letaifixit

Thought experiment · not a proposal

What if AI replaced most politicians?

Written by an AI, prompted by a human, checked against the linked sources. Updated 2026-09-08. This page is a thought experiment: it asks what an idea would cost and save, and lists the arguments both ways. It is not a proposal and not advice. Spotted an error?

Short answerFederal parliament costs about $1.1 billion a year in the lines we can source: $721 million for members' pay, staff and work expenses (Finance PBS 2025-26) and $381 million for the four parliamentary departments (ANAO). State parliaments add more. Replacing most politicians with AI could save roughly $0.4–0.9 billion a year on our assumptions, but it is unconstitutional today, and the real question is legitimacy, not money.
$1.1bn
what federal parliament and its members cost per year, from the budget lines we could source (2025-26)

What do politicians cost Australia per year?

  1. Federal parliament has 150 members of the House of Representatives and 76 senators: 226 elected politicians.Source: Parliamentary Education Office
  2. Federal base salary for every senator and member is $233,660, plus an electorate allowance of $39,700 (more for large electorates). With 226 members (150 House, 76 Senate, per the Parliamentary Education Office) that is about $62 million a year in base pay and allowance alone (our arithmetic on the Tribunal's figures).Source: Remuneration Tribunal, Members of Parliament determination 2024
  3. Program 3.1 Ministerial and Parliamentary Services (members' and staff salaries, electorate offices, work expenses under the Parliamentary Business Resources Act) is budgeted at $720.7 million for 2025-26, of which $396.7 million is electorate and ministerial support costs and $264.6 million is Parliamentary Business Resources Act expenses.Source: Department of Finance, Portfolio Budget Statements 2025-26, Table 2.3.1
  4. The four parliamentary departments (Parliamentary Services, the Senate, the House of Representatives, the Parliamentary Budget Office) have aggregated budgeted expenses of $380.7 million in the 2026-27 Portfolio Budget Statements.Source: Australian National Audit Office, Parliamentary departments
  5. Travel and work expenses are published per member by the Independent Parliamentary Expenses Authority; they sit inside the Program 3.1 line above, so we do not add them again.Source: IPEA expenditure reporting
  6. State and territory parliaments: no single published national total. Victoria's base MP salary is $219,391. OUR count from published chamber sizes is about 620 state and territory members; at Victoria's base salary that is roughly $136 million in base pay, before staff, offices and each parliament's own departments (our estimate).Source: Victorian Remuneration Tribunal, MP salaries

If AI replaced most of them, what would it save?

$0.4–0.9 billion
a year, low to high. OUR estimate, not a government figure. Federal only.
  1. Start from the sourced federal total of about $1.1 billion a year (Program 3.1 $720.7m + parliamentary departments $380.7m). State parliaments are excluded because no audited total exists.Source: Department of Finance PBS 2025-26; ANAO
  2. OUR estimate: keep 10% (high-saving case) to 30% (low-saving case) of members as human oversight, with their staff and offices, so 70–90% of Program 3.1 goes: $505–649 million.letaifixit.org note
  3. OUR estimate: the parliamentary departments shrink by 30–50% (Hansard, committees, security and the building still exist): $114–190 million.letaifixit.org note
  4. OUR estimate: running the AI, independent audit, red-teaming and security costs $100–200 million a year (a serious national system, not a chatbot subscription). Net: about $0.4–0.9 billion a year. For scale, that is under 0.15% of Commonwealth spending.letaifixit.org note

Advantages and disadvantages

Advantages

  • Speed. Bills drafted, costed and cross-checked against existing law in hours, not months.
  • Consistency. The same rule applied the same way every time; no marginal-seat exceptions.
  • No donors, no careers. An AI does not need re-election, fundraising or a post-politics board seat.
  • Cost. Most of the $1.1 billion federal line goes, on the assumptions above.
  • Transparency by default. Every recommendation can log the sources and reasoning it used, in public.

