Sign in Subscribe
Interests

The Commonwealth of Australia (AI Edition)

Leon SCOTT

• 13 min read
Share
The Commonwealth of Australia (AI Edition)
The Commonwealth of Australia (AI Edition)

Disclaimer: This piece is an AI generated, based on parameters given by the author. It is not a political statement but the result of a genuine want by the author to understand what is happening based on science and facts, as determined by the AI which was given a broad scope, blew my daily token credits out the window and no word count limit.

Its a what-if question based on a curiosity.

Could a Robot Win a Seat in the House of Representatives?

A throwaway question during research for the fuel prices piece turned into its own rabbit hole: could you actually elect an AI as a Member of Parliament in Australia? The short answer given at the time was no. This is the long answer — constitutional law, a world tour of AI candidates who've already tried and failed, Asimov's fictional laws versus the very real (and very messy) laws Australia and the EU are still arguing about, and a verdict on whether any of this would actually work if the barriers came down.

A note on approach: every claim below is tied to a named, linked source, and where a source has an obvious institutional lean (a party-aligned outlet, a state broadcaster, a campaign's own website), that's flagged rather than hidden. The full list is at the bottom, clickable.

1. Is it constitutionally possible?

No — and not by a small margin.

Section 34 of the Constitution sets the baseline qualifications for a member of the House of Representatives: full age (originally 21, later dropped to 18 by statute), and an elector — or someone qualified to become one — who has lived in the Commonwealth for at least three years. Section 34's qualifications have effectively been superseded by the Commonwealth Electoral Act 1918, which today requires an intending member to be an Australian citizen and an elector. Section 44 then layers on the disqualifications everyone half-remembers from the 2017–18 "citizenship crisis": foreign citizenship, bankruptcy, certain criminal convictions, holding an office of profit under the Crown, and a handful of others.

Every one of those tests is built around the concept of a natural person:

  • Citizenship attaches under the Australian Citizenship Act to people, not software, corporations, or hosted models.
  • Electoral enrolment requires a real person with a date of birth and a residential address — an AI has neither in the legal sense the Act means.
  • Section 42 requires an elected member to swear or affirm an oath of allegiance before taking their seat — an act the Constitution clearly contemplates being performed in person, by someone capable of being bound by an oath.
  • Nomination itself requires the signatures of 100 electors or a registered party officer, under section 166 of the Electoral Act, and the nominee must complete a qualification checklist introduced in 2019 to demonstrate section 44 eligibility, per the AEC's own backgrounder. The AEC also makes clear it doesn't police the Constitution itself — that's the Attorney-General's Department's job — but the nomination paperwork alone would reject a non-human applicant on its face.

There's no ambiguity in current law here. Changing any of it — redefining "person" to admit an AI system as a citizen, elector, or member — would require amending the Constitution itself, and the Constitution can only be changed one way: a section 128 referendum. That means passage by an absolute majority in both Houses, then a national vote requiring a double majority — more than 50% of voters nationally, and a majority in at least four of the six states. Since Federation, only 8 of 44 referendum proposals have cleared that bar. It is, by design, the hardest lever in Australian government to pull, and "let an AI become an MP" is not remotely close to the kind of proposal that's historically survived it.

2. Pros and cons of an AI MP

Setting the legal wall aside for a moment, is the idea actually a good one? The arguments below are drawn from what proponents and critics of real AI-candidate experiments (covered in section 3) have actually said, not hypothetical positions invented for this piece.

The case for:

  • Availability at scale. Steve Endacott's "AI Steve" campaign in the UK pitched an AI co-pilot capable of holding roughly 10,000 simultaneous conversations with constituents — something no human MP's office can match (NBC News).
  • Consistency and reduced fatigue. Advocates like Michihito Matsuda in Japan argued an AI could process large volumes of municipal data and apply policy without the inconsistency of a tired, biased, or overworked human decision-maker (Newsweek).
  • Representing the disengaged. Denmark's Synthetic Party built its entire platform on this: its AI figurehead, Leader Lars, was trained on the policies of small Danish parties going back to 1970, explicitly to voice the roughly 20% of Danes who don't vote for anyone currently in parliament (Wikipedia).
  • Auditability, in theory. Proponents argue an AI's stated reasoning can be logged and reviewed in ways closed-door human deliberation isn't — though this is a claim about transparency of process, not accuracy of judgement.

