The future of AI

ChatGPT has thrust the power and possibility of artificial intelligence (AI) into the mainstream. 

The disruptive technology has been used by students to complete assignments to politicians delivering speeches.

Examples emerge every day of its capacity to translate human thoughts and prompts into prose that is almost indistinguishable to that created by humans.

The fast-paced evolution of AI technology has divided opinion, with critics concerned bots like ChatGPT will cost humans jobs and erode true personal interaction.

But global estimates are that AI will contribute up to USD$16 trillion to the global economy by the end of the decade. CSIRO research has predicted Australian industry will need up to 161,000 new specialist workers in AI by 2030.

Adelaide is at the forefront of the research and development in machine learning and computer vision through the University of Adelaide’s Australian Institute for Machine Learning (AIML) based at Lot Fourteen.

Machine learning underpins business models of the largest corporations and has the potential to deliver massive social, economic and environmental benefits.

AIML’s director Professor Simon Lucey, says AI is not something to be feared but has huge opportunities to enhance Australia’s industries, create new jobs, provide better healthcare and drive business efficiency.

“AI is set to transform every aspect of our lives. It’s a revolution and we’re living through it right now,” he says.

“The real value of AI is not in the technology itself, but the huge improvements it’s making across a range of areas.

“In agriculture, AI is being used to improve harvest yields and help manage environmental resources and in medicine; researchers are using AI to improve patient diagnosis and reveal insights to deliver better health care outcomes, both at a population and an individual level.”

“To meet the huge demand for AI skills in the next few years, we really need to ramp up our education and training efforts, to make sure we have great diversity of talent across the whole tech sector.”

Doubling in size since moving in as Lot Fourteen’s first tenant, AIML is now home to more than 170 machine learning researchers, engineers, and postgraduate students. This core of AI talent -particularly in computer vision – has proven a powerful lure for multinational tech companies seeking to grow their AI capability; with both Amazon and Microsoft choosing to establish operations at Lot Fourteen.

Meet two PhD students at AIML who share the critical research they are doing right here at Lot Fourteen, and their thoughts on AI.

Georgia Kenyon, 3rd Year Neuroimaging Analysis PhD Student, Australian Institute for Machine Learning

Q&A with Georgia Kenyon: Using AI systems for diagnostic decisions

What is the research you are doing at AIML?

I’m on a joint PhD between the University of Nottingham (UK) and the University of Adelaide, and I am working on using AI, specifically deep learning, to analyse brain MRI scans. My project focuses on combining the knowledge of human anatomy with deep learning models, to improve the segmentation of the brain into its multiple structures. I am particularly focused on extracting the blood vessels from the brain, which are often ignored. Understanding where vessels are in the brain can help to identify anomalies like microbleeds to doctors, as increased brain microbleeds can indicate risk factors such as stroke.

So, using machine learning to highlight areas of the anatomy on imaging that might be more difficult to see?

Definitely – this could be a way of highlighting normal anatomy, or it could also highlight issues we don’t expect to see. So, you could have the capacity to highlight anomalies like tumours; or in my case, microbleeds. The difference between my project and a lot of other deep learning projects, is that I am going to focus on incorporating what we know about anatomy and training the deep learning models in a way that provides almost a set of ‘rules’ that should help deliver more anatomically plausible segmentations.

Why are people interested in using machine learning on medical imaging such as MRIs?

When we go to the doctor, it’s very hard to know what’s going wrong without doing some sort of imaging, whether it’s blood tests, CT or MRI scans. And at times it can be very difficult to see certain aspects of the body in these scans, particularly at fast pace. Having something that could split a brain MRI into its different structural parts, or highlight anomalies, can be beneficial for a doctor because it can help highlight something that they may not otherwise notice immediately.

Will this replace doctors?

No. It’s just going to potentially be a really good tool that could improve efficiency and workflow for doctors.

What would the ideal outcome be for your research in 10 years’ time?

I would love to see aspects of my work go into clinical practice and used on different aspects of the body, not only blood vessels. For medicine, I see the benefits of multi-modality, combining a patient’s healthcare data to create a good patient-to-doctor clinical workflow, that results in optimised care for every patient.

What advice would you give to students about to leave high school, or girls and women thinking about a career in STEM?

Follow your passion. I think whatever field you go into; technology and artificial intelligence is going to have an impact. So, pursue your passion but then keep thinking about how technology can improve what you’re doing, and hone your skills around that.

Lana Tikhomirov, PhD Candidate, Australian Institute for Machine Learning

Q&A with Lana Tikhomirov: Training AI for health outcomes

What is your research at AIML?

I’m a cognitive scientist in Artificial Intelligence (AI) safety research. Cognitive scientists test and understand unseen thinking processes such as decision making, memory and language. I’m trying to understand what happens to these processes when you use AI. Specifically, I am investigating radiologists’ decision making when they use AI systems for diagnostic decisions, as this is the most common application of AI in medicine. This knowledge is going to be important from a safety and ethical perspective, because we’ve seen previously that there are a lot of difficulties in introducing technology to humans – especially in high-risk workplaces. Through my work, I hope to gain better understanding of how AI impacts decisions, then we can design AI that complements the decision making of radiologists.

What would be the ideal outcome of your research and impact in ten years from now?

Firstly, that we would be able to better outline why we should create and implement AI, so it can complement an existing task that humans want to do better. Secondly, would be to create a framework for designing and implementing AI that tries to minimise any misuse or overuse of AI.

What might misuse and overuse be?

It could be examples where the technology may not, for instance, account for a particular demographic or a minority. And, if you accept that decision right off the bat, you could end up having a worse outcome for a patient. There may also be cases where AI can’t pick-up incredibly rare diseases, or sometimes there’s certain contextual knowledge that humans have that AI can’t understand.

Where do you see AI going in the future?

I see two potential outcomes. The first is where AI is used to assist humankind in areas where we need it, and we understand why we need the technology. The alternative is where we can sometimes unfairly erase certain human decisions in favour of AI, because we don’t understand the value of those human decisions. These outcomes depend on how well we have a good ethical and safe societal approach to AI.

What career advice would you give a high school student wanting to enter a similar field?

I think the most important area in the future will be human and AI interaction and implementation. We don’t just need people to create AI, we need people to understand what its applications are. That means if you enter a human or social science, you will be important in these spaces. 

AI is still very much a male-dominated industry, although diversity is increasing. Do you have any thoughts about the role of women in this space?

I think diversity is increasing. The culture and environment can be very welcoming. I knew that I wasn’t necessarily going to know a whole lot about machine learning or know a whole lot about technology, but I had an interest in it… and that’s all you need really. I think the days where computer science seems more like a geeky male pursuit is coming to an end, as more people interact with technology.