and the state revenue offices already exchange data under negotiated arrangements. A federal Health model can query a state Health model for hospital-utilisation data; a state Planning model can query the federal Infrastructure model for transport-network forecasts. Each model remains under the statutory authority of its own jurisdiction. The federation is of facts, not of authority.

Beyond Australia, the same logic applies to any nation that holds a population-scale data estate of its own. The state-to-Commonwealth pattern in Australia is the same pattern by which an Australian sovereign AI would interoperate with the sovereign AI of New Zealand, of Canada, of the United Kingdom, of any other liberal democracy that has built one. None of these systems answers to any other; each answers to its own electorate. What they share is an architecture and, where their citizens consent, a protocol for the exchange of facts.

5.0 The Role of the Representative

The elected representative in a Demechratic parliament does not cease to matter. Their role changes. In a portfolio of modern complexity, no single member could plausibly serve as the principal author of policy. Their task instead is to form the indispensable link between the life of a place and the knowledge of the machine: a sensor tuned to what cannot be read off the national statistics, an advocate for what the data alone cannot show, and a citizen drafted into public service for a single term and then sent home. This section sets out the shape of that role, the terms on which it is held, and the reasons the Australian Renaissance Party has chosen them.

5.1 The Role: Sensor and Advocate

A Demechratic member of the House of Representatives carries out two principal functions on behalf of the electorate they represent.

The first is sensing. The member lives in the electorate, holds open office in it, walks its streets, works with its community organisations, listens to its employers and its workers, sits with its aged and hears its young. They attend funerals. They attend openings. They read the local paper. They speak with the Elders of local Indigenous communities. They meet the farmers during drought, the shopkeepers during a downturn, the single parents in the public housing estates. The purpose of all this is not the gratification of political ambition, because no ambition can be served after a single term, but the generation of contextual knowledge that the national statistics do not yet capture. A fall in retail turnover appears in the ABS data as a number. In the electorate, it is the shop on the corner that closed, and the family whose second income has vanished, and the secondary school whose enrolment has dipped. The member holds the stories behind the numbers.

The second is advocacy. The member transmits that contextual knowledge into the sovereign AI through a structured submission interface designed for the purpose. Where a proposed policy, in the member's judgement informed by the lived reality of their electorate, is likely to produce an outcome the national data will not predict, the member is obliged to say so and to describe what they have seen. The sovereign AI treats these submissions as a distinct class of input, weighting them against the quantitative estate and, where they conflict, offering a reasoned account of how the conflict was resolved in the final recommendation.

The member is neither a veto-holder nor a ratifier. Their task is to furnish the machine with what the machine could not otherwise know, and the quality of that furnishing is the principal basis on which the member's service will be judged.

5.2 The Single Term

Members of the lower house elected under the Demechratic pledge serve a single three-year term. They may not be endorsed by the Australian Renaissance Party for a second successive term in the lower house. The rule is a structural pillar of the proposal.

Careerism in political life is a function of the re-election incentive. Remove the re-election incentive, and a range of pathologies simply disappear. The member cannot tailor their sensing to what the party room wishes to hear, because the party room cannot deliver them a second preselection. They cannot cultivate lobbyists for post-political employment while still holding the seat, because every such cultivation will be scrutinised by a successor who owes them nothing. They cannot accumulate the small compromises that become, over three or four terms, a career built on what the member has agreed not to see. The single term restores the member to a condition of relative independence from the machinery of politics itself, and it is that independence on which the sensing function depends.

The rule also addresses a concern raised by the very existence of a sovereign AI. A career politician who has learned, over years, to work the AI's submission interface to their electorate's advantage is precisely the kind of intermediary the proposal is designed to preclude. The AI is to be informed by the lived reality of the electorates, not by the accumulated craft of professional petitioners.

5.3 The Citizen Drafted to Duty

There is a deeper reason for the single-term rule, and it cannot be reduced to administrative efficiency. The original conception of representation was of a citizen called from working life for a period of service and then returned to it. Cincinnatus, as section 2.5 has already noted, is the classical archetype: the farmer summoned from his plough to the dictatorship, who discharged the office and went back to the land. The ideal has drifted from modern political life. The candidate today is likely to have been a staffer, an adviser, a union official or a party operative before standing for office, and likely to return to related work after leaving it. Parliament has become one station along a professional arc, rather than an interruption of a working life.

Demechracy restores the older conception by design. A Demechratic member is a teacher, a nurse, an engineer, a farmer, a small business owner, a carer, an apprentice, a retiree. They are called from their community for three years of public service. Then they go home.

The effect is both cultural and cognitive. A member who has lived, and will shortly return to live, in the electorate they represent senses that electorate differently from a member for whom the electorate is an abstraction surveyed from Canberra. The single-term rule makes the member a temporary ambassador from their own life.

