The debate over the need for “sovereign artificial intelligence” (AI) is intensifying in Australia.
According to the CSIRO report, the country lags behind other countries in AI research and development. And experts warn that this is leaving us behind when it comes to leveraging AI to boost productivity and boost the economy.
Australia's AI lag
CSIRO points out that the underlying models, a type of machine learning (ML) algorithm trained on vast amounts of data, that power products such as OpenAI's ChatGPT are primarily manufactured overseas. .
At least 125 basic models have been developed around the world in recent years. Of these, most are manufactured by private technology companies. 73% come from the US, 15% from China, and most of the rest from Europe.
Anton van den Hengel is Professor of Computer Science at the University of Adelaide and Founding Director of the University's Australian Institute for Machine Learning (AIML).
he says universe “A big part of what we need in an AI economy is efficiency gains.” Introducing AI in industry will make goods and services cheaper and more competitive, he added. This could take another 20 years, van den Hengel says.
The remaining impacts of AI are new applications that do not yet exist and can only be realized through research. “Having advanced sovereign AI capabilities is essential for nations to be able to capture all of these economic opportunities,” van den Hengel said.
“We had a pretty streamlined and competitive AI research department around the world,” he added. “had”
“We have some very good software engineers, but they're not really ML engineers. And we have a very small number of very good researchers, but very few in between. I miss everything.”
What does sovereign AI mean?
“We're all impressed by the way these models write wedding speeches and poems. But the speed, power, and huge scale of data analysis they can achieve is what sets our biggest… It has the potential to help solve challenges, increase productivity and save lives,” said Dr Stefan Hajković, lead author of the CSIRO report.
“For example, the Fundamentals of Healthcare Model helps uncover the complex relationships hidden in patient health records, helping to reduce more than 140,000 medical costs. [human-error] Medical misdiagnosis occurs every year in Australia. ”
Van den Hengel added: “The simplest infrastructure would be Australia's large language model.”
“Everyone says why don't we just download the American model? It's not a million miles away.”
According to van den Hengel, it's not that simple, and such an approach can cause problems.
“These large language models form the core infrastructure on which all other AI is built. When you use someone else's model, everything else has to do with that person's value judgments, intonation, and preferences. The ranking will be reflected.
“If children in our schools are being educated using a system that Joe Biden considers a democratic leader, what does that mean for our national identity?”
CSIRO scientists agree.
“Tweaking existing models has significant advantages in terms of cost and speed of innovation, but using foreign models poses security and reliability risks,” says CSIRO's Digital and National Facility・Professor Elanor Huntington, Executive Director of Collections, said: “It could also result in tools that are not culturally appropriate in the Australian context or that do not deliver the benefits for workers that we want.”
education issues
Van den Hengel argues that research is essential to enabling AI-powered education and industry in Australia.
“If you want to have more AI-enabled teachers, or if you want to have more AI high school teachers, or if you want to have more AI courses in universities, you can't just go and print them. You have to train them. No. That means you need a trainer. You can't even print a trainer. Someone has to train the trainer.
“Graduating PhD students conduct research and then train their own PhD students, who in turn train PhD students in areas such as: their Own. Those doctoral students are then taught at universities and train undergraduate students. Undergraduate students then train teachers, and teachers train high school students.
“But the information has to come from somewhere. And where that comes from is PhD graduates. And the way to get more PhDs is by funding research. We are in the unenviable position of starting at least 10 years late in this process.”
Van den Hengel said the number of PhDs in AI in Australia is equivalent to that of one medium-sized US university.
Where is Australia's AI going?
“Australia performs very well in several focused research areas. We are very good at computer vision and have been for quite some time. It’s interesting because that’s the part of the AI that really drove the revolution,” Van den Hengel explains.
He also noted that Australian researchers have strong skills in natural language processing and robotics.
“There are about five or six really world-class groups in Australia,” but van den Hengel said they “are still not really operating on the scale that groups overseas are.” To tell.
He believes Australia has a chance to turn things around.
“This is an opportunity to increase productivity across all industries and build a globally competitive AI-based economy for children, but unfortunately we are currently in retreat.”
