Artificial intelligence (AI) is set to reshape practically every sector of the economy over the next two decades, from healthcare through to agriculture, mining and transport. It has the potential to automate repetitive or dangerous tasks, increase productivity and inform the development of new products and services which change the way that we work and live.
Australia is world-renowned for the quality and volume of AI and computer science research we create, and as the national science agency, CSIRO’s work in artificial intelligence is 89% more cited than the global average, according to a Normalised Citation Impact, InCites (Web of Science Articles & Reviews, 2014-2018).
At CSIRO, AI is currently employed across multiple domains, from agriculture and climate science to energy distribution and precision health care. There are many fields of science within AI including: machine learning, natural language programming, deep learning (neural networks), computer vision and robotics. At D61+ LIVE 2019, we gathered leaders throughout CSIRO to explain how they are using artificial intelligence and machine learning to accelerate their science for social and environmental good.
How artificial intelligence (AI) and machine learning (ML) can advance the tools that we have
The combination of historical tools and cutting edge technology has created a plethora of new opportunities, said Dr Simon Barry, Acting Director of CSIRO’s Data61. With the availability of vast quantities of data and the high-powered tools required to understand it, new pathways to accelerate and optimise every aspect of the future have been created.
Using the example of computer processing, Data61’s interim Director explained that while deep learning was widely understood in the ‘80s and ‘90s, it wasn’t until the late 2000s that the methodology was applied to Graphical Processing Units. This collaboration resulted in more efficient training of Deep Learning Models used to perform simultaneous computer tasks.
The CSIRO has applied this concept to agriculture, allowing farmers to make more efficient and sustainable cultivation choices, explained Dr Jen Taylor, Research Group Leader of CSIRO’s Agriculture and Food. “There’s a 70% uptake of unmanned tractors and similar machines rolling around farms making decisions about water, pests and pathogens without a farmer walking up and down the fields.”
“It’s so much more precise and efficient, which is fantastic for the whole system.”
Optimising Australia’s energy network using AI and ML
CSIRO’s Energy Business Unit is using AI and ML to model Australia’s extensive energy supply network via a digital simulation, known as a Digital Twin, to improve efficiency, disaster management, and emerging power sources. This process will also help manage over investment and energy distribution in the grid of the future, said CSIRO’s Energy Business Unit Director, Dr Tim Finnigan. He added that AI and ML is crucial to developing an efficient and adaptable system that can be used up until 2040 and beyond.
When it was originally created, Australia’s energy grid was designed to manage only a small number of energy sources, such as coal and gas, however, now there are millions of different derivations, explained Dr Finnigan. A Digital Twin simulation provides insights and results from various methods that could be used to update the network, with automatic programming able to determine the most effective solution.
"We’ll avoid over investment in that, and that could be in the billions in savings. And that’s just one benefit.”
“In the situation where you have faults in the system or environmental impacts, for example bushfires, which with climate change we are expecting more and more intense bushfires with higher peak temperatures, how will a grid system in those environmental conditions operate in the future, and how can it potentially repair itself if it is integrated with some machine learning?
If a fault goes out in a certain part of the grid, can machine learning help the system repair itself or reroute power so that we don’t get a South Australia state-wide blackout? We can actually have a smart enough and well-trained grid that could figure out other pathways for power to flow.”
Building solid and dependable models for climate analysis
Machine learning is being used by CSIRO’s Climate Science Forecasting team to help better understand complex environmental patterns and emerging weather trends such as natural disasters like floods and droughts by mining insightful historical data, says team leader Dr Terry O’Kane. Being able to find a common underlying driver in similar climate patterns by using this cutting-edge technology to organise immense amounts of statistical information could help predict the future of Australia’s atmospheric conditions.
"We actually don’t really know the dynamics that drive the major climate teleconnection modes,” explained Dr O’Kane, “so if you think of the El Nino southern oscillation, which drives a lot of the large scale weather in Australia, and it’s relation to something like the Indonesia Dipole, the cause and relationships between these large teleconnections (climate anomalies being related to each other at large distances), we only really know them imperially, and machine learning offers a tool for a purely data-driven approach to pull out and mathematically underpin those relations.”
This approach has been applied to analyse drought prediction, a multi-faceted challenge with a plethora of variables that must be considered when forecasting the future. "The work we’ve been doing is to try and find, in a mechanistic sense, what other common underlying drivers
"We have a better understanding of what we don’t know, which is a lot.”
Using AI to guide farmers future plans
As the Research Leader of CSIRO’s Agriculture and Food department, Dr Jen Taylor appreciates how essential climate prediction is for Australia’s agriculture and aquaculture industries. Dr Taylor explained how a single gram of soil can contain billions of microbes that can, or cannot help a plant grow, machine learning and artificial intelligence provides a system that can interpret a complex growth system to ensure future crops set up for success.
“Farmers have been working hard to get efficient practises for a while now, but there’s a lot more urgency in the system. Everything needs to get a lot more efficient, more sustainable. So farmers need to make some fairly complex decisions about how they use their water, whether they need to worry about pests or pathogens and that sort of stuff, and there are systems around that do that.”
Using the example of state-of-the-art wearable and hands-free sensors Data61 developed for prawn harvesters, Dr Taylor explained that AI and machine learning is helping the nation’s farmers making instantaneous decisions about best practise system management and preventing waste. “There’s a booth outside that’s been talking about how they help aquaculture farmers monitor the water quality in their pond.”
“We wouldn’t have been able to do that without sensors and a way to interpret that data that’s coming in.”
Understanding genetic diseases
Being able to extrapolate data when there are 3 billion letters in the genome was once a near-impossible task before machine learning, says Dr Denis Bauer, CSIRO’s
principal research scientist in transformational bioinformatics. With a research pool of 22,000 individuals with motor neurone disease, each who have 2 million genetic differences between them, equalling roughly 80 million features that need analysing, the resulting matrix included 1.7 trillion statistical examples. A dataset of this size could not be analysed fully if using traditional methodology, explains Dr Bauer. “Using machine learning, particularly when we’re using it as a random forest, that builds individual decision trees that look at subsets of the data, and then ultimately come up with a model that captures the whole 3 billion letters for the genome and
In partnership with the Dementia Team Grant led by Prof Ian Blair of Macquarie University, CSIRO’s Transformational Bioinformatics used machine learning to analyse the full genome simultaneously, identifying the genomic location that jointly contributes to the disease rather than being limited to detecting only the strong individually contributing genes. This approach resulted in the effective analysis of genomic data from sizeable sample pools, boosting the statistical power needed for disease gene classification. “Now, for the first time, we’ve built our own machine learning model that deals with this ultra-high dimensional data that others have not dealt with before. For example, the Googles and the Yahoos of the world typically deal with
“It’s nice to know that a little Australian team is at the forefront of big data analysis.”