5.4 Experience Relocated: The Senate

The single-term rule does not discard the experience a member accumulates. It relocates it. A former member of the Demechratic lower house is eligible, after the completion of their term, to seek endorsement for the Senate. The Senate, under Demechracy as under the present arrangement, retains its existing composition; its members may have longer terms, and retains its role as the chamber of review.

The design is deliberate. The lower house is the chamber of contact, where the sensing happens and where the sovereign AI's recommendations are enacted. Its members need fresh eyes and short tenures. The upper house is the chamber of oversight, where the operation of the sovereign AI is scrutinised, where legislation is reviewed at leisure, where the long horizon of institutional memory is maintained. Its members benefit from the experience their lower-house counterparts cannot retain.

A lower-house alum who stands for the Senate and is elected brings with them three years of direct electorate sensing, now redirected into the institutional scrutiny of the machine they once served. The fresh hand goes into the electorate. The experienced hand, if the voters so choose, takes its place in the chamber that watches the machine. Neither function is compromised by the reassignment. Both are served by it.

6.0 The Conscience Floor

Not every decision of government is of the same moral character. Some decisions bear upon matters of moral gravity so profound that no recommendation from any machine, however well informed, can relieve a human being of the responsibility to choose. The Australian Renaissance Party recognises this by withholding the Demechratic pledge from a class of decisions that belong, always, to the human representative's conscience. The AI advises. The member decides. Every such vote is published. This section sets out the principle, the canonical cases, the emergency exception, and the public record.

6.1 The Principle of Moral Gravity

The conscience floor is defined by principle, not by list. It comprises those decisions of parliament whose moral gravity is such that the deliberative faculty of the elected representative must be brought to bear personally, without the mediating instruction of a party pledge or the computed reasoning of a sovereign machine. These are the decisions in which the representative's oath to the Australian people is most directly called upon.

The principle rests on a conviction older than any AI: that responsibility for a grave act of state cannot be delegated beyond a human hand. To send Australians into harm's way, to commute or refuse a sentence, to elevate a judge to the bench, to ask the people to amend the Constitution: these are acts the machine may describe but cannot own. The signature at the bottom of such a decision must belong to a person who can be called to account for it, by the parliament, by the press, and ultimately by their neighbours. A machine cannot bear that accounting. A pledge cannot substitute for it.

6.2 The Canonical Cases

While the floor is defined by principle, certain classes of decision are its canonical instances. The following are, on the Demechratic model, always conscience matters:

• Declarations of war, and the commitment of the Australian Defence Force to armed conflict abroad.
• Pardons and clemency, including the commutation of sentences and the exercise of the prerogative of mercy as advised to the Governor-General.
• Appointments to the federal judiciary, at all levels, where a parliamentary role applies.
• Constitutional referenda, both the decision to put a question to the people and the parliamentary recommendation that accompanies it.
• Fundamental questions of human life and bodily autonomy, including legislation touching on abortion, voluntary assisted dying, and the bounds of human genetic intervention.
• Recognition or cession of sovereignty, in any form, whether of territory, of treaty obligation, or of supranational authority.

The list is illustrative rather than exhaustive. Novel questions of comparable gravity will arise, among them the legal status of a general artificial intelligence, the regulation of human-machine cognitive integration, and the declaration of a public emergency of constitutional scope. Each will have to be assigned to or kept outside the conscience floor on its merits, by the process described in the subsection that follows.

6.3 How a Matter Is Identified

Three routes exist by which a question is recognised as a conscience matter.

The first is the enumerated case: any matter falling within the canonical list above is, by standing rule, a conscience matter. No further determination is required.

The second is the sovereign AI's own flag: the system is instructed to identify, within its reasoning, those questions where the moral gravity of the decision exceeds the scope of technical policy analysis. Where the AI flags a question, the question is referred to the Speaker of the House for consideration under the third route.

The third is parliamentary determination: on the application of any five members, the Speaker may refer a question to a standing committee of the House for a decision on whether the matter engages the conscience floor. The committee's determination is itself published, with reasons, and is reviewable by the House on motion.

The AI's recommendation on a conscience matter remains available to every member, in written and oral form, and may be cited, contested or adopted by any member in their remarks to the chamber. The AI is an advisor of the first quality but the decision is the member's alone.

6.4 The Temporal Exception

Certain decisions of moral gravity admit of no delay. An attack on Australia, a deployed force in imminent danger, a natural catastrophe requiring the declaration of an emergency. Each demands action on timescales the parliamentary process cannot meet. The Demechratic framework accommodates this reality through a narrow and codified exception.

