AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 296

Mr Biranchi Poudyal

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.

Open in the interactive guide

Summary

Read the original submission for the full text.

A doctoral candidate, writing in a personal capacity, recommends a national pilot testing whether higher education providers can make accountable, explainable decisions about AI use while preserving students' judgement. Proposes assessment redesign, fair AI decision procedures and procurement safeguards, and describes its supporting studies as provisional and documentary.

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.

  1. Says Australia's AI capability depends partly on people who can question outputs, verify evidence and take responsibility for decisions.

    Australia’s AI capability depends partly on people who can question an output, verify its evidence and take responsibility for a decision.

  2. Argues that naming a person responsible for AI does not establish workable criteria, authority or review, and that education policy must address both problems.

    Equally, an institution naming a person responsible for AI does not establish that the person has workable criteria, sufficient authority or a review process. Education policy must address both problems together.

  3. Recommends that assessments make explicit where students must show their own reasoning and where AI collaboration is an intended outcome.

    For term (g), the Department of Education should support providers and professional bodies to identify where students must demonstrate their own reasoning and where competent AI collaboration is an intended outcome. Assessments should make those expectations explicit before students begin.

  4. Argues that AI-detection scores must never alone establish misconduct; cites bias in tested detectors while cautioning that the study does not establish every current product’s performance.

    AI-detection scores should never be the sole basis for misconduct findings. Liang and colleagues found bias against non-native English writing in the detectors they tested. That study does not establish the performance of every current product, but it reinforces the need for independent evidence and procedural safeguards.

  5. Recommends model procurement clauses on where educational data is processed, who can access it, training use, deletion and exit.

    develop model clauses requiring providers to know where educational data is processed, who can access it, whether it is retained or used for training, and how deletion and exit work. Student work and research materials should not default to vendor-model training.

  6. States that the Higher Education Standards Framework already addresses assessment, integrity and governance, so the pilot should test implementation within it.

    The Higher Education Standards Framework already addresses assessment, integrity, complaints and governance. [4] The pilot should test implementation within that setting.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Fund a sector-led pilot, coordinated by the Department of Education with TEQSA, testing documented AI decision pathways.

    Ask the Australian Government to fund a sector-led pilot, coordinated by the Department of Education with TEQSA’s participation, testing documented decision pathways against realistic AI scenarios.

  2. Include diverse providers, educators, students and independent evaluators in the Education/TEQSA pilot; adapt the author’s governance stress testing into an openly documented, reproducible procedure.

    The Department of Education should coordinate an initial pilot with TEQSA, diverse providers, educators, students and independent evaluators. It should adapt my governance stress-testing approach into an openly documented procedure that other institutions can reproduce.

  3. First fix a dated set of policies, procedures and assessment instructions; apply common scenarios covering routine assistance, conflicting instructions, accessibility adjustments, AI-assisted marking and delegated action, allowing supplementary local scenarios.

    Each provider would first fix a dated set of relevant policies, procedures and assessment instructions. Reviewers would then apply a common set of scenarios, including routine assistance, conflicting instructions, an accessibility adjustment, AI-assisted marking and a tool taking delegated action. Local scenarios could supplement the common set.

  4. Require pilot evidence of applicable rules and conflict priority, authority and its limits, procedures and response opportunities, decision criteria, privacy/accessibility/fairness safeguards, unsafe-use stopping authority, reasons, reconsideration and remedies.

    Evidence that the provider must identify Applicable rule Which current provision governs this use, and which rule prevails if documents conflict? Authority Who may authorise, restrict or decide the case, and what is the limit of that authority? Procedure What steps must staff and students follow, including notice and an opportunity to respond? Decision criteria What evidence distinguishes permitted assistance, inadequate learning and misconduct? Safeguards What protects privacy, accessibility and fairness, and who can stop an unsafe use? Review How can a person obtain reasons, seek reconsideration and secure a remedy?

  5. Record exact supporting clauses and explicit, ambiguous or absent pathways; do not offset missing essential safeguards with strong results elsewhere; assign unresolved cases an owner, correction date and retesting.

    Reviewers should record the exact supporting clauses and distinguish an explicit pathway, an ambiguous pathway and no identified pathway. Strong performance on one check should not cancel a missing essential safeguard. Unresolved cases need a named owner, a correction date and retesting.

  6. Independently double-review a sample and document disagreements; conduct staff/student walkthroughs comparing decisions, located evidence and time required to test practical usability.

    A second reviewer should independently assess a sample and document disagreements. The pilot should also run staff and student walkthroughs. Documentation can appear complete yet remain unusable. Compare the decision each person reaches, the evidence they locate and the time required. This adds a practical check that my documentary studies cannot provide.