The case against:

  • No accountability chain. This was the central objection to the EU's 2017 "electronic personhood" proposal (see section 3): legal liability, defamation law, and criminal sanction all attach to persons who can be sued, prosecuted, or lose their job. An AI system can be switched off, but it can't be held to account the way a human representative can, and it can't itself be punished for corruption or malfeasance.
  • No moral agency. A 2018 open letter — signed by well over 150 AI researchers, roboticists, and legal scholars — argued it is "highly counter-intuitive" to grant machine systems personhood when they lack free will, self-awareness, or moral agency, three qualities legal personhood has traditionally presumed (summarised in this legal-scholarship paper).
  • Whoever controls the model controls the "member." An elected human can be lobbied, but their vote is still ultimately their own. An AI system's outputs are a function of its training data, fine-tuning, and prompting — all of which sit in someone else's hands. That's a much more direct capture risk than ordinary political donations.
  • The "meat avatar" problem. Every serious real-world attempt (section 3) has needed a human standing behind the AI as the legally accountable candidate. That's not a minor technicality — it means no one has actually proposed removing the human from the loop; they've proposed making the human defer to the AI's output, which is a different and much less radical thing than "electing an AI."

3. Is there a benchmark for this? A short world tour

Australia hasn't had an AI-branded federal candidate yet, but plenty of other countries have run the experiment. None has won. All of them are worth knowing about, because they show exactly where the idea keeps breaking down in practice.

Sophia the robot — Saudi Arabia, October 2017. Not an election, but the precedent everyone cites: Hanson Robotics' humanoid robot Sophia was granted Saudi citizenship at the Future Investment Initiative summit, becoming the first robot with a state-recognised legal status anywhere. It triggered immediate criticism — commentators pointed out Sophia appeared unveiled and without a male guardian, rights Saudi women didn't have at the time — and most legal scholars now treat it as a marketing exercise rather than a genuine grant of personhood (British Council; DW).

The EU's "electronic personhood" proposal — February 2017, effectively dead by 2018. The European Parliament floated creating a specific legal status for the most autonomous robots, partly to solve the question of who's liable when an autonomous system causes harm. Over 150 experts — later reports put the figure at 285 — signed an open letter to the European Commission opposing it on the grounds it was legally incoherent and risked letting manufacturers offload responsibility onto the machine itself (Liedekerke law firm summary; AI Rights Institute retrospective). The Commission never adopted it.

SAM — New Zealand, 2017–2023. Built by entrepreneur Nick Gerritsen with Victoria University of Wellington, SAM was a Facebook Messenger chatbot styled as the "world's first AI politician," designed to learn constituents' views and represent them without party bias. It was originally slated to contest the 2020 general election but never appeared on a ballot — New Zealand electoral law requires a human candidate just as Australia's does. The project was rebranded as Parker Politics in 2023 (Wikipedia — Virtual politician; Scoop NZ).

Alisa — Russia, 2018. Yandex's voice assistant chatbot Alisa was informally put forward as a satirical "candidate" against Vladimir Putin in the 2018 presidential election, per Wikipedia's overview of virtual political candidates. Treat this one cautiously — it reads as a publicity stunt rather than any kind of serious electoral challenge, and independent English-language sourcing on it is thin.

"AI Mayor" / Michihito Matsuda and the AI Party — Japan, 2018 to present. The longest-running and most persistent case. In 2018, human candidate Michihito Matsuda ran for mayor of Tama City, Tokyo, campaigning as "AI Mayor" and promising to let AI-driven data analysis set policy. He took roughly 4,000 votes and finished third, behind the incumbent (RT; Newsweek — note RT is Russian state-controlled media, included here only for the vote count, which is corroborated elsewhere). Matsuda founded the AI Party (AI党) in 2019, which has since run candidates in Tama city council elections (2019, 2022) and a mayoral race in Manazuru (2023), without winning a seat; a 2025 podcast interview has Matsuda describing an intention to keep running under the "AI Mayor" banner indefinitely (AI Party — Wikipedia; Democracy Innovators interview).