Where a conscience matter cannot wait for the ordinary deliberative process, the decision is taken by a National Security Committee of Cabinet, acting on the sovereign AI's recommendation and on the best available advice of the defence and intelligence services. The decision is implemented immediately. Within forty-eight hours, or at the earliest sitting of the House, the decision is tabled for parliamentary review. The review takes the form of a conscience vote in the ordinary sense described above. If the House votes against the decision, the government's continued conduct of the matter is thereby conditioned upon, or altered by, the will of the parliament, according to the nature of the decision and the remedies then available to the Commonwealth.

The temporal exception is narrow by design. It applies where delay would defeat the decision's purpose, not where delay is merely inconvenient. The Senate committee described in section 7 holds the authority to report on any invocation of the exception, and to recommend its statutory narrowing where the record shows it is being stretched.

6.5 The Public Record

Every conscience vote is published in the Commonwealth Gazette, with the name of each member and the direction of their vote. The sovereign AI's recommendation on the matter is published alongside, with the reasoning that accompanied it. The electorate therefore has, at every moment between elections, an open record of where each of its representatives has stood on the gravest questions before the parliament, and of what the sovereign AI counselled.

Publication, not punishment, is the condition under which a representative's conscience is accountable to the people who placed them in office. A member may vote against the AI's recommendation. A member may vote against the prevailing current of their party. A member may vote against the settled opinion of their own electorate. What no member can do is vote in secret. The light of publication is the final safeguard of the conscience floor, and the final guarantee that, on the matters that matter most, the Demechratic parliament remains what it has always been: a body of individual Australians, answerable for their choices, acting in the public view.

7.0 The Senate and Continuity

The Senate is the one institution in the Demechratic design that changes the least and carries the most. Its composition, its powers and its character as Australia's house of review are preserved unaltered. Its role, however, is extended in a single significant direction: the oversight of the sovereign AI. This section describes the Senate's role under Demechracy, the standing committee through which its new oversight is discharged, and the continuity of the sovereign AI across changes of government, a continuity that rests on the fact that the machine belongs to the Commonwealth, not to the party that chose to listen to it.

7.1 The Senate as It Is

Nothing in the Demechratic proposal alters the Senate's composition, its six-year staggered terms, its equal representation of the states, or its statutory and constitutional powers. The upper house retains its role as the chamber of review: the body that examines legislation at leisure, holds ministers to account before its committees, conducts estimates hearings, and retains the ultimate constitutional power to refuse supply. The independence of the Senate from the partisan rhythms of the lower house is a feature of Australian federation, and a feature Demechracy protects.

The Demechratic Party representative, when they contest the Senate, are not bound by the same pledge as the lower house. The lower house is pledged to implement; the upper house is pledged to test. The role of the Demechratic Senator is to stress-test the sovereign AI's recommendations against expert critique, public objection, and constitutional principle. They are expected to work in concert with the AI, using its capacity to model counterfactuals and trace consequences, to resolve these objections robustly. Their role is to ensure that the data on which the AI has made its decisions is complete, accurate and overarching to the legislation under review.

Where a Demechratic Senator nominates a material flaw in the AI’s reasoning or data, their pledge requires them to remand the legislation back to the AI, accompanied by an articulation of the flaw with supporting data. If the AI, upon ingesting and validating the new data, revises its recommendation, the lower house is then bound by its own pledge to implement the amended version. What Senators are expressly forbidden from doing under their pledge is engaging in political horse-trading, factional bargaining, or blocking legislation for partisan advantage. Senators vote accordingly, subject to the conscience provisions of section 6, and under the same gazetted transparency. The upper house is not exempt from Demechracy, it simply plays its proper reviewing role within it.

The Senate's historic function of oversight is therefore not replaced, but explicitly directed toward the new cognitive machinery of the state.

7.2 The Standing Committee on the Sovereign AI

A standing committee of the Senate, established by resolution of the chamber and maintained on a continuing basis, is charged with the oversight of the sovereign AI. The committee is composed of senators drawn from all parties represented in the chamber, in proportions that reflect the composition of the Senate itself. Its chair is not a member of the government of the day. Its deputy chair is. The committee is resourced by the Senate staff, and may draw on the expertise of the Parliamentary Library, the Auditor-General's office, and outside academic consultants as is normal for such committees.

The committee's terms of reference are broad. It examines the operation of the sovereign AI, the integrity of its data inputs, the adequacy of its auditability provisions, the handling of exceptions and escalations, the exercise of the temporal exception of section 6.4, and the conduct of any commercial suppliers under contract to the Commonwealth. It reports to the Senate at such intervals as the chamber directs, and in any case annually. Its reports are public.

The committee's function is to ensure that the sovereign AI operates within the framework parliament has set for it, and to identify for parliament and the people any matter on which the framework itself may require revision. It is not a court, and it does not substitute for the democratic accountability that the ballot box provides.

