Chair and Committee members, thank you for the opportunity to appear today to represent the more than 235,000 scientists, technologists, engineers, mathematicians and STEM educators who we represent.
Australian researchers have been developing and using artificial intelligence for decades — from machine learning and natural-language processing to advanced data analysis. Generative AI has brought some of these technologies into the mainstream, but AI is not new to research.
Researchers are simultaneously developers, users and evaluators of AI. Their experience offers important evidence about its opportunities, risks and barriers.
As we noted in our submission, Science & Technology Australia, in partnership with MetaMelb at the University of Melbourne, recently surveyed Australian researchers about how they use AI, based on a similar UK survey. A direct country comparison will be undertaken soon.
Our survey closed on 25 September with 1,034 responses. Eighty per cent were from higher-education institutions, with the remainder from public research agencies, medical research institutes and other organisations. Respondents represented a broad range of disciplines and career stages. We have undertaken some initial analysis, and I am pleased to have provided an early copy to the Committee.
That document shows that AI use is indeed already widespread in the research industry. Nearly three-quarters of respondents use AI, while 70 per cent reported moderate-to-extensive experience with generative AI tools. There is a slight concentration of AI use in administrative and supporting tasks.
The survey also identifies a gap between individual uptake and readiness. 45% said the AI training and skills-building opportunities provided by their institution were inadequate. Guidance was frequently described as unclear or poorly coordinated.
Computing capacity is another critical issue. With only around 10% of respondents quantifying their future requirements, figures should not be interpreted as a national demand forecast. Nevertheless, responses revealed a potential indicative demand of approximately 53.5 million CPU-hours and 14.3 million GPU-hours over the next 12 months — just from those who responded to this question. This is insightful given the 2026 National Computational Merit Allocation Scheme in 2026 was almost three times oversubscribed across Australia's two national facilities; the National Computational Infrastructure and the Pawsey.
Researchers also expressed clear perspectives on acceptable AI use. They generally accept AI as a tool to assist researchers to do their work. They drew a much firmer line when AI was substituted for human judgement, such as determining funding scores or assessing individual or institutional researcher performance.
These findings reinforce our submission recommendations, which had three central points.
First, Australia needs sovereign AI capability. But that capability must be built here, not merely hosted here, and it does not require Australia to replicate the entire global AI supply chain.
Second, Australia needs sustained investment in researchers, publicly accessible computing, secure data, technical support and long-term national infrastructure. International AI and hyperscaler. businesses operating here should make meaningful contributions to Australian R&D capability.
Finally, AI governance should be risk-based and human-centred. AI can assist researchers and institutions, but people must remain accountable for consequential decisions about research funding, careers and performance.
As the Government's own consultation paper on developing standards for AI infrastructure and training has noted, AI can lift productivity, accelerate discovery and strengthen Australia's resilience but only if we, collectively, get the settings right. Realising those benefits will require investment in people, infrastructure, research capability and public trust.
Thank you. I welcome the Committee's questions.
Ryan Winn
CEO, Science & Technology Australia
