How do we build trust in AI?

While AI is becoming more common in business and the daily lives of Australians, trust remains a key driver in maximising the technology’s potential.

Andrew Hinchliff, Group Chief Risk Officer, CommBank, opened a panel discussion at CommBank’s Accelerate AI event by recognising that there is “extraordinary value to be unlocked” through AI, and it presents an opportunity of $45–115 billion a year by 2030.

“Importantly, we only get to recognise that value if we can scale AI and we cannot scale AI unless we build trust,” he says.

“So no trust, no scalability. And in that context, the country does run the risk of being left behind and not recognising that value.

“As we know, trust can take years to build, but can be destroyed in minutes.”

“We only get to recognise value if we can scale AI and we cannot scale AI unless we build trust.”
– Andrew Hinchliff, Group Chief Risk Officer, CommBank

What drives trust in AI?

Chair in Trust and Professor of Management at the University of Melbourne, Dr Nicole Gillespie, outlined the results of a recent survey of 48,000 people across 47 countries, which identified AI literacy as a key driver of trust.

“We know it’s a lot harder to trust AI systems when you don’t understand them,” says Gillespie. “When people have low AI literacy, it makes them really vulnerable to some of the hype, as well as some of the scaremongering around AI.

“So, what our research highlights is that AI literacy is a cross-cutting enabler.”

The research found that four-in-five Australians were “very concerned” about the societal risks of AI.

“Everyone is concerned about the risks. The difference here in Australia is that 70% of people don’t think the current regulation and governance is sufficient to make AI use safe,” says Gillespie.

“We find 80% of people globally say they would trust AI systems more when they know what these mechanisms are, and the Australian Government has gone a long way in simplifying and providing very clear guidance.”

“We find 80% of people globally say they would trust AI systems more when they know what these mechanisms are, and the Australian Government has gone a long way in simplifying and providing very clear guidance.”
– Dr Nicole Gillespie, Chair in Trust and Professor of Management, University of Melbourne

Building in safety

Stephanie Crowe is Head of the Australian Cyber Security Centre at the Australian Signals Directorate. She says to generate trust in AI, it is important to use AI products that are secure by design and created by reputable companies that are explicit on their cybersecurity policies, and how they update and configure their AI models.

“If it hasn’t got basic cyber hygiene or good security controls in place, that AI model will inherit those vulnerabilities,” says Crowe.

“If it hasn’t got basic cyber hygiene or good security controls in place, that AI model will inherit those vulnerabilities.”
– Stephanie Crowe, Head of the Australian Cyber Security Centre, Australian Signals Directorate

“If you want to use an internet-facing AI model, for example, you probably don’t want it having access across all your sensitive datasets on your network and then funneling that all back up to the AI owner.”

Instead of attempting to build complete trust in AI, however, Crowe says there is a need to find some “middle ground between full trust and scepticism”.

“I think there is a middle ground, which is confidence in AI,” she explains.

Crowe also acknowledges that AI can be a powerful cybersecurity tool. Seeing evidence of AI being used to prevent harmful activity online can demonstrate the value of AI for people who may be concerned about the technology and how it is being used by hackers.

“We need to use AI defensively, because we are probably in a period over the next 12–18 months of uncertainty and probably more vulnerabilities being discovered by AI,” she says.

A responsible mindset

Technology company Google is at the forefront of AI, and the company’s Managing Director and Vice President, Google Australia and New Zealand, Mel Silva, says “guardrails and principles” have to be at the core of AI products.

“When we’re launching products, we know they will go into the hands of billions of people around the world and you don’t take that responsibility lightly,” says Silva.

“When we’re launching products, we know they will go into the hands of billions of people around the world and you don’t take that responsibility lightly.”
– Mel Silva, Managing Director and Vice President, Google Australia and New Zealand

“So, when it comes to the combination of research, culture and governance, I agree 100% that it has to go across everything you do.”

While it is essential that staff and customers have confidence in AI, one of the biggest risks is managing the integrity of the data.

“Think about the data from a supply-chain perspective,” says Silva. “Is it coming from reputable sources? Is it coming from an untrusted website and are you thinking about reducing the likelihood of data poisoning?”

Underpinned by good governance

One Australian company that is introducing AI in the retail sector is hardware chain Bunnings.

Bunnings’ Chief Customer Officer, Rachael McVitty, says that guardrails and team education were critical as the company rolled out its customer-facing “buddy AI shopping assistant”.

“We made sure we had really strong AI guardrails and that we built all our use cases in line with these guardrails,” says McVitty. “Some of these things include being really transparent on how we’re using AI, making sure AI is really fair, inclusive and objective, and making sure it’s reliable.

“We ensure it’s using reliable data sources, we protect the privacy of users and we make sure that the AI is accountable.”

“Some of these things include being really transparent on how we’re using AI, making sure AI is really fair, inclusive and objective, and making sure it’s reliable.”
– Rachael McVitty, Chief Customer Officer, Bunnings