Submission 396
Ms Lotus Harkness
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.
An individual practitioner argues AI can support disabled people’s work when users can refuse, declare, check and leave, but becomes coercive when users absorb liability without authority. Recommends literacy funding, human attestation, separating present harms from catastrophe, and a Safety Institute with testing access and publication duties. Says daily practice does not establish provider safety or resolve catastrophic risk.
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 four questions to ask before locking in an AI rule: can the person refuse the output, declare use of the tool, check it, and walk away.
Before you lock in a rule for AI, we should ask four things. Can the person refuse the output? Can they declare they used the tool? Can they check it? Can they walk away?
Argues AI can be assistive infrastructure improving disabled users’ access to work and services and belongs in the opportunity discussion.
Put disabled use in the opportunity chapter. Report where AI is assistive infrastructure, where it improves access to work and services, and where blanket bans are blocking that use. Vulnerable Australians are users, not only people to whom AI is done.
Argues that intensive or disabled AI users should not be classified as a hybrid legal class.
Do not classify intensive users or disabled users as a hybrid legal class. Consent, disability rights and accountability attach to the person.
Argues that speculative loss-of-control should be distinguished from present implications such as fraud, deepfakes, child safety, labour intensification and accessibility.
Catastrophe scenarios can be examined without emptying the inquiry of present harm and present gain. We must distinguish speculative loss-of-control, from real implications like fraud, deepfakes, child safety, labour intensification, and accessibility.
Argues for funding practical AI literacy for the APS, small business and exposed occupations as a risk control and a prospect for job creation.
Fund practical literacy for the APS, small business, and exposed occupations. Fear without skill produces either rejection or undeclared use. Skill is how the first design is taught. This is a huge prospect for job creation in this country.
Argues that an empowered Safety Institute with access and a duty to publish is preferable to a pause without an institution that can test.
A pause without an institution that can test is a press release. An empowered Safety Institute with access and a duty to publish is a state.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Require data centres to pay for new power and river water use through planning.
Big computer sheds should pay for new power and not drink the river for free. That is planning.
Allocate resources both to speculative loss-of-control and to present implications such as fraud, deepfakes, child safety, labour intensification and accessibility.
We must distinguish speculative loss-of-control, from real implications like fraud, deepfakes, child safety, labour intensification, and accessibility. We must allocate resources however to both to deal with both (again prosperity creation).
Write a two-design test into the report: before locking in a rule for AI, ask whether the person can refuse the output, declare use of the tool, check it, and walk away.
Write the two-design test into the report with rules/skills and prompts. Before you lock in a rule for AI, we should ask four things. Can the person refuse the output? Can they declare they used the tool? Can they check it? Can they walk away?
Fund practical AI literacy for the APS, small business and exposed occupations.
Fund practical literacy for the APS, small business, and exposed occupations. Fear without skill produces either rejection or undeclared use.
Keep consent, disability rights and accountability attached to people; do not classify intensive or disabled users as a hybrid legal class or recode current users when discussing robotics and future systems.
Do not classify intensive users or disabled users as a hybrid legal class. Consent, disability rights and accountability attach to the person. Discuss robotics and future systems without recoding current users.
Give the AI Safety Institute the job of frontier evaluation, and the Office of AI, education and industry the job of fraud, deepfakes, child safety, labour design and access.
Give the Australian AI Safety Institute the job of frontier evaluation. Give the Office of AI, education and industry the job of fraud, deepfakes, child safety, labour design and access.
Make human attestation the default for official AI use, naming the assistance and the verifying human, who should have the right to refuse the generation.
Make human attestation the default for official use and pair it with authority. Name the assistance. Name the human who verified it. Give that human the right to refuse the generation.
Report on AI as assistive infrastructure in the opportunity chapter, including where it improves access to work and services and where blanket bans block that use.
Put disabled use in the opportunity chapter. Report where AI is assistive infrastructure, where it improves access to work and services, and where blanket bans are blocking that use. Vulnerable Australians are users, not only people to whom AI is done.
Weight practitioner evidence: organisational briefs should be heard but not replace named, accountable use as evidence of what systems do now.
Organisational briefs on pacing, data centres and Safety Institute powers should be heard. They do not replace named, accountable use as evidence of what systems do now.
If Australia attaches conditions to large training deployments, make them verifiable, supported by an empowered Safety Institute with access and a duty to publish.
If Australia attaches conditions to large training deployments, make them verifiable. A pause without an institution that can test is a press release. An empowered Safety Institute with access and a duty to publish is a state.