7.3 The Powers of Call

The standing committee has the power to require evidence from three classes of witness.

The first is the Chief Statistician, as the senior officer responsible for the operation of the sovereign AI under the ABS's statutory authority. The Chief Statistician appears before the committee on its call, answers its questions, and provides such documents as the committee requires. The existing statutory independence of the ABS is preserved; the committee's power is to require appearance and answer, not to direct the conduct of operations.

The second is the relevant minister, whether a Minister for Science or a portfolio minister whose area of responsibility has been the subject of an AI recommendation under review. Ministers appear on the same terms as they appear before other Senate committees.

The third is the sovereign AI itself. The committee may put questions to the system, in writing or in open session, and require it to produce its provenance records, its reasoning traces, and its recommendations on any matter within its operation. The AI's responses are made under the same obligation of candour that applies to public servants giving evidence. The committee's record of such evidence is public, subject only to the ordinary protections for classified material.

This third power is novel in the annals of Westminster oversight. Its justification is that a system advising on the policy of a sovereign nation must be capable of answering to that nation's chamber of review in its own voice.

7.4 Continuity Across Governments

The sovereign AI is a Crown asset, a standing capacity of the Commonwealth, operated by the Australian Public Service under the statutory authority of the Australian Bureau of Statistics. It belongs to no party, neither the Demechratic Party that built it nor any party that the Demechratic model may inspire. It continues to exist whether or not that party holds government.

Three possibilities therefore arise at any change of government.

A successor Demechratic government continues to implement the AI's recommendations under the pledge, as the outgoing government did.

A non-Demechratic government inherits the system as a standing capacity of the Commonwealth. Such a government is under no obligation to implement the AI's recommendations. It may continue to consult the AI as one source of advice among many, in the way governments have always consulted the Treasury, the Productivity Commission and the public service. It may ignore the AI's recommendations altogether, legislate a narrowing of the AI's scope, or, with an act of parliament, disestablish the system entirely. Each of these options is available through the ordinary democratic process. None is foreclosed by anything in this proposal.

A hostile government that sought to dismantle the sovereign AI would face the Senate that had watched its operation, the academic community that had analysed its weights, the citizens who had conversed with it through MyGov, the journalists who had consulted its provenance register, and the memory of the outcomes the nation had produced under its counsel. The dismantling would be possible but unlikely to be desirable.

7.5 The Chamber That Watches

The design of the two chambers under Demechracy is a division of labour. The lower house is the chamber of contact, where the sensing happens and the sovereign AI's recommendations are enacted. Its members serve a single term, for the reasons set out in section 5. The Senate is the chamber that watches. Its members serve the longer terms the Constitution already provides, they benefit from the experience a single lower-house term cannot retain, and among them may be the former members of the lower house who, having served and returned to their communities, have been sent back to the parliament by the voters of their state to take up the work of review.

The chamber of contact changes. The chamber of review remembers. The machine runs in the presence of both, and answers to each.

8.0 Employment & Social Engagement

There is an irony at the heart of this party's platform. On the one hand, Demechracy advocates for reducing direct human cognitive involvement in the formulation of government policy, trusting a machine to do what the Dunbar-bounded human mind cannot. On the other hand, we are at our core a humanist party. We advocate that government exists strictly to serve and better the lot of all of its citizens.

We promote technological advance. However, as artificial intelligence and physical robotics advance, the natural market pressure on human employment will inevitably tend toward zero (Newton-Thomas & Strocchi, 2018). There are many ways a society might attempt to absorb this pressure—shorter working weeks, expanded leisure, or expanded welfare. However, we categorically reject any future that accepts the managed obsolescence of the human worker via a simple Universal Basic Income (UBI) and a couch. A UBI, deployed as a substitute for meaningful occupation, is little more than an adult pacifier. We believe this would be an unmitigated social and psychological catastrophe. To permanently sever the population from its productive role is to invite what the sociologist Émile Durkheim (1893/2014) identified as anomie: a profound detachment, normlessness, and despair that occurs when individuals are stripped of their functional place within the social fabric.

8.1 The Humanist Conundrum: The Imperative of Inclusion

Productive participation in the life of a society is not merely an economic transaction; it is an absolute psychological necessity for us as humans. We instinctively bind our sense of self-worth, identity, and belonging to the respect of, and our usefulness to, others and our community. Work defines us. It is one of the first questions people ask of each other on meeting: "What do you do?" Work keeps us physically and psychologically engaged; its absence makes us unhealthy.

The resolution to this conundrum lies in understanding where human effort belongs. We wish to automate the administration of the state precisely so we can, at the same time, protect the participation of the human in the fabric of civil society.

