Defence is laying more foundations for its contribution to pillar II activity under AUKUS, the barely-spoken-of tranche of work focused on AI, quantum and other emerging technology.
The Defence Science and Technology Group (DSTG) is in the process of sourcing up to three data engineers to prepare field data for ingestion into artificial intelligence and machine learning models.
Initially, the contracted team will focus on “optical imagery” before expanding into data preparation “for other sensors”, according to a statement of work.
The work feeds into the development of “advanced capabilities in artificial intelligence and autonomy” under AUKUS pillar II.
Most of the public-facing focus around AUKUS is on the delivery of nuclear-powered submarines, under what is officially known as Pillar I.
There is considerable industry interest, however, in Pillar II, which encompasses technology domains including AI, cyber security and quantum computing.
However, as the Center for Strategic and International Studies (CSIS) notes, Pillar II is the “the lesser-known – and poorly understood – part of AUKUS.”
“If pillar I was historic, then pillar II promises to be revolutionary,” CSIS writes.
Similar sentiment is apparent in some of the submissions to the active “Performance of the Department of Defence in supporting the capability and capacity of Australia’s defence industry” senate inquiry in Australia.
In a submission [pdf], data centre appliance maker SoftIron states that “nearly two years on from the AUKUS announcement … pillar II remains fuzzy, poorly defined and opaque in nature.”
“Whilst there is obviously, and quite properly, a limit to how much can be shared around such capabilities, the vacuum of publicly available information risks impeding the development of the agreement itself,” SoftIron states.
There was some public-facing movement in late May when “the first instance of jointly developed Australian, UK and US AI capability [was] deployed on coalition autonomous systems” for a UK-based trial.
Months later, the recruitment of a data engineering team to focus on data preparation work suggests more movement on the pillar II front is impending.
DSTG’s requirements are that the team works “independently” with the agency’s “support” to “acquire data which has been collected from exercises and operations”.
This will help identify “data collection gaps and inform data collection plans for future exercises and operations including ground truth data to simplify data preparation”.
On the data preparation front, the contracted team is expected to “prepare and analyse the data to develop diverse data sets fit for training, testing and validating models”, develop a data labelling taxonomy, as well as some “prototype machine learning models to determine fitness for purpose of the data for machine learning outcomes.”
“The data prepared through this contract will be used to support engagement with industry and academia in the development of AI and machine learning models,” DSTG notes.
“It will also be used within the test and evaluation pipeline to assess these models.”
While the initial work of this team is likely to be intensive, insofar as being a hands-on approach to data preparation, the DSTG requirements specifically asks the contracted team to investigate more hands-off approaches as well.
The agency is particularly interested in “automated approaches to data labelling potentially using weak supervision, semi supervised or active learning methods.”
These are all sub-domains of machine learning that aim to reduce manual effort in data labelling, potentially reducing the time to build a training dataset for a model.
iTnews filed detailed questions with Defence about the specific data preparation work and the extent to which pillar II capabilities more broadly have been stood up in Australia, but received a one-line response quoting the same requirements documentation.
