Submission 78
Transition Level
Submitter group: Companies.
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.
A commercial AI adviser recommends three priorities for AI adoption policy: measuring practical benefits, developing human expertise, and testing whether oversight works. Asks for evaluation of publicly supported adoption programs, funded research on skill formation, and evidence of effective human oversight in Commonwealth AI deployments, while cautioning the cited studies do not show economy-wide returns or general workplace deskilling.
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 evidence of AI benefits warrants adoption and evaluation in Australian workplaces but does not establish an economy-wide return.
AI can deliver substantial benefits. We know this from mounting evidence. Findings like in [1] warrant adoption and evaluation in the settings where Australians work; they do not establish an economy-wide return.
Recommends vendor-neutral practical support for SMEs to manage change, select tasks, redesign workflows and assess results.
Provide SMEs with practical, vendor-neutral support to manage cultural change, select suitable tasks, redesign workflows and assess results.
Argues that performance with AI and learning to perform independently are different outcomes, citing mounting evidence.
Performance with AI and learning to perform independently are different outcomes. This, too, has mounting evidence.
Argues that oversight should test whether staff can identify errors and challenge outputs, rather than rely on a human-in-the-loop principle, citing evidence that humans are poor at active monitoring.
Test whether designated staff can identify material errors, challenge outputs and activate fallback arrangements under realistic conditions. Reassess after material system or workflow changes, and use the results to improve implementation guidance. The oversight should not be designed with the principle of just having a ‘human-in-the-loop’; there is strong evidence humans are bad at active monitoring.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Evaluate publicly supported AI adoption programs against work quality and productivity, including verification, rework and implementation costs.
Evaluate publicly supported AI adoption programs against improvements in work quality and productivity, including verification, rework and implementation costs.
Provide SMEs with practical, vendor-neutral support to manage cultural change, select suitable tasks, redesign workflows and assess results.
Provide SMEs with practical, vendor-neutral support to manage cultural change, select suitable tasks, redesign workflows and assess results.
Publish findings from adoption programs, including unsuccessful trials, to help other businesses make informed choices.
Evaluate publicly supported AI adoption programs against improvements in work quality and productivity, including verification, rework and implementation costs. Provide SMEs with practical, vendor-neutral support to manage cultural change, select suitable tasks, redesign workflows and assess results. Publish findings, including unsuccessful trials, to help other businesses make informed choices.
Fund evaluations with employers and education providers of how AI affects skill formation and retention.
Fund evaluations with employers and education providers of how AI affects skill formation and retention.
Where independent competence matters for professional judgement or resilience, test learning approaches and assess unassisted performance over time alongside assisted productivity.
Where independent competence matters for professional judgement or resilience, test learning approaches and assess unassisted performance over time alongside assisted productivity.
Use the skill-formation findings to update training and entry-level development pathways.
Fund evaluations with employers and education providers of how AI affects skill formation and retention. Where independent competence matters for professional judgement or resilience, test learning approaches and assess unassisted performance over time alongside assisted productivity. Use the findings to update training and entry-level development pathways.
Require proportionate evidence of effective oversight for consequential Commonwealth AI deployments, through procurement and assurance processes.
Require proportionate evidence of effective oversight for consequential Commonwealth AI deployments through procurement and assurance processes.
Test staff’s ability to identify material errors, challenge outputs and activate fallbacks realistically; reassess after material system or workflow changes and use results to improve implementation guidance.
Test whether designated staff can identify material errors, challenge outputs and activate fallback arrangements under realistic conditions. Reassess after material system or workflow changes, and use the results to improve implementation guidance.
Do not design oversight around a 'human-in-the-loop' principle alone; use other, better methods, citing evidence that humans are poor at active monitoring.
The oversight should not be designed with the principle of just having a ‘human-in-the-loop’; there is strong evidence humans are bad at active monitoring. Other, better, methods exist and should be utilised.