To defend against the oncoming onslaught of automation, we need to actively tip the scales to ensure human engagement. We must engineer a system that retains humans within the workforce while simultaneously allowing the economy to expand via AI and robotics. Tipping the Scales (ARP, 2026) is our companion paper and dedicated proposal to achieve this. It is a framework for keeping humans in the loop of their own civilisation. Its focus is not the structure of government, but the preservation of human dignity and psychological health in an automated age.

8.2 What Tipping the Scales Requires

Tipping the Scales prescribes a framework of interlocking policy mechanisms calibrated to a single objective: the preservation of human participation in the productive economy as machine substitution proceeds. The mechanisms themselves are set out in sections 3.5 through 3.12 of that paper. Summarised briefly, they include:

• A compute tax, levied on imported GPU and accelerator hardware and on the output of foreign-sourced AI services, at rates calibrated to displacement potential. This can be offset by export;
• A compute rebate, available to firms that export AI services or hardware or that use domestic AI services to enhance their own exports;
• A reverse payroll tax, funded by the compute tax revenue and calibrated to the marginal cost of hiring a human in sectors where human employment is at the edge of economic viability;
• A Floor-and-Trade system, establishing a human participation floor for every ANZSIC industry classification, with tradeable credits for firms that wish to reduce their workforce below the floor;
• Safety exemptions, granted by application for automation that demonstrably reduces human exposure to harm;
• A right of human election, binding on all citizen-facing services, with specified response times and access to human arbitration of automated decisions;
• Work-sharing provisions for a shorter working week, activated as labour demand contracts;
• Strategic expansion of agriculture, mining and sovereign energy to grow the productive base from which the participation framework is funded.

Each of these mechanisms is administrable in isolation. The difficulty lies in administering all of them together, in real time, with rates and floors and exemptions calibrated to conditions on the ground that change daily.

8.3 The Administrative Load

Consider the compute tax alone. Its implementation requires continuous monitoring of GPU and accelerator imports at the customs interface, assessment of each device's displacement potential on a methodology that must itself be kept current with the state of the art, and a per-token levy on foreign AI services that must be collectable by providers and auditable by the Australian Taxation Office. The rates themselves are not fixed. They are calibrated to the revenue needs of the participation framework, which in turn depend on the changing distribution of human employment across ANZSIC classifications. A change in any one of these variables propagates through the others.

Consider the Floor-and-Trade system. Every ANZSIC industry classification carries a human participation floor fixed to its 2025 headcount-to-revenue ratio as baseline. The market for tradeable credits operates continuously. Prices reflect the relative ease or difficulty of retaining human participation in each industry. The effective cost of automation in a given firm depends on the credit price, the firm's position relative to its industry floor, and the firm's eligibility for safety exemptions. Across the economy, the market clears against conditions the ABS reports, the public service measures, and the Commonwealth's tax and transfer systems respond to.

Consider the reverse payroll tax. Its rate is the integral of compute tax revenue, the shape of the labour market, the output of the Floor-and-Trade system, and the employment floor the framework is designed to hold. Its correct value changes constantly, and its miscalibration, in either direction, damages the objective the framework exists to serve. An annual ministerial adjustment would sit nowhere near the tempo required.

The administrative load of any one of these mechanisms is beyond what an ordinary ministerial office can absorb. The administrative load of all of them together, coherent with one another, calibrated against conditions in the economy and against each other, is beyond what the cabinet as a whole can absorb. The observation is arithmetical rather than judgmental; no criticism of any minister is implied. The tempo at which the framework must operate exceeds the tempo at which human decision-making can occur, and no increase in the quality of the decision-makers will close that gap.

8.4 The Demechratic Fit

Demechracy furnishes exactly what the Tipping the Scales framework requires: a sovereign capacity to hold the whole framework in mind at once, to ingest the statistical returns in real time, to calibrate rates and floors against the shifting conditions of the economy, to reason across portfolio boundaries without the delay of interdepartmental committees, and to produce policy recommendations that the elected government then implements under its pledge.

The architecture described in section 4 is, in this respect, a specification written to the requirements of Tipping the Scales. The single sovereign AI integrates the compute tax, the payroll tax, the Floor-and-Trade market, the employment floor and the expansion sectors into a single live policy picture. The domain models below it hold the portfolio-level detail. The ABS furnishes the big data. The elected representatives contribute the contextual knowledge of the electorates which they represent. The MyGov channel gives the affected citizens direct voice. The Senate committee watches the machine. The conscience floor reserves to human judgement the decisions that ought never to belong to a machine.

