Submission 360
Mr Jonathon Rose
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 'Evidence Integrity' approach to reduce bias and misinformation in generative AI answers, starting with user-pasted preference prompts. Outlines a staged path from public advice to voluntary provider controls to a possible legislated requirement, plus higher-assurance systems for the APS and Parliament.
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.
Proposes that users copy and paste one of two evidence-checking prompts into their AI chat and AI-assisted search engine.
The main part of this submission proposes using this existing capability to improve evidence integrity and reduce avoidable bias. Users are provided with and encouraged to copy and paste one of two prompts into their AI Chat and AI-assisted search engine.
A program corrected obvious text-recognition errors in this quote for display.
Proposes that government could ask major AI and AI-search providers to offer three choices: Off, Moderate and High.
Government could ask major AI and AI-search providers to offer three simple choices: © Off - normal operation. e@ Moderate - extra evidence and bias checks mainly for important or contested issues. e High - stronger source checking, counter-evidence, competing views and uncertainty analysis.
A program corrected obvious text-recognition errors in this quote for display.
Proposes that if testing shows material improvement, a legislated requirement for major providers to offer the controls is the stronger long-term option.
If independent testing shows that the approach materially improves information quality and reduces avoidable bias, the stronger long-term option is a legislated requirement for major providers serving Australians to make the controls available.
States that the proposed safeguard is not a content filter and would not prevent users from accessing lawful information or competing political views.
the proposed safeguard is not a content filter. These Evidence Integrity safeguards would not prevent users from accessing lawful information or competing political views.
Proposes a separate parliamentary evidence-assurance system for parliamentarians, ministers and political staff, institutionally separate from the APS system.
A separate parliamentary system could provide similar evidence assurance for parliamentarians, ministers and political staff while protecting parliamentary privilege, confidentiality and political independence. The APS and parliamentary systems should remain institutionally separate.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Test a simple Evidence Integrity personal-preference approach across major generative AI chats and AI-assisted search services; if it improves answers, make the tested instructions widely available and easy to try.
Australia should test a simple Evidence Integrity personal-preference approach across major generative AI chats and AI-assisted search services. If it improves answers, government should make the tested instructions widely available and easy to try.
Include major Chinese-language services used by Australians, including Qwen and ERNIE, in testing.
It should also be tested with major Chinese-language services used by Australians, including Qwen and ERNIE.
Keep public guidance short, current and specific to each service.
The exact features will change over time, so any public guidance should be short, current and service-specific.
If testing shows clear benefit, seek voluntary Off, Moderate and High evidence-checking controls from major AI and AI-search providers.
If testing shows clear benefit, the next step could be to make the same idea easy to select inside AI services. Voluntary provider option Government could ask major AI and AI-search providers to offer three simple choices: © Off - normal operation. e@ Moderate - extra evidence and bias checks mainly for important or contested issues. e High - stronger source checking, counter-evidence, competing views and uncertainty analysis. This would be easier for users than copying instructions, but uptake and design would depend on provider cooperation and could vary between companies.
A program corrected obvious text-recognition errors in this quote for display.
If independent testing demonstrates material information-quality improvement and reduced avoidable bias, consider requiring major providers serving Australians to offer controls across generative chats and relevant search/browser answers, defaulting to Moderate with Off/High choices.
If independent testing shows that the approach materially improves information quality and reduces avoidable bias, the stronger long-term option is a legislated requirement for major providers serving Australians to make the controls available. Under this approach, Moderate would be the default. It would add extra checking when the topic is important or contested. Users would remain free to select Off or High. The requirement would apply to generative AI chats and, where relevant, AI-generated answers in search and browser-based services.
For a mandated approach, explain the change and choices, monitor effectiveness and ensure provider compliance.
Government's ongoing role would then be much simpler: explain the change and its benefits, make clear that users can choose Off, Moderate or High, monitor whether the system is working, and ensure the company's compliance.
Measure the additional computing, energy, response-time and financial costs of Moderate and High settings in testing.
Testing should also measure the additional computing, energy, response-time and financial costs of Moderate and High Evidence Integrity settings, so that any improvement in information quality can be weighed against their resource use.
Invite further development of higher-assurance Evidence Integrity and Decision Assurance options for the APS and Parliament.
In parallel, the Committee should invite further development of higher-assurance Evidence Integrity and Decision Assurance options for the APS and Parliament.
Keep APS and parliamentary assurance systems institutionally separate; protect parliamentary privilege, confidentiality and political independence.
A separate parliamentary system could provide similar evidence assurance for parliamentarians, ministers and political staff while protecting parliamentary privilege, confidentiality and political independence. The APS and parliamentary systems should remain institutionally separate.
Consider APS assurance using protected government information, original-source tracing, source-independence checks, coordinated/foreign-influence checks, contradictory evidence, uncertainty and consequences.
An APS system could add stronger security and evidence checks, including protected use of government information, tracing important claims to original sources, checking whether apparently independent sources actually come from the same source, looking for coordinated or foreign influence, actively searching for contradictory evidence, and reporting important uncertainty and consequences.
Keep safeguards from filtering lawful information or competing political views, preserving users’ Off and High choices.
the proposed safeguard is not a content filter. These Evidence Integrity safeguards would not prevent users from accessing lawful information or competing political views. Their purpose is to broaden and test the evidence considered before an AI answers. Users would remain free to select Off or High.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.