In this thought piece, NSSN Co-Director Professor Julien Epps, explains how Artificial Intelligence is helping important societal and industry challenges and why it is a core part of NSSN projects.

Artificial Intelligence has received a huge amount of media attention in recent times, in particular generative AI.

However, “AI” is a broad suite of tools, including modelling, prediction, and classification.

It’s important to understand how other types of AI are already being beneficially applied to a huge range of important societal and industry challenges, and what the future might hold.

It was clear to NSSN’s founders that – as well as creating and deploying innovative new sensor technologies – the clever use of raw sensing data was essential to translate measurements of the environment into actionable insights.

This is why AI has been a core part of our growing portfolio of projects from the very beginning.

A couple of examples of smart sensing projects which have used other types of AI are our flood intelligence and wildlife monitoring projects.

The Improved Operation Flood Intelligence project is developing a proof-of-concept software system using modelling and prediction to help the NSW State Emergency Service plan for flood events that can stretch for weeks and months as flood waves travel downstream.

Drawing on raw water level data from a range of sensors along river systems, the system will combine machine learning methods with uncertainty estimates and schematics to provide new intelligence on where flooding may cause the greatest impact.

The Improved Operation Flood Intelligence project could help the SES plan for flood events like those which affected Windsor in 2021 (PHOTO). The NSW town experienced dramatic flood heights in February 2020, March 2021 and March, 2022. Credit: Shutterstock.

EcoEar is an acoustic (audio) sensing device that uses classification, a form of machine learning, to automatically detect koala calls from a diverse background of environmental sounds.

It is designed to automatically classify the raw audio signals within the sensor device itself, using optimised low-power computation so that only the relevant koala calls are marked and saved for further analysis.

This saves time for ecologists, who no longer need to listen through days of sound recordings to estimate koala populations.

The classification model is pre-trained on previously collected koala calls from the NSW government, so that it can achieve a high enough accuracy for use in the wild.

Interestingly, smart sensing projects like these have mostly not used generative AI, the type of AI that has been mainly in the news over the past year, but have used other forms of AI: modelling, prediction or classification.

There are two reasons why we could expect that the future for AI is likely to be even more exciting than what we’ve heard in the media.

Firstly, many from the world of research translation will be familiar with the hype curve, which is one way of expressing Amara’s Law.

Amara’s Law says that we tend to overestimate the impact of a new technology in the short term and underestimate its impact in the long term.

For those of us who are old enough, think back to the dot-com boom.

Who could forget the wild speculation in the early days of the Internet, even though most websites were extremely basic (sometimes just “Under construction…”) and used by very few?

Now it is almost impossible to imagine a functioning first-world society without the Internet, which is more important than our wildest predictions from the dot-com era.

There are many other technologies that have experienced the same hype trajectory.

This suggests that after the hype of generative AI has subsided, the real work of applying it meaningfully to business and societal problems will make its biggest gains.

Secondly, generative AI is just one part of what we call “AI”, which includes a broader suite of tools to gain insight from large amounts of data.

Most systems that collect large quantities of sensor data can benefit from some form of AI to automatically ‘make sense’ of the data, e.g., to plan better for a flood that is in progress or select only the koala calls from thousands of hours of recordings.

For this reason, AI is likely to have a place in most systems that include a sensor, and our understanding of how best to use it will only improve further.

On top of this, there is growing consensus that industry must adopt AI or fall behind.

The good news is that although many of the innovations in the news have been coming from the US, Australia has deep capability in AI, ranking seventh in the world for AI research and development, even though it trails the OECD in business adoption of AI.

Further, there are recent calls to increase investment in AI research and development.

In the years ahead, the deep neural networks that sit underneath generative AI, and their many variants, will also provide increasingly accurate systems for generating actionable insights from raw sensor data.

These systems will depend heavily on the availability of large amounts of data that are representative of the problem to be solved, just as generative AI’s impressive abilities depend on hundreds of billions of words of training data.

So, they will work best when we can accumulate lots of data to train them with.

For those considering introducing AI to their solutions, it is never too early to begin, because these systems are likely to steadily improve in accuracy over time, well after the generative AI media storm has blown over.

Perhaps the best news of all is that partnering with the NSSN simplifies the process of engaging with hundreds of researchers with deep AI expertise in universities, by creating a single point of contact for NSW's and ACT's leading research-intensive universities.

Together, we can build on this expertise and technology to create enduring solutions to some of society’s big problems.

Professor Julien Epps is Co-Director of the NSSN

Koala monitoring, floods: how AI is benefiting smart sensing projects