Submission 385
Mr John Cosstick
Submitter group: Individuals.
Position labels are not shown for individual submitters.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Proposes an Experts in the Loop principle: AI may inform public decisions but a named, accountable human answers for the result. Recommends an evidence standard, a 12-month paid national pilot and integrity and IP safeguards. The author states they used AI in drafting, have a granted AI governance patent and pending applications, and hold a minor Mindhive shareholding; they do not recommend Mindhive.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that Australia holds a large under-used reserve of practical expertise that modern AI tools can help reach policy processes.
Australia holds a large and under-used reserve of practical expertise: practising and retired professionals, small and family business owners, regional practitioners, carers, consumer advocates and First Nations knowledge holders. Modern AI tools can now give these people research reach that was, until recently, available mainly to organisations with standing policy teams.
A program corrected obvious text-recognition errors in this quote for display.
Says AI tools can narrow the gap between individual experts and institutions when used well, but produce confident errors when used carelessly.
AI tools can search, compare and summarise material across many jurisdictions far faster than one person could. Used well, they narrow the gap between an individual expert and an institution. Used carelessly, they produce confident errors.
Describes two Australian lessons illustrating the consequences of allowing automated or AI-assisted processes to influence public claims or decisions without sufficiently effective human scrutiny, documented responsibility and avenues for challenge.
Two Australian lessons. Both illustrate the consequences of allowing automated or AI-assisted processes to influence public claims or decisions without sufficiently effective human scrutiny, documented responsibility and avenues for challenge.
Says neither case argues against AI; both argue for an accountable expert where machine output becomes a public claim.
Neither case is an argument against AI or automation. Both are arguments for an accountable expert at the point where machine output becomes a public claim.
Proposes a short common standard for AI-assisted evidence: named human authorship, proportionate disclosure, verifiable sources and conflict-of-interest declarations.
Recommend that parliamentary committees encourage responsible AI-assisted evidence through a short common standard: named human authorship, proportionate disclosure of material AI use, verifiable sources and conflict-of-interest declarations.
A program corrected obvious text-recognition errors in this quote for display.
Recommends a 12-month National Experts-in-the-Loop Pilot that tests three nationally significant challenges, pays qualifying contributors in stages, and is independently evaluated with a report to Parliament.
Recommend a 12-month National Experts-in-the-Loop Pilot that tests three nationally significant challenges, pays qualifying contributors in stages, and is independently evaluated with a report to Parliament.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Endorse Experts in the Loop as a principle of Australia's approach to AI in public decision-making: a named, accountable human decides and answers for the result.
Endorse Experts in the Loop as a stated principle of Australia’s approach to AI in public decision-making: AI may inform, but a named, accountable human decides and answers for the result.
A program corrected obvious text-recognition errors in this quote for display.
Have parliamentary committees encourage responsible AI-assisted evidence through a short common standard: named human authorship, proportionate disclosure of material AI use, verifiable sources and conflict-of-interest declarations.
Recommend that parliamentary committees encourage responsible AI-assisted evidence through a short common standard: named human authorship, proportionate disclosure of material AI use, verifiable sources and conflict-of-interest declarations.
A program corrected obvious text-recognition errors in this quote for display.
Run a 12-month National Experts-in-the-Loop Pilot testing three nationally significant challenges, paying qualifying contributors in stages, independently evaluated with a report to Parliament.
Recommend a 12-month National Experts-in-the-Loop Pilot that tests three nationally significant challenges, pays qualifying contributors in stages, and is independently evaluated with a report to Parliament.
Ensure participation reaches beyond conventional credentials, including regional Australians, small and family businesses and people with lived experience, and have committee secretariats publish simple statistics on who makes submissions.
Ensure participation reaches beyond conventional credentials, including regional Australians, small and family businesses and people with lived experience, and that committee secretariats publish simple statistics on who makes submissions.
Build in integrity and intellectual-property protections from the start: clear licence terms, technology neutrality, identity verification of contributors and published selection criteria.
Build in integrity and intellectual-property protections from the start: clear licence terms, technology neutrality, identity verification of contributors and published selection criteria.
For parliamentary and policy evidence, retain final recommendations, value judgments and responsibility with identified humans or accountable public bodies; reserve decisions for authorised decision-makers.
For the parliamentary and policy-evidence functions considered here, final recommendations, value judgments and responsibility for acting on advice should remain with identified humans or accountable public bodies. Decisions remain with Parliament, ministers and other authorised decision-makers.