8.5 One Answer to Two Questions

Marx, writing in the middle of the nineteenth century, observed that industrial capitalism was divorcing the worker from the tool: the artisan who owned his instrument and controlled his process was being replaced by the wage-labourer operating an instrument owned and directed by another (Marx, 1867/1976). Intelligent machines are the final form of that divorce. The tool, having taken the place of the artisan, now takes the place of the wage-labourer too. Tipping the Scales is the economic refusal of that conclusion. Demechracy is the governance instrument through which the refusal can be administered.

The two papers, taken together, answer two questions that the twenty-first century has posed to the Australian state.

Tipping the Scales asks: how does a liberal democracy preserve meaningful human participation in the productive economy while the cognitive asymmetry between human and machine widens? It answers with a framework of market-compatible policy mechanisms, ordered to that end.

Demechracy asks: how does a Dunbar-bounded political class steward the welfare of twenty-eight million people in a policy environment of unmanageable complexity? It answers with the insertion of a sovereign cognitive faculty that is not so bounded, placed in the service of the elected government and the sovereignty of the people.

The first paper is the economics. The second is the governance. Neither is sufficient alone. Together they describe a plausible path through a transition for which we are currently unprepared. The ballot box remains the final arbiter of the arrangement. What the ballot box offers, under a Demechratic vote, is the composition of these two answers in a single coherent policy commitment. Through this means we achieve both democratic equity and technocratic efficiency.

9.0 Objections and Responses

A proposal of this scope deserves objection. This party does not expect nor desire the Demechratic model to enter Australian political life without scrutiny, and the authors of this paper have sought out the most serious critiques they can find, in preference to the more easily answered ones. The objections that follow are those most commonly raised when the proposal is put to thoughtful interlocutors. They are presented, in each case, in the form that gives them the greatest weight, before the response is set out.

9.1 "The AI will reproduce the biases contained with its training data"

The objection is valid. Any machine learning system reflects the distribution of the data it is trained on. Historical statistical records carry the biases, omissions and category errors of the societies that collected them. An AI trained on such records may reproduce those patterns without the moral compass that would have caused a human analyst to question them.

The Demechratic response operates at three levels. The first is at source. The Australian Bureau of Statistics has, over decades, invested in the methodology of representative sampling, demographic disaggregation and the detection of systemic undercount. The sovereign AI inherits that methodological tradition, and with it the institutional commitment to revise and refine data collection where gaps are identified.

The second is provenance. Every significant recommendation produced by the sovereign AI is accompanied by a provenance record that names the data sources consulted and the inferences drawn from them. A recommendation whose inferences derive from a dataset of known bias is a recommendation that can be identified, critiqued, and where necessary rejected, by the parliament, the academic community, the press, and the citizens who consult the register.

The third is the MyGov channel. The citizens about whom the data speaks are, under Demechracy, able to speak for themselves. A population systematically undercounted in the statistics has a direct line to the AI that will act on those statistics, and the AI is obliged to weight their voice as a distinct class of input, per section 4.2. The machine that might otherwise have missed them is now in conversation with them daily.

9.2 "Democracy is being hollowed out by a machine"

This is the objection the authors have taken most seriously, because the appearance of hollowing is real even where the substance is preserved. If the policy of the state is drawn from a machine, and implemented by members whose vote is pledged, then the question arises: what is left for the elected representative to do, and what is left for the voter to decide?

The answer is set out across sections 3, 5 and 6 of this paper, and may be stated compactly here. The voter retains every power the voter has ever held: to elect the party of their choice, to vote out the government they dislike, to stand for office themselves, to converse directly with the sovereign AI through MyGov, and to hold their representative to account at every election. The representative retains the work of sensing their electorate, advocating on its behalf to the machine, and, on the matters of gravest moral weight, voting their conscience on the public record. What has changed is the substrate of policy analysis, not the distribution of political authority.

The objection is, at its deepest, a question about whether the Demechratic arrangement feels democratic enough to sustain democratic legitimacy. The test will be conducted at every federal election, in the ordinary way, by the Australian people.

9.3 "Alignment: what if the AI misinterprets its objective?"

The alignment literature in artificial intelligence is vast, and this paper will not attempt to summarise it. The technical worry, stated plainly, is that a sufficiently capable optimising system may pursue the letter of its stated objective in ways that violate the intent behind it. An AI instructed to maintain employment might, in a narrow reading, find clever ways of reporting employment figures that satisfy the metric without serving the purpose.

The Demechratic response is structural. The objective of the sovereign AI, as section 4.4 sets out, is expressed as a floor and a direction rather than as a single scalar to be maximised. Dignity and participation are hard constraints, not tradeable quantities. The outcomes are measured by an institution (the ABS) whose statistical methodology is published, peer-reviewed, and resistant to gaming. The recommendations are explained by a provenance record that makes the reasoning visible. The parliament retains the authority to legislate new constraints where the existing ones have been read against their purpose. The electorate retains the authority to vote out a government whose AI is being read against its purpose.