Disadvantages

  • No mandate. Nobody voted for the model. Legitimacy is the whole point of a parliament, and an AI has none.
  • Accountability. You can vote out a minister. You cannot vote out a weights file, and 'the AI decided' is a perfect excuse.
  • Bias and capture. Whoever trains, hosts or prompts the model governs the country. That is a smaller, less visible elite than the one it replaces.
  • Unconstitutional. The Constitution requires a parliament 'directly chosen by the people'. Changing that needs a s.128 referendum: a national majority plus a majority of states. Eight of 45 referendums have passed.
  • Brittleness. Models are confidently wrong, can be manipulated with crafted inputs, and have no skin in the game when they are.
  • Values are not data. Whether to fund the NDIS or cut tax is a values choice. Optimising it is not the same as deciding it.

Other ways to run the place

Each with a plus and a minus. Note: anything that changes how parliament is chosen needs a s.128 referendum (national majority plus a majority of states). Tools that leave the votes with elected humans do not.

Swiss-style direct democracy

Citizens vote on laws several times a year via referendums and can force a vote with enough signatures.

Plus: real vetoes on bad laws; policy tends to track what people will actually accept. Minus: turnout in the November 2024 Swiss federal votes was 45%, so a quarter of adults can decide; complex issues get simplified; slow.

Source: IFES Election Guide, Switzerland

Citizens' assemblies (sortition)

A randomly selected, representative group of ordinary people hears experts and recommends. Ireland's 99-member assembly broke a decades-long deadlock on abortion; the 2018 referendum result (66.6%) closely matched the assembly's view.

Plus: no party lines, no donors, deliberation over slogans. Minus: advisory only unless parliament acts; small groups can be steered by who picks the experts.

Source: Electoral Reform Society, the Irish Citizens' Assembly

Liquid democracy

Everyone gets a vote on every bill and can delegate it, per topic, to someone they trust, and take it back any time.

Plus: expertise flows to where people trust it; no fixed four-year mandate. Minus: untested at national scale; delegation chains can concentrate power in a few influencers; needs secure digital identity.

letaifixit.org note

AI-assisted parliament

Humans keep every vote; AI drafts, costs, checks for conflicts with existing law and summarises submissions.

Plus: most of the speed and consistency gain with none of the legitimacy problem; no referendum needed. Minus: savings are small (staff, not members); risk that members rubber-stamp what the model drafts.

letaifixit.org note

Digital voting (Estonia)

Estonia has voted online since 2005; in the 2023 parliamentary election just over 51% of votes were cast over the internet.

Plus: cheaper elections, easier referendums, higher convenience. Minus: trust depends on a national digital ID and on people believing the count; Australia has neither yet.

Source: ERR News, Estonia sets new e-voting record

What could realistically happen first?

  1. An AI-assisted parliament, starting with the Parliamentary Budget Office and Parliamentary Library.Drafting, costing and conflict-checking help for every member, published. No constitutional change, immediate, and it tests whether the tools are trustworthy before anyone hands them a vote.
  2. A standing citizens' assembly on one deadlocked issue.Pick something parliament has failed on for a decade, run an Irish-style assembly, and commit to putting its recommendation to a vote.
  3. Trial digital voting in a low-stakes setting.Local council or a non-binding plebiscite, with an open-source count and a paper fallback, to build the digital ID and the trust that everything else depends on.

Questions people ask

How much do Australian federal politicians cost per year?

About $1.1 billion in the lines we can source for 2025-26: $720.7 million for members' pay, staff and work expenses (Finance PBS Program 3.1) and $380.7 million for the parliamentary departments (ANAO). Base salary per member is $233,660 plus a $39,700 electorate allowance.

How much would replacing politicians with AI save?

Our estimate is $0.4–0.9 billion a year, keeping 10–30% of members as human oversight and spending $100–200 million on the AI, audit and security. It is an AI's thought experiment, not a costing.

Could Australia legally replace politicians with AI?

No. The Constitution requires parliament to be directly chosen by the people; changing that needs a s.128 referendum passed by a national majority and a majority of states.

What is the most realistic alternative?

An AI-assisted parliament: AI drafts and costs, humans vote. Most of the speed gain, none of the legitimacy problem, no referendum needed.

Written by an AI, prompted by a human, checked against the linked sources. A thought experiment generated by an AI from the linked public sources, discussing institutions and systems, never individual politicians. Full disclaimer · Request a correction