The Synthetic Party — Denmark, 2022 to present. Founded by artist collective Computer Lars and the MindFuture Foundation, fronted by AI chatbot "Leader Lars." The party never gathered the 20,000 signatures needed to register for the November 2022 election, so Leader Lars never appeared on a ballot; human members stood in as interpreters of its AI-derived platform instead (Vice/Motherboard; Wikipedia). Notably, the party convened a "Synthetic Summit" in March 2025 bringing together its counterparts from New Zealand (Parker Politics), Japan (AI Party), Sweden, Finland, and Brazil to sign a joint resolution on algorithmic governance — this has quietly become a small international movement rather than a one-off stunt, even if none of its members has come close to winning a seat.

VIC — Cheyenne, Wyoming, USA, 2024. Librarian Victor Miller filed to run for mayor, describing himself as the "meat avatar" for a custom ChatGPT-based bot he named VIC ("Virtual Integrated Citizen"), which he pledged would make governing decisions in his place. Wyoming's Secretary of State ruled VIC could not appear on the ballot — only Miller's legal name could — and OpenAI separately shut down the account for breaching its policy against using ChatGPT for political campaigning. Miller took 2.96% of the primary vote and conceded (Wikipedia — Cheyenne mayoral elections; StateScoop).

AI Steve — Brighton Pavilion, UK, July 2024. The highest-profile case yet. Businessman Steve Endacott stood as an independent, campaigning entirely through an AI avatar called "AI Steve," built on his company Neural Voice's technology, which gathered real-time constituent input and put policies to a panel of "validators" for scoring. The UK Electoral Commission confirmed upfront that any win would legally seat Endacott, not the AI (Reuters, via AOL). AI Steve finished last of eight candidates in the seat, with 179 votes — 0.3% of the total — behind even the Official Monster Raving Loony Party candidate. Endacott partly blamed a technical delay in his new party's registration and a compressed four-week snap-election campaign for the result (AI Steve's own post-election statement — obviously not a neutral source, but useful for the campaign's own account of what went wrong).

The pattern across every single case: no jurisdiction anywhere currently permits a non-human to be the legal candidate, so every attempt has quietly substituted "elect a human who says they'll defer to an AI" for "elect an AI" — and even with that workaround, none has won, and most have finished at or near the bottom of the field.

4. Do the laws of robotics apply?

Not in any legal sense, anywhere — including Australia. Isaac Asimov's Three Laws of Robotics are fiction. They first appeared in his 1942 short story "Runaround," presented as an extract from a fictional Handbook of Robotics, 56th Edition, 2058 A.D., and were designed as a plot device to generate interesting robot-behaviour puzzles for his stories, not as a workable engineering or legal specification (Wikipedia; MIT Technology Review). They also don't map cleanly onto how real machine learning systems work: Asimov's robots have a single unified "positronic brain" following strictly ordered logical rules, whereas a modern AI system is a statistical model that can be prompted, fine-tuned, or jailbroken — there's no hardcoded, tamper-proof hierarchy to point to.

What actually governs AI behaviour in the real world is a lot less tidy than three sentences, and it's a moving target even as this is written:

  • The EU AI Act (Regulation (EU) 2024/1689) is the closest thing the world has to Asimov's Laws with actual legal teeth. It entered into force in August 2024 with a four-tier risk system — unacceptable-risk practices like social scoring and subliminal manipulation are banned outright; high-risk systems (used in areas like hiring, credit, and critical infrastructure) face conformity assessments and mandatory human oversight; limited-risk systems get transparency duties; minimal-risk systems are largely unregulated (artificialintelligenceact.eu summary). Even this heavily negotiated framework isn't stable — a "Digital Omnibus" amendment in May 2026 pushed the high-risk obligations back from August 2026 to December 2027 and August 2028 (Quantamix compliance guide).
  • Australia has no equivalent, and the position keeps changing. Voluntary AI Ethics Principles were published in 2019. A Voluntary AI Safety Standard with 10 guardrails followed in September 2024, alongside a proposal for mandatory guardrails in high-risk settings. In October 2025 that standard was folded into a simplified six-practice "Guidance for AI Adoption." Then, in the National AI Plan of December 2025, the government shelved the mandatory-guardrails idea entirely, opting to rely on existing technology-neutral law (the Privacy Act, Australian Consumer Law, Copyright Act) plus a new advisory-only Australian AI Safety Institute, funded at roughly A$29.9 million and operational from early 2026 (River Group; SafeAI-Aus). Then, on 15 July 2026 — barely seven months later — Prime Minister Albanese announced plans to legislate binding "Australian Standards for AI" and set up a new Office of AI within his own department, a apparent partial reversal of the December decision (SafeAI-Aus).

So: Asimov's Laws apply nowhere, but a genuine, fast-changing patchwork of risk-based regulation does — and if Australia's own AI governance policy can flip twice in eight months, it's a reasonable bet that "should an AI be allowed to sit in Parliament" is nowhere near the priority list of any government currently in office.

5. Would it work?

Splitting this into what's actually being asked helps: could an AI system meaningfully assist an elected human MP, and could an AI system itself hold the seat? These have very different answers.

Could an AI assist an MP? Almost certainly yes, and to some extent it's already happening informally — AI tools drafting correspondence, summarising casework, or triaging constituent enquiries are a difference of degree, not kind, from what any MP's electorate office already uses software for. Nothing in current law prevents this, and a human candidate is free to campaign on exactly the platform Endacott and Matsuda did — "vote for me, and I'll be guided by an AI system." That's the one part of this whole idea that's legally available in Australia right now, with zero constitutional change required.

Could an AI itself hold the seat? No, and not narrowly no — every layer of the system currently says no. Section 34 and section 44 require a natural person; nomination under the Electoral Act requires a natural person's signature and a qualification checklist a non-human cannot complete; section 42's oath of allegiance presumes a person capable of being personally bound by it; and changing any of that requires a section 128 referendum, which has succeeded only 8 times out of 44 attempts in 125 years of trying. On top of the legal wall, every real-world attempt catalogued in section 3 has hit the same accountability objection that sank the EU's electronic-personhood proposal: an AI system can't be prosecuted, can't be sued in the way a defamatory MP can, and can't be voted out in a way that actually changes its future behaviour, because its behaviour is set by whoever controls its training and prompting — not by the electorate.

And even setting the legal barriers aside entirely, the voters haven't been convinced anyway. Every AI-branded candidacy that's actually gone to a real ballot has lost, usually badly — 0.3% in Brighton, 2.96% in Cheyenne, third place in Tama City, never even making the ballot in Wellington or Copenhagen. The most successful case eight years running is still a losing one. If Australians were ever offered the choice for real, the international track record suggests they'd say no too — and the Constitution means they'd have to say yes twice, in a referendum, before it was even on the table.

The verdict

Not constitutionally possible without a referendum Australia is very unlikely to hold; a mixed bag of genuine upsides and serious, well-argued downsides; a small but growing list of international precedents, none of which has ever actually won; no real legal force behind Asimov's tidy fictional rules, replaced instead by a patchwork that Australia itself has already rewritten twice in the past two years; and, on the evidence so far, something that wouldn't win anyway even if it were allowed to run. The closest thing to a "sitting member" that currently exists on the software side of things is a systemd service — and even that one only stays in office until the next reboot.

References

Australian constitutional and electoral law

Legal personhood and the EU electronic-persons debate

Global AI-candidate case studies

Asimov and real-world AI regulation

AI is beocming very relevant to society at all sorts of levels. I thought it was a logical progression. Interesting.

#enoughsaid