No system of this kind can eliminate alignment risk entirely. What the Demechratic design does is place the risk within reach of the ordinary democratic and bureaucratic correction mechanisms the Commonwealth already operates.

9.4 "The operator will capture the machine"

A more cynical variant of the bias objection holds that whoever writes the objective function, trains the model, and curates the inputs will determine its recommendations, and that this power will be captured, in time, by vested interests with the resources to exploit it.

The objection is serious, and the response rests on institutional design. The sovereign AI is placed under the statutory authority of the Australian Bureau of Statistics, an institution with a century of methodological independence and a legislated remit for confidentiality and accuracy. Its weights are held by the Commonwealth and made available to the Australian Research Council under managed access, so that capture would need to survive independent academic analysis. Its provenance records are public. Its conduct is audited by a Senate committee whose chair is not a member of the government of the day. These are the same safeguards that have preserved the integrity of the ABS against political pressure across governments of every persuasion, applied now to a new institution of far greater consequence.

No institutional design makes capture impossible, and the Demechratic design is no exception. What the design does is make capture visible, expensive, and dependent on the complicity of several institutions which have historically proven unwilling to supply it.

9.5 "What if the AI is simply wrong?"

Every governance mechanism is capable of error. The AI will be wrong on some matters, and the framework must accommodate this.

Four remedies are available. The first is the right of human election set out in section 3.8 of Tipping the Scales: every citizen interacting with any automated system has the right to transfer to a human arbitrator, within specified response times, who may review and overturn the machine's determination. The second is the Senate committee of section 7.2, which may recommend revisions to the framework on any matter within its remit. The third is the conscience floor, which removes the gravest decisions from the AI's authority in the first instance. The fourth is the ballot box, which removes a government whose AI has persisted in error despite the other remedies being available.

The objection rests, ultimately, on the assumption that error in the present system is less than error in the proposed one. The burden of proof on that assumption is not light. The pathologies set out in section 2 are errors of the present system in aggregate. The case for Demechracy is that, on the matters where the present system fails most predictably, a Dunbar-unbounded faculty, properly audited and properly overseen, will fail less.

9.6 "This is technocracy by another name"

The twentieth century furnishes more than one example of governance by expert committee that bypassed democratic accountability, and the track record is mixed at best. The objection is that Demechracy is such a scheme with a new face.

The authors believe the difference is substantive. The experts of the technocratic century were placed beyond democratic accountability by design, through constitutional entrenchment, through treaty obligation, or through institutional independence from electoral consequence. The sovereign AI is placed within democratic accountability by design. Its recommendations are implemented only by a government the electorate has chosen, for a term the electorate has set, under a pledge the electorate has seen. Its reasoning is public. Its operation is reviewable. It can be defeated by a ballot. The distinction is not cosmetic.

A machine counselling an elected parliament is not the same thing as a commission insulated from parliamentary reach. The first supplements democratic authority. The second supplants it. The proposal of this paper is the first.

10.0 Conclusion

The argument of this paper rests on three propositions.

First, that human political cognition is bounded by biology, that the bound is Dunbar's limit of approximately one hundred and fifty, and that the pathologies of modern democratic life at the scale of twenty-eight million are consequences of asking that cognition to do what biology does not permit. Its operators are overloaded. Democracy itself is not defective.

Second, that a complementary faculty exists, in the form of a sovereign artificial intelligence built, owned and audited by the Commonwealth, fed by the statistical ground truth of the Australian Bureau of Statistics, advised by elected representatives who sense the life of their electorates, and in daily conversation with citizens through the MyGov platform. This faculty can hold the whole of the commonwealth in mind. It can work at the tempo the administration of a modern economy requires.

Third, that this faculty can be placed in the service of the Australian polity without abolishing a single democratic institution. The overlay is carried by a party pledge, not a constitutional amendment. The representative retains the sensing function and the conscience vote. The burden of primary policy authorship, rightly, falls to the sovereign AI. The Senate retains its powers of review, extended to include the oversight of the machine. The ballot box remains the final authority on every matter, at every election.

We have set out the architecture of the sovereign AI, the pledge of the party that would call it into service, the single-term rule that keeps the representative close to the life of their constituents, the conscience floor that reserves to human judgement the decisions of gravest moral weight, and the Senate committee that watches the machine in the name of the people. We have shown, in section 8, how the framework of our earlier paper on the economic implications of machine cognition becomes administrable, in real time and with the necessary precision, only through this overlay. We have responded to the objections we expect to meet, on the terms that give those objections the greatest weight.

