Australia’s AI challenge

Think ChatGPT is the big problem? When it comes data Aus is undergunned

The Kingston Group of professors of AI* sets out the challenge and some responses in a new paper.

First up, Australia must work out how to do more work with what we have got. “The core research design for Australia to solve is how to train AI faster and more accurately with smaller amounts of data and in close collaboration with humans.

“We need this ‘small data’ capability to give us a competitive edge against much larger economies and companies that rely more on vast datasets, computing infrastructure, and a larger engineering workforce to build their AI capability.”

And that will take a national strategy, says the Kingston Group (as in the Canberra suburb where they met).

What is needed

* “research into the careful and clever use of mathematics and coding to achieve disproportionate performance with more limited resources. Some of the specific research areas include: efficient learning, lifelong learning, transfer learning, expert knowledge, common sense and reasoning.”

* “a profound increase in skills development at all levels,” from VET techs to core research and postgraduate skills development “at between two and ten times the current rate to match investment by global peers.”

The Kingston Group is, * Joanna Batstone (Monash U) * Peter Corke (QUT) * Stephen Gould (ANU) * Richard Hartley (ANU) * Anton van den Hengel (Uni Adelaide) * Sue Keay (OZ Minerals; Robotics Australia Group) *  Dana Kulić (Monash University) * Jie Lu (UTS) * Simon Lucey (Uni Adelaide) * Michael Milford (QUT) * Ian Reid (Uni Adelaide) * Ben Rubinstein (Uni Melbourne) * Svetha Venkatesh (Deakin U) * Toby Walsh (UNSW)