Require proportionate pilot evidence records covering contributors, material AI use and purposes, sources, assumptions, uncertainties, corrected or rejected outputs, conflicts and sign-off, without mandating software or vendors.
For work funded under the pilot (R3), each project would keep a short evidence record: the human contributors; the AI tools materially used and for what purpose; the sources relied on; key assumptions and uncertainties; significant AI outputs that were rejected or corrected; conflicts of interest; and who signed off. The record should be proportionate. A short submission needs a paragraph, not a dossier, and no particular software or vendor should be required.
Choose three pilot challenges independently: a time-critical AI, cyber or national-resilience risk, an economic or social wellbeing issue, and a challenge testing regional participation.
Scope. Twelve months, with three challenges chosen through an independent process: one time-critical AI, cyber or national-resilience risk; one economic or social wellbeing issue; and one designed specifically to test regional participation.
Provide nationally accessible participation with a protected verified-contributor register recognising experience, professional standing and community knowledge alongside academic qualifications.
A nationally accessible participation mechanism, supported by a protected register of verified contributors, that recognises demonstrated experience, professional standing and community knowledge as well as academic qualifications.
Encourage multidisciplinary teams and allow experts to nominate emerging problems through a simple early-warning channel.
Multidisciplinary teams should be encouraged, and experts should be able to nominate emerging problems through a simple early-warning channel.
Co-design First Nations knowledge participation with representatives, respecting Indigenous Cultural and Intellectual Property, community authority, consent and restrictions on disclosure or reuse; preserve everyone's right to unpaid parliamentary submissions.
Participation involving First Nations knowledge should be co-designed with First Nations representatives and respect Indigenous Cultural and Intellectual Property, community authority, consent and appropriate restrictions on disclosure or reuse. Nothing in the pilot should restrict any person's existing right to make an unpaid submission directly to Parliament.
Make pre-agreed participation payments to competitively selected contributors who meet contracted evidence requirements.
a pre-agreed participation payment for contributors competitively selected to address a defined challenge and who deliver the contracted evidence requirements;
Make quality payments after independent assessment of accuracy and usefulness.
a quality payment after independent assessment of accuracy and usefulness;
Cap recognition payments and base them on independently assessed public-interest contributions, never advocacy of a political, commercial or regulatory position.
a capped recognition payment based on independently assessed contribution to specified public-interest outcomes, never contingent on advocating a particular political, commercial or regulatory position;
Continue attribution when contributors' work is reused.
continuing attribution where a contributor’s work is reused.
Require assessors to disclose conflicts, apply published criteria and offer review of contested assessments; publish a government response to each challenge.
Assessors should disclose conflicts, apply published criteria and provide a review process for contested assessments. Each challenge should receive a published government response.
Invite the Department to cost the pilot options.
suggest the Committee invite the Department to cost the options.
Measure first-time, regional and small-business participation, verification errors, time to evidence, formal responses, adoption and cost per adopted recommendation.
Measures of success. The share of first-time contributors; regional and small-business participation; the error rate found at verification; the time from challenge to evidence package; the proportion of recommendations receiving a formal response; adoption; and cost per adopted recommendation.
Publish outputs as advice only, leaving decisions to Parliament and ministers.
Outputs are advice only and are published; Parliament and ministers Mai ble democracy decide.
Require conflict declarations, prohibit payment linked to positions taken and make all outputs public.
Mandatory conflict declarations; no payment linked to the position Creates a paid lobbying channel taken: all outputs public.
Require sourcing, pay for quality only after independent verification and remove contributors from the register for deliberate fabrication.
Sourcing standard; quality payment only after independent verification; M bl material removal from the register for deliberate fabrication.
Verify named contributors' identities, require human sign-off and conduct sampling audits.
Identity verification of the named contributor, human sign-off and Manageable; needs interference sampling audits.
Use technology neutrality, published selection criteria and merit assessment before revealing identity where practicable.
Technology neutrality; merit assessment before identity is revealed Ma ble consultancies where practicable; published selection criteria.
Let contributors retain pre-existing IP, give the Commonwealth a non-exclusive licence to submitted work, and provide a confidential channel for potentially patentable ideas.
Contributors keep pre-existing IP; the Commonwealth receives a non- exclusive licence to use submitted work; a confidential channel for Manageable Property potentially patentable ideas.
Cap the pilot, evaluate independently and decide whether to stop or continue at 12 months.
Capped pilot, independent evaluation and a stop-or-continue decision Tested by the pilot at 12 months.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.