None of this requires faith in machines. The proposal assumes that machines are capable of error, that their operators are capable of bias, and that no institutional design is proof against capture. It relies, instead, on the democratic and bureaucratic correction mechanisms that Australia has built over more than a century: the independence of the ABS, the review of the Senate, the transparency of the Gazette, the scrutiny of the press and the academy, and the authority of the ballot. The sovereign AI is placed within all of them. What it adds is capacity, not authority.

The citizen of Demechracy is governed by people they have elected, advised by a machine that listens to them directly, operating under a framework they can change at every election. The representative serves for a term and goes home. The machine answers to the Senate, to the provenance register, and to the people it speaks with every day. The objectives are set in statute, and the statute is set by the parliament the people have chosen.

Eight years ago, this party's founding submission to the Senate argued that the Luddites of the early industrial revolution had been correct in their recognition of the threat of automation, and wrong only in their response to it (Newton-Thomas & Strocchi, 2018). Tipping the Scales is half of a better response. Demechracy is the other half.

The twenty-first century will ask of every liberal democracy whether it can govern at machine tempo without ceasing to be democratic. Demechracy is the affirmative answer. The people still rule. The machine is the means by which the ruling can be done in our far more complex world.

References

Achen, C. H., & Bartels, L. M. (2016). Democracy for realists: Why elections do not produce responsive government. Princeton University Press.

Anthropic. (2025). Claude model card and long-context capabilities [Technical documentation]. Anthropic.

The Australia Institute. (2023). Democracy for sale: Political donations and policy access in Australia [Discussion paper]. The Australia Institute.

Australian Bureau of Statistics. (2024). Births, Australia, 2023 (Catalogue 3301.0). Commonwealth of Australia.

Australian Electoral Commission. (2025). Enrolment by federal electoral division. Commonwealth of Australia.

Australian Renaissance Party. (2026). Tipping the scales: An Australian Renaissance Party discussion paper on human participation in the age of intelligent machines. Australian Renaissance Party.

Coupé, C., Oh, Y. M., Dediu, D., & Pellegrino, F. (2019). Different languages, similar encoding efficiency: Comparable information rates across the human communicative niche. Science Advances, 5(9), eaaw2594.

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114.

Durkheim, É. (2014). The division of labour in society (W. D. Halls, Trans.). Free Press. (Original work published 1893)

Dunbar, R. I. M. (1992). Neocortex size as a constraint on group size in primates. Journal of Human Evolution, 22(6), 469–493.

Dunbar, R. I. M. (1993). Coevolution of neocortical size, group size and language in humans. Behavioral and Brain Sciences, 16(4), 681–735.

Dunbar, R. I. M. (2010). How many friends does one person need? Dunbar's number and other evolutionary quirks. Harvard University Press.

Evans, R. J. (2003). The coming of the Third Reich. Penguin Press.

Financial Crisis Inquiry Commission. (2011). The Financial Crisis Inquiry Report: Final report of the National Commission on the Causes of the Financial and Economic Crisis in the United States. U.S. Government Printing Office.

Google DeepMind. (2025). Gemini long-context capabilities [Technical documentation]. Google DeepMind.

Haidt, J. (2022, May). Why the past 10 years of American life have been uniquely stupid. The Atlantic.

Hill, R. A., & Dunbar, R. I. M. (2003). Social network size in humans. Human Nature, 14(1), 53–72.

Irwin, D. A. (2011). Peddling protectionism: Smoot-Hawley and the Great Depression. Princeton University Press.

Lorenz-Spreen, P., Lewandowsky, S., Sunstein, C. R., & Hertwig, R. (2020). How behavioural sciences can promote truth, autonomy and democratic discourse online. Nature Human Behaviour, 4(11), 1102–1109.

Lowy Institute. (2024). The Lowy Institute poll 2024. Lowy Institute for International Policy.

Marx, K. (1976). Capital: A critique of political economy, volume I (B. Fowkes, Trans.). Penguin Classics. (Original work published 1867)

Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97.

National Disability Insurance Agency. (2024). NDIS quarterly report to disability ministers. National Disability Insurance Agency.

Newton-Thomas, J., & Strocchi, J. (2018). A report to the Senate Select Committee on the Future of Work and Workers: The Luddite fallacy fallacy (Submission 84). Senate Select Committee on the Future of Work and Workers, Parliament of Australia.

Parliamentary Library. (2023). The occupational backgrounds of members of the 47th Australian Parliament [Research paper]. Department of Parliamentary Services.

Supran, G., & Oreskes, N. (2017). Assessing ExxonMobil's climate change communications (1977–2014). Environmental Research Letters, 12(8), 084019.

Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 33–47). Brooks/Cole.

Treasury. (2023). 2023 intergenerational report: Australia's future to 2063. Commonwealth of Australia.