  7. Test implementation within existing higher-education standards; separately assess legal authority, benefits and administrative costs before any later mandatory requirement.

    The Higher Education Standards Framework already addresses assessment, integrity, complaints and governance. [4] The pilot should test implementation within that setting. Any later mandatory requirement would require a separate assessment of legal authority, benefits, and administrative costs.

  8. Run an 18-month pilot: convene partners, select diverse providers and publish protocol, baselines and costs in months 1–3; test, change assessments and retest in months 4–12; independently evaluate and publish de-identified findings in months 13–18.

    I propose an 18-month pilot from commencement. During months 1-3, the Department of Education should convene the partners, select a diverse group of providers, and publish the protocol, baseline measures and costed work plan. Months 4-12 should cover testing, assessment changes and retesting. Months 13-18 should support independent evaluation and publication of de-identified findings.

  9. Support assessment redesign demonstrating independent understanding and evaluation of AI-assisted work, using discipline-specific methods and accessible alternatives.

    Support assessment redesign that demonstrates students’ independent understanding and their ability to evaluate AI-assisted work, with discipline-specific methods and accessible alternatives.

  10. Have the Department of Education support providers and professional bodies to identify where students must demonstrate their own reasoning and where AI collaboration is an intended outcome, with assessments making this explicit in advance.

    the Department of Education should support providers and professional bodies to identify where students must demonstrate their own reasoning and where competent AI collaboration is an intended outcome. Assessments should make those expectations explicit before students begin.

  11. Develop model procedures separating permitted assistance, disclosure errors and misconduct, with reasons, a chance to respond and human review.

    Develop model procedures that separate permitted assistance, disclosure errors, and substantive misconduct, with reasons, a meaningful opportunity to respond, and human review.

  12. Have pilot partners distinguish permitted assistance, actual conduct, disclosure compliance and demonstrated learning; base responses on evidence and applicable policy, without automatically equating disclosure errors with substantive misrepresentation.

    For term (i), the pilot partners should develop model procedures that ask what assistance was permitted, what the student did, whether the disclosure met the stated requirement, and whether the learning outcome was demonstrated. The evidence and applicable policy should guide the severity and nature of a response. A disclosure error in otherwise permitted work should not automatically attract the same response as substantive misrepresentation.

  13. Require adverse findings to state rules, evidence and reasons, allow a response and provide reconsideration by a human reviewer; never base misconduct findings solely on AI-detection scores.

    An adverse finding should identify the rules, relevant evidence, and reasons; allow the student to respond; and provide a review by a person who can reconsider the decision. AI-detection scores should never be the sole basis for misconduct findings.

  14. Publish pilot evidence on learning, fairness, workload and governance gaps before deciding whether further obligations are warranted.

    Develop common contractual requirements for educational AI and publish evidence from the pilot on learning, fairness, workload, and governance gaps before deciding whether additional obligations are warranted.

  15. Have Education convene providers and procurement specialists to develop clauses covering data processing location, access, retention/training use, deletion and exit; prevent student work and research materials defaulting to vendor-model training.

    the Department of Education should convene providers and procurement specialists to develop model clauses requiring providers to know where educational data is processed, who can access it, whether it is retained or used for training, and how deletion and exit work. Student work and research materials should not default to vendor-model training.

  16. Require meaningful human oversight and a challenge route for uses affecting marks, progression or discipline.

    Uses affecting marks, progression or discipline should require meaningful human oversight and a route to challenge the outcome.

  17. Require contracts to provide relevant audit information, material model-change notices, incident reporting and the ability to suspend use.

    Contracts should provide relevant audit information, notice of material model changes, incident reporting and the ability to suspend a use.

  18. Evaluate unresolved scenarios before/after correction, reviewer agreement, staff time, students’ explanation/verification of AI-assisted work and review accessibility; report complaint outcomes alongside access and compare student groups with consent/privacy protections.

    Evaluation should report unresolved scenarios before and after correction, reviewer agreement, staff time, students’ ability to explain and verify AI-assisted work, and the accessibility of review procedures. Report complaint outcomes alongside access to complaints: fewer complaints alone do not establish improvement. Compare outcomes across relevant student groups with appropriate consent and privacy protections.

  19. Fund staff release, student participation and evaluation; condition scale-up on demonstrated improvements relative to burden while preserving providers’ discretion over teaching methods.

    Funding should cover staff release, student participation and evaluation. Scale-up should depend on demonstrated improvements relative to burden. Providers should retain discretion over teaching methods,

Read the original

Open in the interactive guide